Women in Telecommunications (Women in Engineering and Science) 3031219740, 9783031219740

Citation preview

Women in Engineering and Science

Maria Sabrina Greco Dajana Cassioli Silvia Liberata Ullo Margaret J. Lyons   Editors

Women in Telecommunications

Women in Engineering and Science Series Editor Jill S. Tietjen, Greenwood Village, CO, USA

The Springer Women in Engineering and Science series highlights women’s accomplishments in these critical fields. The foundational volume in the series provides a broad overview of women’s multi-faceted contributions to engineering over the last century. Each subsequent volume is dedicated to illuminating women’s research and achievements in key, targeted areas of contemporary engineering and science endeavors. The goal for the series is to raise awareness of the pivotal work women are undertaking in areas of keen importance to our global community.

Maria Sabrina Greco • Dajana Cassioli • Silvia Liberata Ullo • Margaret J. Lyons Editors

Women in Telecommunications

Editors Maria Sabrina Greco University of Pisa Pisa, Pisa, Italy Silvia Liberata Ullo University of Sannio Benevento, Italy

Dajana Cassioli University of L’Aquila L’Aquila, Italy Margaret J. Lyons Freehold, NJ, USA

ISSN 2509-6427 ISSN 2509-6435 (electronic) Women in Engineering and Science ISBN 978-3-031-21974-0 ISBN 978-3-031-21975-7 (eBook) https://doi.org/10.1007/978-3-031-21975-7 © The Editor(s) (if applicable) and The Author(s), under exclusive license to Springer Nature Switzerland AG 2023 This work is subject to copyright. All rights are solely and exclusively licensed by the Publisher, whether the whole or part of the material is concerned, specifically the rights of translation, reprinting, reuse of illustrations, recitation, broadcasting, reproduction on microfilms or in any other physical way, and transmission or information storage and retrieval, electronic adaptation, computer software, or by similar or dissimilar methodology now known or hereafter developed. The use of general descriptive names, registered names, trademarks, service marks, etc. in this publication does not imply, even in the absence of a specific statement, that such names are exempt from the relevant protective laws and regulations and therefore free for general use. The publisher, the authors, and the editors are safe to assume that the advice and information in this book are believed to be true and accurate at the date of publication. Neither the publisher nor the authors or the editors give a warranty, expressed or implied, with respect to the material contained herein or for any errors or omissions that may have been made. The publisher remains neutral with regard to jurisdictional claims in published maps and institutional affiliations. This Springer imprint is published by the registered company Springer Nature Switzerland AG The registered company address is: Gewerbestrasse 11, 6330 Cham, Switzerland

Contents

1

Women Pioneers in Telecommunications . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Jill S. Tietjen

1

2

Twenty-First-Century Women in Telecommunications . . . . . . . . . . . . . . . . Margaret J. Lyons

35

3

TLC Transversal and Strategic Role . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Carolina Botti

65

4

Recent Advances in Bayesian Inference for Complex Systems . . . . . . . . Mónica F. Bugallo

85

5

Hardware-Limited Task-Based Quantization in Systems . . . . . . . . . . . . . . 105 Derya Malak, Rabia Yazicigil, Muriel Médard, Xing Zhang, and Yonina C. Eldar

6

Satellite Communications Toward a Sustainable 3D Wireless Network . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 165 Ana I. Pérez Neira

7

Integrating AI into Radar System Design: Next-Generation Cognitive Radars . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 187 Sevgi Z. Gurbuz, Kristine L. Bell, and Maria S. Greco

8

Passive Radar: A Challenge Where Resourcefulness Is the Key to Success . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 223 Francesca Filippini and Fabiola Colone

9

Remote Sensing Through Satellites and Sensor Networks . . . . . . . . . . . . . 249 Silvia Liberata Ullo and Afreen Siddiqi

10

Uphill on Two Fronts . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 277 Carole Perry

v

vi

Contents

11

Access, Inclusion, and Accommodation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 285 Katherine Grace August

12

Private Land Mobile Radio Services In-building System Design Considerations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 335 Asuncion (Beng) Connell

13

On the Entanglement Role for the Quantum Internet. . . . . . . . . . . . . . . . . . 357 Jessica Illiano and Angela Sara Cacciapuoti

14

Space Sustainability: Toward the Future of Connectivity . . . . . . . . . . . . . 375 Ernestina Cianca and Marina Ruggieri

15

The Security of Wireless Communication Protocols Used in Mobile Health Systems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 393 Eleonora Losiouk

16

Emerging Technologies in Wireless Communications . . . . . . . . . . . . . . . . . . 413 Dajana Cassioli

Index . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 439

Chapter 1

Women Pioneers in Telecommunications Jill S. Tietjen

1.1 Introduction Women have contributed to telecommunications and its underlying technology in many ways. Some worked with radio or telegraph. Others helped develop the computers and computer methodologies that make modern telecommunications possible. One started a school to teach radio skills. Another developed the engine technology that powers satellites used for communication. One invented the technology that forms the basis of Bluetooth. Women have contributed in so many ways to telecommunications. In this chapter, we discover information about a number of women telecommunications pioneers.

1.2 Ada Byron Lovelace (1815–1852) The daughter of the English poet Lord George Byron, Ada Lovelace, now has a computer language (Ada) named after her. A somewhat sickly child, Lovelace was tutored at home and was competent in mathematics, astronomy, Latin, and music by the age of 14. Totally enthralled by Charles Babbage’s Difference Engine (an early computer concept), at 17 years old, Lovelace began studying differential equations. As proposed, his second machine, the analytical engine, could add, subtract, multiply, and divide directly, and it would be programmed using punched cards, the same logical structure used by the first large-scale electronic digital computers in the twentieth century.

J. S. Tietjen () Technically Speaking, Inc., Greenwood Village, CO, USA © The Author(s), under exclusive license to Springer Nature Switzerland AG 2023 M. S. Greco et al. (eds.), Women in Telecommunications, Women in Engineering and Science, https://doi.org/10.1007/978-3-031-21975-7_1

1

2

J. S. Tietjen

Fig. 1.1 Ada Byron Lovelace – Note G. (Courtesy of Wikipedia)

In 1842, the Italian engineer, L.F. Menabrea published a theoretical and practical description of Babbage’s analytical engine. Lovelace translated this document adding “notes” in the translation. Her notes constitute about three times the length of the original document, and, as explained by Babbage, the two documents together show “That the whole of the development and operations of analysis are now capable of being executed by machinery.” These notes include a recognition that the engine could be told what analysis to perform and how to perform it – the basis of computer software. Her notes (Fig. 1.1) were published in 1843 in Taylor’s Scientific Memoirs under her initials, because although she wanted credit for her work, it was considered undignified for aristocratic women to publish under their own names. Ada Lovelace is considered to be the first person to describe computer programming (Alic 1986; Morrow and Perl 1998).

1.3 Mattie “Ma” Kiley (1880–1970) Famous “telegrapher” Mattie “Ma” Kiley was an expert at telegraph operation and Morse code for the railroads. The railroads needed instant communication to operate safely, and the telegraph was the technology available to meet that need at the time. Although most early telegraph operators were men – and it was thought that women

1 Women Pioneers in Telecommunications

3

didn’t have electronic or mechanical capabilities – Kiley proved that stereotype incorrect. In 1902, she was hired for her first railroad telegraphy job in Mexico. She gained fame as “Ma” Kiley (the last name of her second husband). She worked in both Texas and Mexico as a telegrapher, station operator, and dispatcher until her retirement in 1942. Although not the first woman telegrapher, Kiley was the best known. In 1950, Railroad Magazine ran an autobiographical story on her titled “The Bug and I” (Ma Kiley 2021; Hatch 2006).

1.4 Helen Campbell (?–?) Helen Campbell, whose pictures are archived in the Library of Congress (Figs. 1.2 and 1.3), was a Hunter College student who received a license as a wireless operator. She then became the first wireless operator of the National League for Women’s Service. This program was set up by the US government jointly with the Red Cross in anticipation of World War I to use women to fill jobs vacated by the men who were drafted into military service (The Radio Club of America 2021). Fig. 1.2 Helen Campbell, wireless operator. (Courtesy of Library of Congress)

1.5 Mary Texanna Loomis (1880–1960) Mary Texanna Loomis (Fig. 1.4) founded the Loomis Radio School in Washington, DC, in 1920 to train men to enter the radio profession and in 1927 wrote the textbook, Radio Theory and Operating for the Radio Student and Practical Operator. She explains her rationale for establishing the school, for which she served as President and Lecturer, in the Dearborn Independent in 1921:

4

J. S. Tietjen

Fig. 1.3 Helen Campbell. (Courtesy of Library of Congress)

Fig. 1.4 Mary T. Loomis. (Courtesy of Library of Congress)

There were two reasons why I launched into this fascinating work. In the early stages of the World War, I was eager to do something useful for my country and therefore mastered wireless telegraphy. The United States Department of Commerce thought sufficiently well of my ability to grant me a first grade radio license, and by the time the armistice was signed I was so fascinated with the work that I just hated to give it up and return to what seemed like ordinary everyday endeavors. Suddenly recalling the fact that a cousin of mine, Dr.

1 Women Pioneers in Telecommunications

5

Mahlon Loomis, was really responsible for giving to the world wireless telegraphy, having invented and demonstrated it some years before Mr. Marconi was born, the happy thought came to me that right now was my opportunity to do something worth while in honor of his memory. I, therefore, dug right down to the bottom of my bank account and founded a school in honor of that pioneer electrical inventor who, in 1865, sent the first aerial telegraph message between two peaks of the Blue Ridge Mountains in Virginia. My great ambition is to obtain the world-wide credit that is due his memory.

The wireless apparatus used at the school was constructed on site. She also said: No man can graduate from my school until he learns how to make any part of the apparatus. I give him a blue print of what I want him to do and tell him to go into the shop and keep hammering away until the job is completed. I want my graduates to be able to meet any emergency or mishap that may arise some day far out on the sea. (http://www.loomis. mysite.com/page2.html; https://www.facebook.com/RadioClubOfAmerica/)

1.6 Gladys Kathleen Parkin (1900–1990) As a 15-year-old, Gladys Kathleen Parkin obtained a first-class commercial radio operator’s license (Fig. 1.5). She had been interested in wireless telegraphy since she was 5 years old and operated an amateur wireless station in her home with her brother. Six years previously, she had obtained her amateur license. The Parkin’s wireless station was one of the first wireless stations in California. In the October 1916 edition of the Electrical Experimenter, Parkin said: With reference to my ideas about the wireless profession as a vocation or worthwhile hobby for women, I think wireless telegraphy is a most fascinating study, and one which could very easily be taken up by girls, as it is a great deal more interesting than the telephone and telegraph work, in which so many girls are now employed. I am only fifteen, and I learned the code several years ago, by practicing a few minutes each day on a buzzer. I studied a good deal and I found it quite easy to obtain my first grade commercial government license, last April. It seems to me that every one should at least know the code, as cases might easily arise of a ship in distress, where the operators might be incapacitated, and a knowledge of the code might be the means of saving the ship and the lives of the passengers. But the interest in wireless does not end in the knowledge of the code. You can gradually learn to make all your own instruments, as I have done with my 1/4 kilowatt set. There is always more ahead of you, as wireless telegraphy is still in its infancy

As an adult, she worked with her brothers in the family radio manufacturing business, Parkin Manufacturing Company (Parkin 2021; Oral History: Parkin, Kathleen 1977).

6

J. S. Tietjen

Fig. 1.5 Gladys Kathleen Parkin

1.7 Grace Murray Hopper (1906–1992) Admiral Grace Murray Hopper (Fig. 1.6) was famous for carrying “nanoseconds” around with her. These lengths of wire – just less than one foot – represented the distance light traveled in a nanosecond, one billionth of a second. She was renowned for trying to convey scientific and engineering terms clearly and coherently to nontechnical people. Hopper, also known as “Amazing Grace” and “The Grandmother of the Computer Age,” helped develop languages for computers and developed the first computer compiler – software that translates English (or any other language) into the zeroes and ones that computers understand (machine language). Actually, her first compiler translated English, French, and German into machine language, but the Navy told her to stick with English because computers didn’t understand French and German! Computers truly only understand numbers, but humans can translate those numbers now into English, French, German, and even Chinese and Japanese. She was also part of the group that found the first computer “bug” – a moth that had gotten trapped in a relay in the central processor. When the boss asked why they weren’t making any numbers, they responded that they were “debugging” the computer. Although Admiral Hopper loved to lay claim to the discovery of this first

1 Women Pioneers in Telecommunications

7

Fig. 1.6 Admiral Grace Murray Hopper. (Courtesy Library of Congress)

computer “bug” – and it is in the Smithsonian’s National Museum of American History – the term bug had been in use for many years by then.1 Hopper received the Society of Women Engineers’ (SWE) Achievement Award in 1964 “in recognition of her significant contributions to the burgeoning computer industry as an engineering manager and originator of automatic programming systems.” She was the first woman to attain the rank of Rear Admiral in the US Navy. The Arleigh-Burke-class-guided missile destroyer USS Hopper (DDG-70) was commissioned by the US Navy in 1997. Hopper received the National Medal of Technology from President Bush in 1991, the first individual woman to receive the medal: “For her pioneering accomplishments in the development of computer programming languages that simplified computer technology and opened the door to a significantly larger universe of users.” She was inducted into the National Women’s Hall of Fame in 1994. Hopper said she believed it was always easier to ask for forgiveness than permission. “If you ask me what accomplishment I’m most proud of, the answer would

1 Zuckerman reports that Thomas Edison referred to a “bug” in his phonograph as early as 1889. Edison is reported to have defined a bug as “an expression for solving a difficulty, and implying that some imaginary insect has secreted itself inside and is causing all the trouble.”

8

J. S. Tietjen

be all of the young people I’ve trained over the years; that’s more important than writing the first compiler.” Admiral Hopper helped make computers accessible to everyone (Hopper 1906–1992; www.swe.org/SWE/Awards.achieve3.htm; Hopper 2020; Billings and Hopper 1989; Zuckerman 2000; Stanley 1995; Kass-Simon and Farnes 1990).

1.8 Hedy Lamarr (1914–2000) Hedy Lamarr (Hedy Kiesler Markey) (Fig. 1.7) was not only an Austrian-born American sex symbol and movie star of the 1930s and 1940s, but she also coinvented a key technology used in cell phones today – known as frequency hopping spread spectrum. She and composer George Antheil developed the jam-proof radio guidance system for torpedoes. They filed for a patent, and patent number 2,292,387 (Fig. 1.8) was granted in 1942. Too complex to be used at the time with the existing technology, it was implemented by the US Navy during the 1962 Cuban Missile Crisis. Today, spread spectrum techniques are used in CDMA, Wi-Fi, and Bluetooth technologies. Lamarr and Anthell were inducted into the National Inventors Hall of Fame in 2014 (National Inventors Hall of Fame 2021). Fig. 1.7 Hedy Lamarr in Heavenly Body – MGM, 1944

1 Women Pioneers in Telecommunications

Fig. 1.8 Hedy Kiesler Markey and George Antheil Patent 2,292,387

9

10

J. S. Tietjen

1.9 Catherine Eiden (1914–1990) Catherine Eiden (Fig. 1.9) spent her career with the Illinois Bell Telephone Company. She started working in the engineering department when she was 20 years old – as an Engineering Clerk Special. Although she did not have a formal engineering education, she learned quickly, and, with her excellent math skills, she learned the telephone technology. She progressed up the ladder becoming an Engineering Assistant, then an Engineer, and then a Staff Engineer. She helped plan the telephone system. Eiden analyzed the existing systems and compared that against the forecast of customer demands in the future. Balancing the cost of equipment versus expected revenues and taking into account equipment retirements, she recommended the most economic way to expand the system to meet those needs. Eiden’s methods became a de facto standard throughout the company (Hatch 2006; Eiden 2021). Fig. 1.9 Catherine Eiden. (Courtesy, Society of Women Engineers Photograph Collection, Walter P. Reuther Library, Wayne State University)

1.10 Joan Curran (1916–1999) British physicist Joan Curran worked in the Telecommunications Research Establishment starting in 1940 assigned to the radar countermeasures group. There she worked on “Operation Window” with the express objective of developing a method to conceal aircraft from enemy radar. By 1942, she had invented chaff, aluminized paper strips thrown out of leading aircraft that concealed aircraft behind a large signal and could even lead the enemy to believe a large air force had arrived instead of one or two planes. In 1943, Curran’s chaff was used in air raids over Hamburg, Germany. It was also used for D-Day in 1944 and other bombing raids including over Duisburg, Germany in 1944 (Fig. 1.10).

1 Women Pioneers in Telecommunications

11

Fig. 1.10 Royal Air Force plane dropping chaff over Duisburg, Germany. (Courtesy of Wikipedia)

Curran completed all of her course work for an honors degree in physics at Cambridge University in 1937, but it was not awarded because Cambridge didn’t yet award women degrees; that would not occur for another 10 years. World War II diverted her from her PhD studies – instead she contributed to the war effort. She did later receive an honorary Doctor of Laws from the University of Strathclyde. Curran’s other efforts during World War II included the invention of a proximity fuse as well as work on the Manhattan Project – the effort that resulted in the development of the atomic bomb (Fischer-Hwang 2018; Curran 2021; 106: Joan Strothers (Lady Curran) 2019).

1.11 Betty Shannon (1922–2017) The wife of the man who is considered the “father of information theory” and who laid the foundation for digital theory, Betty Shannon, was a mathematician herself and collaborated on many of Claude Shannon’s inventions and critical advances. She worked at Bell Laboratories as a “computer” after graduating Phi Beta Kappa from the New Jersey College for Women (today part of Rutgers University). At Bell Labs, she worked first in the microwave research group and then moved to radar. The Shannons met in 1948 at Bell Labs. They worked together on many of his papers; she did much of the writing and added historical references. She even completed the wiring on the famous “Theseus” the mouse – a mechanical mouse invented in 1950. She made his work and ideas publishable, helping lay the foundation for today’s world of telecommunications (Betty Shannon Unsung Mathematical Genius 2021).

12

J. S. Tietjen

1.12 Yvonne Brill (1924–2013) Yvonne C. Brill expanded the frontiers of space through innovations in rocket and jet propulsion. Her accomplishments and service had major technical and programmatic impacts on a very broad range of national space programs. Her most important contributions were in advancements in rocket propulsion systems for geosynchronous communication satellites. She invented an innovative satellite propulsion system that solved complex operational problems of acquiring and maintain station (keeping the satellite in orbit and in position once it is aloft). Her patented hydrazine/hydrazine resistojet propulsion system (3,807,657 – granted April 30, 1974) provided integrated propulsion capability for geostationary satellites and became the standard in the communication satellite industry (Fig. 1.11). Two aspects of Brill’s invention are of special significance: she developed the concept for a new rocket engine, the hydrazine resistojet, and she foresaw the inherent value and simplicity of using a simple propellant. Her invention resulted in not only higher engine performance but also increased reliability of the propulsion system and, because of the reduction in propellant weight requirements, either increased payload capability or extended mission life. As a result of her innovative concepts for satellite propulsion systems and her breakthrough solutions, Brill earned an international reputation as a pioneer in space exploration and utilization. Through her personal and dedicated efforts, the resistojet system was then developed and first applied on an RCA spacecraft in 1983. Subsequently, the system concept became a satellite industry standard. It has been used by RCA, GE, and Lockheed Martin in their communication satellites. The thruster has stood the test of time; more than 200 have been flown. Satellites using her invention form the backbone of the worldwide communication network – 77 of them form the Iridium mobile telephony constellation of satellites, and 54 are geosynchronous communications satellites. The impact of global satellite communications extends to all walks of life, from national security to commercial telephone, from remote medicine and education to international trade. The invention of the hydrazine/hydrazine resistojet and its extensive use on current communications is just one of the many contributions Brill made to expanding space horizons. Her other significant technical achievements include work on propellant management feed systems, electric propulsion, and an innovative propulsion system for the Atmosphere Explorer, which, in 1973, allowed scientists to gather extensive data of the earth’s thermosphere for the first time. Brill defined, successfully advocated for, and conducted a program to evaluate capillary propellant management for three-axis stabilized spacecraft. Capillary propellant management is now routinely used on a significant fraction of US space systems. Her system has led to major improvements in the capabilities and competitiveness of very large numbers of US spacecraft.

1 Women Pioneers in Telecommunications

Fig. 1.11 Yvonne Brill Patent 3,807,657

13

14

J. S. Tietjen

Brill managed the fabrication, assembly, integration, and test of a complex Teflon solid propellant pulsed plasma propulsion system (TSPPS, also called pulsed plasma thrusters PPT). She resolved many technical and design problems in the process of bringing TSPPS from experimental to operational use in satellites, including the NOVA I spacecraft launched in May 1981, which formed part of the US Navy’s Navigational Satellite System. Her efforts both provided the solution for an unprecedented navigational capability and opened the way for the now routine use of electric propulsion on commercial Western space systems. In addition, PPTs, which are direct descendants of her design, are now being developed for propulsion functions on small/microgovernment spacecraft for many applications. Brill brought society the benefits of her bountiful knowledge and wisdom by consulting with governments and space agencies throughout the world. She was instrumental in the success of several satellite system developments for the International Maritime Satellite organization (INMARSAT) and for Telenor the Norwegian Telecommunications organization. She served as one of nine members of NASA’s Aerospace Safety Advisory Panel (ASAP) which was created in 1968 after the Apollo 204 Command Module spacecraft fire in 1967 to focus on safety issues. During her period of service, ASAP defined and recommended many technical and programmatic changes to enhance Orbiter safety that were subsequently implemented by NASA. During her long and stellar career, Brill was a pioneer in the field of space technology. Brill became a member of the National Academy of Engineering in 1987 and was a Fellow of SWE and the American Institute of Aeronautics and Astronautics. Among her many awards were the 1986 SWE Achievement Award “for important contributions in advanced auxiliary propulsion of spacecraft and devoted service to the growing professionalism of women in engineering,” the 1993 SWE Resnik Challenger Medal for expanding space horizons through innovations in rocket propulsion systems, and induction into the Women in Technology International Hall of Fame in 1999. After induction into the New Jersey Inventor’s Hall of Fame (first woman) (2009) and the National Inventors Hall of Fame (2010), in 2011, Brill received the nation’s highest honor, the National Medal of Technology and Innovation from President Obama (Fig. 1.12) “For innovation in rocket propulsion systems for geosynchronous and low earth orbit communication satellites, which greatly improved the effectiveness of space propulsion systems.” (Stanley 1995; www.swe.org/SWE/Awards.achieve3.htm; Brill 2001; President Obama Honors Nation’s Top Scientists and Innovators 2015).

1 Women Pioneers in Telecommunications

15

Fig. 1.12 Yvonne Brill receives the National Medal of Technology and Innovation. (Courtesy of the author)

1.13 Isabelle French (1924–2014) In 1944, Dr. Isabelle French (Fig. 1.13) was the first woman to graduate from TriState College’s School of Engineering with a degree in radio engineering. She worked first on the engineering and development of radar tubes at Sylvania in Massachusetts and then did similar work at Capehart-Farnsworth. In 1954 she joined Bell Telephone Laboratories in Allentown, Pennsylvania, and worked there on the technical staff until she retired. At Bell Labs, French served as a technical editor preparing Technical Information Bulletins for semiconductor devices and electron tubes. Tri-State’s Alumni Association honored her with a Distinguished Service Award in 1962. Then she received an honorary doctorate from her alma mater in 1966. French served as National President of SWE and was elected a SWE Fellow (Isabelle F. French 1924–2014; Profiles of SWE Pioneers Oral History Project 2001; Isabelle French 1980).

1.14 Anita Longley (1931–) Anita Longley (Fig. 1.14) was born in the province of Saskatchewan, Canada, and received her BA in Physics from McMaster University. She received her MS in Physiological Chemistry from the University of Minnesota. After serving as a

16

J. S. Tietjen

Fig. 1.13 Isabelle French. (Courtesy, Society of Women Engineers Photograph Collection, Walter P. Reuther Library, Wayne State University)

Fig. 1.14 Anita Longley. (Courtesy of NTIA)

research assistant and teaching at the high school and university levels, she joined the National Bureau of Standards’ (NBS) Central Radio Propagation Laboratory (CRPL) in Boulder, Colorado, in 1955.

1 Women Pioneers in Telecommunications

17

At the dawn of the computer age, Anita Longley, Phil Rice, and their colleagues at the CRPL realized that the computational capabilities of the new electronic computers could result in vast improvements of the modeling – and design – of radio systems used by the military and the public in a wide array of applications. The necessity for propagation modeling became evident during World War II, as it was vital for pilots to have reliable communications systems. In the late 1960s, she and her colleagues developed what is called the Longley-Rice model. This new model married an empirical model (based on electromagnetic theory) with measured data (terrain features and radio measurements). It provided better engineering and resulted in algorithms still in wide use today. The Longley-Rice model predicts tropospheric radio transmission loss over irregular terrain for a radio link (from transmitter to receiver) and is referred to as the Irregular Terrain Model (ITM) model. The model was designed for frequencies between 20 MHz and 20 GHz and for path lengths between 1 km and 2000 km. The allocations made by the US Federal Communications Commission for commercial and non-federal spectrum as well as the allocation of federal spectrum to the military and non-military government agencies rely on computer models that incorporate the rapidly evolving technology developments of the last 50 years. Many of the more sophisticated computer models in the twenty-first century that take advantage of satellite data and advanced programming languages for determination of radio transmission loss have the basic Longley-Rice methodology embedded in their programming. Longley wrote or collaborated on more than 20 research reports. She was awarded the US Department of Commerce Silver Medal for joint authorship of NBS Technical Note 101 on “Transmission Loss Predictions for Tropospheric Communication Circuits,” which is still widely referenced today (Tietjen Jill 2013).

1.15 Vivian Carr (1925–2018) In 1981, when Vivian Carr (Fig. 1.15) received an honorary bachelor’s of science in mechanical engineering degree from Stevens Institute of Technology, the citation read on that occasion summarized her career and contributions: Vivian Alling Carr. District Manager of the Fundamental Planning Division of the Tariffs and Costs Department of American Telephone and Telegraphy Company. Talented and versatile, Mrs. Carr has excelled in numerous positions of corporate responsibility with the Bell System, ranging from engineering to cost accounting, planning and management. Her awareness of environmental concerns and her understanding of the effects and implications of machines on people have been important elements in her successful career. Mrs. Carr’s responsibilities at AT&T have demanded diverse skills, which she applied in the Transmission, Government Communications and Switching Divisions, and the Executive, and Tariffs and Costs Departments. Among her assignments have been judging the impact of new technology on the utility’s system, business and service as well as analyzing, evaluating and administrating basic tariff policies.

18

J. S. Tietjen

Fig. 1.15 Vivian A. Carr

All these accomplishments have been achieved by Mrs. Carr without the formality of a college degree. Her preparation for corporate challenges, however, was assisted by Stevens through the Institute’s War Industries Training School, where she studied in 1943, and by Iowa State University, where she participated in more recent years in graduate courses in engineering economics. Yet, Vivian Carr remains essentially self-educated. Her election as a member and subsequently as a fellow in the Radio Club of America were firsts for a woman, as were her memberships in the Engineers’ Club of New York and in the New York Section of the Institute of Electrical and Electronics Engineers, for all three of which she has served as chairman or board member. She is an enthusiastic and active member of the Stevens National Development Council for the Technology for Tomorrow program and has participated in our Women in Engineering program as a speaker and fund raiser. By all measures, Vivian Carr is an accomplished engineer, whom we at Stevens are delighted to honor. I ask you, sir, to confer upon Vivian Alling Carr the baccalaureate degree of MECHANICAL ENGINEER, HONORIS CAUSA.

Carr had a long and storied career in the telecommunications industry at Bell Telephone Labs and its successor, American Telephone & Telegraph Co. (AT&T). She worked at Bell Labs from 1943 to 1954 and then at American Telephone & Telegraph Co. until her retirement in 1980. Her active service within Institute of Electrical and Electronics Engineers (IEEE) was recognized through a number of awards including the 1984 Centennial Medal and the 2000 Millennium Medal. Carr received the 1982 President’s Award from Radio Club of America (RCA) for “Unselfish dedication to the support of the Radio Club of America.” She was the first woman president of RCA. In 2014, RCA created the Vivian A. Carr Award and presented it as a tribute to her. It is now awarded to women who have contributed significantly to the wireless industry (Carr 2021; Carr 1925–2018).

1 Women Pioneers in Telecommunications

19

1.16 Alice Morgan Martin (?– 1998) Shortly after the Japanese attack on Pearl Harbor, Alice Morgan Martin (Fig. 1.16) was selected to become a member of the first group of women to be trained for the Drafting Department at Bendix Radio in Baltimore. After learning the fundamentals of drafting, radio, and shop work, she was assigned to the Systems Department and began evening courses in mechanical engineering. Alice recalled that this thrilled her dad who was also a mechanical engineer and who had tried in vain to interest at least one of his four sons to follow his profession. Alice became the first woman engineer in Bendix Radio and was then assigned to an all-male group of draftsmen and engineers to design radio control units for commercial and military aircraft. Within a few years, the manager of the West Coast Branch of Bendix Radio requested that she transfer to California. As Senior Field Engineer, Alice moved to the west coast to offer technical assistance to the sales engineers. Her work included contact with customers such as Douglas, Lockheed, Convair, and Boeing aircraft companies and the major foreign and domestic airlines as well as liaison between the branch office in California and the main plant on the east coast. When Bendix discontinued manufacture of radio control panels, Alice utilized her knowledge of aircraft radio and her customer contacts to start her own electronic company which specialized in the design and manufacture of cockpit control units for remotely controlling radio equipment. After 3 years of successful operation, Alice sold her organization and returned to work with Bendix, this time with the Pacific Division in California where she served as a design engineer in Sonar Engineering and Military Navigation. Subsequently, she worked as a design engineer on the Polaris missile at Lockheed, and later she worked on the Saturn Fig. 1.16 Alice Morgan Martin. (Courtesy, Society of Women Engineers Photograph Collection, Walter P. Reuther Library, Wayne State University)

20

J. S. Tietjen

project for Douglas Missiles & Space Systems. She served as National President of SWE (Martin 2021).

1.17 Erna Schneider Hoover (1926–) Considered the mother of electronic switching, Erna Schneider Hoover graduated Phi Beta Kappa from Wellesley College in 1948 with a B.A. in history and philosophy. She earned her doctorate in philosophy and the foundation of mathematics at Yale University in 1951. Hoover then taught philosophy and logic at Swarthmore College before joining Bell Labs in 1954. One of the few women at Bell Labs, Hoover was the first to be appointed as a technical supervisor. She directed programs for the radar used in the Safeguard AntiBallistic Missile Defense System used to protect the US Air Force’s intercontinental ballistic missile (ICBM) silos. In 1965, Bell Labs announced what it considered its largest project in history, the No. 1 Electronic Switching System (ESS), the first large-scale computerized switching systems in the Bell System, which would revolutionize communication by telephone. Hoover made critical contributions to the system architecture of the first electronic telephone central office. It used “stored program control” which prioritized processes and allowed better performance during peak performance periods. The patent for the system (Patent Number 3,623,007 – Fig. 1.17) belonged to Dr. Hoover, who became one of the world’s first software patent holders. She drew the first sketches for this patent in the hospital after giving birth to one of her daughters. Hoover said the following about her invention: To my mind it was kind of common sense . . . I designed the executive program for handling situations when there are too many calls, to keep it operating efficiently without hanging up on itself. Basically it was designed to keep the machine from throwing up its hands and going berserk.

In 1978, she was named head of the technical department and oversaw software applications with a particular focus on Artificial Intelligence and IMS-IBM/Unixbased systems communications for the next 10 years. She retired in 1987. Hoover received Wellesley’s 1990 Alumni Achievement Award. In 2008, she was inducted into the National Inventors Hall of Fame (Wellesley College 2021; Hoover 2021a, b).

1.18 Gladys West (1930–) Called the “Hidden Figure” behind GPS (global positioning system), Dr. Gladys West grew up with a love for mathematics that was encouraged by her teachers and which she saw as an alternative to the farm and factory work in which she saw her

1 Women Pioneers in Telecommunications

Fig. 1.17 Erna Schneider Hoover Patent 3,623,007

21

22

J. S. Tietjen

family members engaged. She attended what is today Virginia State University on full scholarship where she majored in mathematics and became a member of Alpha Kappa Alpha sorority. After receiving her BS in 1952, and applying unsuccessfully for government jobs, she taught for 2 years before completing her master’s in mathematics in 1955, again at Virginia State University. In 1956, West was hired by the US Naval Weapons Laboratory at Dahlgren, Virginia, where she stayed until her retirement in 1998. The second Black woman hired and the fourth Black employee, West was provided with computer training and worked on the Naval Ordnance Research Calculator. In 1978, she was project manager for SEASAT, the first earth-orbiting satellite that sensed ocean depths. That project led to her work with GEOSAT, an Earth observation satellite, from which she published a guide on how to use GEOSAT data to calculate geoid heights – the shape of the earth’s surface. This made GPS possible. While at Dahlgren, West earned a second master’s – in public administration from the University of Oklahoma in 1973. After her retirement, at 70 years of age, she earned her PhD from Virginia Tech in public administration. In 2018, she received a commendation from the Commonwealth of Virginia’s legislature, “Gladys West for her trailblazing career in mathematics and vital contributions to modern technology.” Also in 2018, the US Air Force inducted West into the Space and Missiles Pioneers Hall of Fame (Johnson 2021; West 2021).

1.19 Evelyn Murray (1937–) The daughter of a radio engineer, Evelyn Murray (Fig. 1.18) started her engineering career as the first female special trainee for Marconi Wireless & Telegraphy Company at Chelmsford, England,2 in 1957. After completing a two-year apprenticeship, she worked as a junior engineer in the transmitter development department. Murray then studied physics and math at the University of Southampton, England, and graduated with her BS degree in 1963. She moved to Massachusetts as a physicist before pursuing higher education at Tufts and MIT. Murray worked at Technical Operations Research, Inc.; Itek; Philco-Ford Corporation; and at MIT’s Lincoln Laboratory after her move to the USA. She served as National President of SWE and was elected a SWE Fellow (Murray 1937; Society of Women Engineers 1982; Curriculum Vitae, Evelyn Murray-Lenthall 1982).

2 Guglielmo Marconi had settled in Chelmsford (England) in the early 1900s and his first factory, Wireless Telegraph and Signal Company, was built there in 1898. Chelmsford is considered the birthplace of radio (Battle to save Marconi Factory 2021; Marconi Company 2021)

1 Women Pioneers in Telecommunications

23

Fig. 1.18 Evelyn Murray. (Courtesy Society of Women Engineers Photograph Collection, Walter P. Reuther Library, Wayne State University)

1.20 Polly Rash Hollis (1939–) Polly Rash Hollis was the first woman inducted into the Space and Satellite Hall of Fame (2005) “for her leadership in the public, educational and health applications of satellites, championing of the ACTS [Advanced Communications Technology Satellite] experimental satellite and her service to SSPI [Space and Satellite Professionals International] and in the founding of the annual Gala and Hall of Fame programs.” Over her 20 years of service in the satellite industry, Hollis played a key role in bringing satellite education to universities and public schools. She worked with Congress to obtain funding for the ACTS which helped facilitate Ka-band technology. Hollis also served as president of Space and Satellite Professionals International (Hollis 2021; Space and Satellite Hall of Fame 2021; Online Journal of Space Communication 2009).

1.21 Arlene Joy Harris Arlene Joy Harris became the first female inductee into the Wireless Hall of Fame in 2007 in recognition of her business success in the telecommunications industry

24

J. S. Tietjen

in paging, mobile telephone, and cellular. An entrepreneur, she is recognized as a pioneer in mobile and wireless and an innovator for consumer products and services. Among other things, Harris successfully developed cellular billing systems, developed and implemented the first prepaid cellular service, and created the first automated wireless management system. She also developed the SOS phone, renamed the Jitterbug, which enabled less tech-savvy customers to have a product that met their needs in the cellular space and won multiple awards. Her patent number 7,286,860 (see Fig. 1.19) is for this type of phone. Best Buy purchased the company she founded that developed Jitterbug, GreatCall, for $800 million. Harris was named an RCA Fellow in 1987 and is called the “First Lady of Wireless.” Harris grew up in the telecommunications business, starting at age 12 as a mobile telephone switchboard operator in her family’s business, Industrial Communications Systems, Inc. (Wireless History Foundation, “Arlene Harris” 2021; Harris 2021a, b)

1.22 Leah Jamieson (1949–) A member of the Purdue University engineering faculty since 1976, Leah Jamieson’s research has focused on speech analysis and recognition, the design and analysis of parallel processing algorithms, and the application of parallel processing to digital speech, image, and signal processing. Her activity in IEEE has been extensive and includes serving as President and CEO of the organization as well as President of the IEEE Foundation. A member of the National Academy of Engineering, she has advocated for women in the STEM fields for many years. Dean Emerita of the College of Engineering at Purdue, the Women in Engineering Program there was named in her honor. Jamieson has received numerous awards over her career including honorary doctorates. She completed her undergraduate mathematics degree at MIT and her master’s and PhD in Electrical Engineering and Computer Science from Princeton University (Jamieson 2021).

1.23 Anita Borg (1949–2003) Anita Borg earned her BS, MS, and PhD (1981) degrees in computer science from Courant Institute of Mathematical Sciences, New York University. Early in her career, Borg was the lead designer and co-implementer of a fault tolerant microprocessor, message-based UNIX system. The system provided users the ability to run programs that would automatically recover from hardware failures. She designed and built the first software system for generating and analyzing extremely long address traces. The knowledge gained from this effort was used in the development of Digital Equipment Corporation’s Alpha technology. Later, she designed and managed the implementation of Mecca, a web-based email system used by thousands of people.

1 Women Pioneers in Telecommunications

Fig. 1.19 Arlene Harris, Patent 7,286,860

25

26

J. S. Tietjen

Borg was known for much more in the computer industry than her significant technical accomplishments. In 1987, she founded the “Systers” email list linking technical women in computing when email was in its infancy. Borg founded the Grace Hopper Celebration of Women in Computing in 1994. After joining Xerox in 1997, she created a center to find ways to apply information technology to assure a positive future for the world’s women. The Institute for Women and Technology (renamed the Anita Borg Institute after her death) researches, develops, and deploys useful, usable technology in support of women’s communities. Borg received many honors and recognitions including induction into the Women in Technology International Hall of Fame, the Melitta Benz Women of Innovation and Invention Award, the Pioneer Award from the Electronic Frontier Foundation, the August Ada Lovelace Award from the Association of Women in Computing, and the Heinz Award for Technology, the Economy, and Environment. She was a fellow of the Association for Computing Machinery. Borg holds two patents. In 1999, she was the presidential appointee (by President Clinton) to the Commission on the Advancement of Women and Minorities in Science, Engineering, and Technology (About Anita Borg 2020; Mieszkowski 1999; President Clinton Names Anita Borg to the Commission on the Advancement of Women and Minorities in Science, Engineering, and Technology 1999; O’Brien 1999; Eng 1998; Borg 1999; Top 25 Women on the Web – Dr. Anita Borg 1999; Corcoran 1999; Method for quickly acquiring and using very long traces of mixed system and user memory references 1993; Hafner 2003).

1.24 Judith Estrin (1954–) Silicon Valley and networking pioneer Judith (Judy) Estrin was involved in the internet in the early days – working with Vincent Cerf, called one of the fathers of the internet, on Transmission Control Protocol/Internet Protocol (TCP/IP) when she was at Stanford. She has founded eight companies since 1981, served as CTO of Cisco Systems, written a book, and served on the boards of directors of companies including FedEx Corporation, Sun Microsystems, and The Walt Disney Company. Estrin earned her bachelor’s degree in mathematics and computer science from UCLA and her master’s in electrical engineering in 1977 from Stanford University. She has been named three times to Fortune magazine’s list of the 50 Most Powerful Women in American Business (Women in Technology Hall of Fame, Judy Estrin 2021; Estrin 2021a; Estrin b).

1.25 Martine Rothblatt (1954–) The founder of Sirius XM satellite radio, Martine Rothblatt was inspired to complete her college education, pursue her law degree and MBA, and establish satellite radio

1 Women Pioneers in Telecommunications

27

after seeing a US Air Force space tracking station on Morne Seychellois, on the island of Mahé in Seychelles (an African country). She says, The technological beauty of that tracking station, and the inter-planetary scope of its mission, slammed into my soul so solidly that I resolved to spread that technology everywhere. I envisioned that satellite technology could unite the world in a way that we would care enough about the earth to stop polluting it, and we would care enough about each other to stop all the wars.

She returned to the USA and completed her bachelor’s degree from UCLA in communications writing her thesis on international broadcast satellites. She then undertook a join MBA/JD degree also at UCLA, graduating with both in 1981. Her focus was telecommunications policy law. After a number of positions, she joined Geostar Corporation and began to conceptualize her vision of satellite radio. In 1990, she launched Sirius XM Holdings. Sirius Satellite Radio was launched in 2002, after she obtained Patent No. 6,105,060 (Fig. 1.20). By 2004, when Howard Stern joined the system, there were tens of millions of subscribers. In the 1990s, Rothblatt underwent sexual reassignment surgery. After her daughter Jenesis was diagnosed with pulmonary hypertension and a dire prognosis, Rothblatt switched her focus from satellite radio to pharmaceuticals and found a treatment for her daughter (Persistence is Omnipotent 2021; Rothblatt 2021).

1.26 Mary Ann Weitnauer (1962?–) Mary Ann Weitnauer became the first female faculty member of Georgia Tech’s School of Electrical and Computer Engineering upon the completion of her doctorate in 1989. Today, the senior associate chair for that school also directs the Smart Antenna Research Laboratory and her research interests include wireless communication, network time synchronization for multi-hop networks, and millimeter wave communication. Weitnauer is the author or co-author of more than 190 academic papers and holds 23 patents and invention disclosures. Dr. Weitnauer received RCA’s Vivian A. Carr Award in 2017 for her achievements in the wireless industry (Weitnauer 2021).

1.27 Andrea Goldsmith (1964?–) In 2020, when Dr. Andrea Goldsmith became the first woman to receive the Marconi Prize, Vincent Cerf said: Andrea has enabled billions of consumers around the world to enjoy fast and reliable wireless service, as well as applications such as video streaming and autonomous vehicles that require stable network performance. Andrea’s personal work and that of the many engineers whom she has mentored have had a global impact on wireless networking.

28

Fig. 1.20 Martine Rothblatt, Patent No. 6,105,060

J. S. Tietjen

1 Women Pioneers in Telecommunications

29

The Prize was awarded for her pioneering contributions to the theory and practice of adaptive wireless communications. It recognizes work that she has conducted for more than 30 years beginning with her PhD research – the theories and engineering techniques that allow wireless networks to adapt to ever-changing configurations of location, signal strength, and interference. Goldsmith earned her BS degree in electrical engineering at the University of California, Berkley in 1986. She returned to her alma mater to earn her masters and PhD degrees, also in electrical engineering. After working in private industry, she taught at the California Institute of Technology before joining the Stanford University faculty in 1999. In 2020, she became the Dean of Engineering at Princeton University. She is a member of the National Academy of Engineering and the American Academy of Arts and Sciences (Stanford News 2020; Schulz 2020; Goldsmith 2021).

1.28 Maha Achour (1964?–) The CEO and founder of Metawave, Maha Achour is now developing radar technology for autonomous vehicles. Her company is building SPEKTRA, an analog beamsteering radar system that can differentiate objects in close proximity, work efficiently in all types of weather conditions, and not lose effectiveness during all types of driving scenarios. Named a Woman of Influence in Silicon Valley in 2020, Achour was founder and CEO of Rayspan which developed and commercialized metamaterials RF systems and antennas. Achour earned her PhD in physics at MIT and earned a masters in electrical engineering, communication theory, and systems from the University of California, San Diego (A Woman of Influence Maha Achour develops radar tech for cars 2020; https://www.metawave. co/about-us; Achour 2021).

1.29 Anne Chow (1966–) Anne Chow was appointed the CEO of AT&T Business in 2019. She is the first woman to be named to this position and the first woman of color to be a CEO in the history of AT&T. Her entire career has been at AT&T, where she started as an engineer and is now the highest ranking Asian American. Her responsibilities include providing three million business customers around the world with fiber and mobile communication solutions. A first-generation American, with parents who emigrated from Taiwan, Chow was good enough at the piano to be in The Julliard School’s pre-college program at age 10. Chow earned her BS and MS degrees in electrical engineering from Cornell as well as her MBA. She sits on the board of directors of the Franklin Covey

30

J. S. Tietjen

Company as well as Girl Scouts of the USA, has written a book, and has received many honors (AT&T Business CEO to give technology lecture 2021; Murphy 2020).

1.30 Maryam Rofougaran (1967–) The co-CEO of Movandi, Maryam Rofougaran is an electrical engineer who was instrumental in starting and building the wireless business at Broadcom. Under her leadership, the business grew to more than $3 billion. Broadcom had acquired Innovent Systems, the company she had co-founded in 1998. Innovent developed low cost Wi-Fi and Bluetooth system-on-chip (SoC) solutions. Rofougaran, who has more than 250 patents, co-founded Movandi to provide solutions in the 5G space. Rofougaran, who came to the US from Iran to attend college, earned her BS and MS in electrical and electronics engineering from UCLA. She has been named one of the 50 Most Powerful Women in Technology (https://movandi.com/about/; https://www.linkedin.com/in/maryam-rofougaran-411aa84/; UCLA 2021).

References 106: Joan Strothers (Lady Curran), Magnificent Women, posted 11/11/2019. https:// www.magnificentwomen.co.uk/engineer-of-the-week/106-joan-stothers-lady-curran. Accessed 4 Apr 2021 A Woman of Influence Maha Achour develops radar tech for cars, May 22, 2020. https://www.bizjournals.com/sanjose/news/2020/05/22/women-of-influence-maha-achourmetawave.html. Accessed 1 May 2021 About Anita Borg. https://anitab.org/about-us/about-anita-borg/. Accessed 21 Apr 2020 Achour M. https://www.linkedin.com/in/mahaachour/. Accessed 1 May 2021 Alic M (1986) Hypatia’s heritage: a history of women in science from antiquity through the nineteenth century. Beacon Press, Boston AT&T Business CEO to give technology lecture March 25, Cornell Chronicle, March 11, 2021. https://news.cornell.edu/stories/2021/03/att-business-ceo-give-technology-lecturemarch-25. Accessed 29 Apr 2021 Battle to save Marconi Factory. https://www.itv.com/news/anglia/update/2015-06-25/the-historyof-marconi-in-chelmsford/. Accessed 22 Apr 2021 Betty Shannon Unsung Mathematical Genius. https://blogs.scientificamerican.com/voices/bettyshannon-unsung-mathematical-genius/. Accessed 18 Apr 2021 Billings CW, Hopper G (1989) Navy admiral and computer Pioneer. Enslow Publishers, Inc., Hillside Borg A. Women in Technology International Hall of Fame. https://www.witi.com/halloffame/ 102852/Dr.-Anita-Borg-Member-of-Research-Staff-Xerox-PARC-Founding-DirectorInstitute-for-Women-And-Technology/. Accessed 17 Aug 1999 Brill Y. www.witi.org/center/witimuseum/halloffame/1999/ybrill.shtml. Accessed 14 Feb 2001 Carr VA (1925–2018). https://www.legacy.com/obituaries/app/obituary.aspx?n=vivian-acarr&pid=188896019&fhid=27019. Accessed 18 Apr 2021 Carr VA. Engineering and technology history wiki. https://ethw.org/Vivian_A._Carr. Accessed 4 Apr 2021

1 Women Pioneers in Telecommunications

31

Corcoran CT (1999) Anita Borg wants more scientists to start listening to women. Red Herring Curran J. https://en.wikipedia.org/wiki/Joan_Curran. Accessed 4 Apr 2021 Curriculum Vitae, Evelyn Murray-Lenthall (1982, June 8) SWE Archives, Walter P. Reuther Library, Wayne State University Eiden C. Society of Women Engineers. https://swe.org/membership/swes-past-presidents/ catherine-eiden/. Accessed 4 Apr 2021 Eng S (1998) Women’s group honors pioneers in technology, 26 June. San Jose Mercury News Estrin J. http://www.jlabsllc.com/. Accessed 22 April 2021a Estrin J. https://en.wikipedia.org/wiki/Judith_Estrin. Accessed 22 Apr 2021b Fischer-Hwang I (2018) The woman whose invention helped win a war — and still baffles weathermen. Smithsonian Magazine, 28 November. https://www.smithsonianmag.com/innovation/ woman-whose-invention-helped-win-warand-still-baffles-weathermen-180970900/. Accessed 4 Apr 2021 French IF (1924–2014). http://www.sweboston.org/isabelle-french.html. Accessed 4 Apr 2021 French IF (1980) Society of Women Engineers Fellow Nomination, October 29, 1980. SWE Archives, Walter P. Reuther Library, Wayne State University Gladys Mae West JB (1930-) https://www.blackpast.org/african-american-history/people-africanamerican-history/gladys-mae-west-1930/. Accessed 1 May 2021 Goldsmith A (engineer) https://en.wikipedia.org/wiki/Andrea_Goldsmith_(engineer). Accessed 1 May 2021 Hafner K (2003, April 11) Anita Borg, 54, creator of Systers list. Rocky Mountain News Harris A (inventor). https://en.wikipedia.org/wiki/Arlene_Harris_(inventor). Accessed 29 Apr 2021a Harris AJ. IT history society. https://www.ithistory.org/honor-roll/ms-arlene-joy-harris. Accessed 29 Apr 2021b Hatch S (2006) Changing our world: true stories of women engineers. American Society of Civil Engineers, Reston Hollis PR. https://www.sspi.org/cpages/hof-hollis. Accessed 1 May 2021 Hoover ES. National Inventors Hall of Fame. https://www.invent.org/inductees/erna-schneiderhoover. Accessed 18 Apr 2021a Hoover ES. Wikipedia. https://en.wikipedia.org/wiki/Erna_Schneider_Hoover. Accessed 18 Apr 2021b Hopper G 1906–1992. https://www.womenofthehall.org/inductee/grace-hopper/. Accessed 1 Sept 1999 Hopper G. National medals of science and technology foundation. https:// www.nationalmedals.org/laureates/grace-hopper. Accessed 9 Apr 2020 http://www.loomis.mysite.com/page2.html. Accessed 22 Apr 2021 https://movandi.com/about/. Accessed 1 May 2021 https://www.facebook.com/RadioClubOfAmerica/. Accessed 22 Apr 2021 https://www.linkedin.com/in/maryam-rofougaran-411aa84/. Accessed 1 May 2021 https://www.metawave.co/about-us. Accessed 1 May 2021 Jamieson LH (2021, March) Biography. https://engineering.purdue.edu/~lhj/biosketch.pdf. Accessed 22 Apr 2021 Kass-Simon GFP (eds) (1990) Women of science: righting the record. Indian University Press, Bloomington Ma Kiley M. Engineering and technology wiki. https://ethw.org/Mattie_Ma_Kiley. Accessed 19 Mar 2021 Marconi Company. https://en.wikipedia.org/wiki/Marconi_Company. Accessed 22 Apr 2021 Martin AM. Society of women engineers. https://swe.org/membership/swes-past-presidents/alicemorgan-martin/. Accessed 4 Apr 2021 Method for quickly acquiring and using very long traces of mixed system and user memory references, Patent 5,274,811 granted 12/28/1993. Patent 4,590, “Backup fault tolerant computer system, granted May 20, 1986 Mieszkowski K (1999) Sisterhood is digital. Fast Company

32

J. S. Tietjen

Morrow C, Perl T (eds) (1998) Notable women in mathematics: a biographical dictionary. Greenwood Press, Westport, Connecticut Murphy SH (2020, April) Meet the Julliard-trained pianist who became leader of a $37 billion business group at AT&T. https://www.dmagazine.com/publications/d-ceo/2020/april/ meet-the-juilliard-pianist-who-became-leader-of-a-37-billion-business-group-at-att/. Accessed 1 May 2021 Murray EME (1937) Society of Women Engineers – Boston section. http://www.sweboston.org/ evelyn-murray.html. Accessed 4 Apr 2021 National Inventors Hall of Fame. https://www.invent.org/inductees/hedy-lamarr. Accessed 4 Apr 2021 O’Brien T (1999) Women on the verge of a high-tech breakthrough, 9 May. San Jose Mercury News Online Journal of Space Communication, Issue No. 15, Spring 2009. https://spacejournal.ohio.edu/ issue15/hollis.html. Accessed 1 May 2021 Oral History: Parkin, Kathleen, September 23, 1977 [interview], ohp_1496.pdf, Anne T. Kent California Room, Marin County Free Library. http://contentdm.marinlibrary.org/digital/collection/ ohp/id/1496. Accessed 23 Mar 2021 Parkin GK. https://en.wikipedia.org/wiki/Gladys_Kathleen_Parkin. Accessed 23 Mar 2021 Persistence is Omnipotent. https://www.uspto.gov/learning-and-resources/journeys-innovation/ field-stories/persistence-omnipotent. Accessed 1 May 2021 “President Clinton Names Anita Borg to the Commission on the Advancement of Women and Minorities in Science, Engineering, and Technology,” White House Press Release, June 29, 1999. https://www.govinfo.gov/content/pkg/WCPD-1999-07-05/pdf/WCPD-199907-05.pdf. Accessed 20 Aug 1999 President Obama Honors Nation’s Top Scientists and Innovators. https://www.whitehouse.gov/ the-press-office/2011/09/27/president-obama-honors-nation-s-top-scientists-and-innovators. Accessed 25 May 2015 Profiles of SWE Pioneers Oral History Project, Walter P. Reuther Library and Archives of Labor and Urban Affairs, Wayne State University. Dr. Isabelle French and Elaine Pitts: An interview conducted by Lauren Kata, Dianne Deturris, and Margaret Pritchard for the Society of Women Engineers, June 29, 2001. Interview LOH001952.14 at the Walter P. Reuther Library and Archives of Labor and Urban Affairs, Wayne State University. https://ethw.org/ Oral-History:Isabelle_French_and_Elaine_Pitts. Accessed 22 Apr 2021 Rothblatt M. https://en.wikipedia.org/wiki/Martine_Rothblatt. Accessed 1 May 2021 Schulz S (2020, April 15) Andrea Goldsmith, entrepreneur and leader in wireless communications, appointed Princeton University dean of engineering. https://www.princeton.edu/news/2020/04/ 15/andrea-goldsmith-entrepreneur-and-leader-wireless-communications-appointed. Accessed 1 May 2021 Society of Women Engineers, Press Release. July 1982. Murray-Lenthall Installed as SWE President. SWE Archives, Walter P. Reuther Library, Wayne State University Space and Satellite Hall of Fame. https://www.sspi.org/cpages/hall-of-fame. Accessed 1 May 2021 Stanford News (2020, April 30) Andrea Goldsmith becomes first woman to win the Marconi Prize, shattering a glass ceiling in the field of telecommunications. https://news.stanford.edu/2020/ 04/30/andrea-goldsmith-first-woman-win-marconi-prize/. Accessed 1 May 2021 Stanley A (1995) Mothers and daughters of invention: notes for a revised history of technology. Rutgers University Press, New Brunswick The Radio Club of America RCA recognizes women of achievein the wireless industry. https://www.radioclubofamerica.org/ ment content.aspx?page_id=22&club_id=500767&module_id=471781. Accessed 29 Apr 2021 Tietjen Jill S (2013) Anita Longley’s legacy: the Longley-Rice model – still going strong after almost 50 years. IEEE Antennas Propag Mag 55(3):237–240 Top 25 Women on the Web – Dr. Anita Borg. wysiwyg://5 http://www.top25.org/ab.shtml. Accessed 17 Aug 1999

1 Women Pioneers in Telecommunications

33

UCLA. Samueli Electrical and Computer Engineering, Maryam Rofougaran: Transforming Wireless Technology. https://www.ee.ucla.edu/maryam-rofougaran-transforming-wirelesstechnology/. Accessed 1 May 2021 Weitnauer MA. https://www.ece.gatech.edu/faculty-staff-directory/mary-ann-weitnauer. Accessed 22 Apr 2021 Wellesley College. Alumni achievement awards’48, Erna Schneider Hoover. https:// www.wellesley.edu/alumnae/awards/achievementawards/allrecipients/erna-schneider-hoover48. Accessed 18 Apr 2021 West G. https://en.wikipedia.org/wiki/Gladys_West. Accessed 1 May 2021 Wireless History Foundation, “Arlene Harris”. http://wirelesshistoryfoundation.org/arlene-harris/. Accessed 29 Apr 2021 Women in Technology Hall of Fame, Judy Estrin. https://www.witi.com/halloffame/240551/JudyEstrin-Chief-Executive-Officer-JLABS,-LLC/. Accessed 22 Apr 2021 www.swe.org/SWE/Awards.achieve3.htm. Accessed 1 Sept 1999 Zuckerman L (2000, April 22) Think tank: if There’s a bug in the etymology, you may never get it out. The New York Times

Jill S. Tietjen, P.E., entered the University of Virginia in the Fall of 1972 (the third year that women were admitted as undergraduates after a suit was filed in court by women seeking admission) intending to be a mathematics major. But midway through her first semester, she found engineering and made all of the arrangements necessary to transfer. In 1976, she graduated with a B.S. in Applied Mathematics (minor in Electrical Engineering) (Tau Beta Pi, Virginia Alpha) and went to work in the electric utility industry. Galvanized by the fact that no one, not even her Ph.D. engineer father, had encouraged her to pursue an engineering education and that only after her graduation did she discover that her degree was not ABET-accredited, she joined the Society of Women Engineers (SWE) and for more than 40 years has worked to encourage young women to pursue science, technology, engineering and mathematics (STEM) careers. In 1982, she became licensed as a professional engineer in Colorado. Tietjen started working jigsaw puzzles at age two and has always loved to solve problems. She derives tremendous satisfaction seeing the result of her work – the electricity product that is so reliable that most Americans just take its provision for granted. Flying at night and seeing the lights below, she knows that she had a hand in this infrastructure miracle. An expert witness, she works to plan new power plants. Her efforts to nominate women for awards began in SWE and have progressed to her acknowledgement as one of the top nominators of women in the country. Her nominees have received the National Medal of Technology and the Kate Gleason Medal; they have been inducted into the National Women’s Hall of Fame and state Halls including Colorado, Maryland and Delaware; and have received university and professional society recognition. Tietjen believes that it is imperative to nominate women for awards – for the role modeling and knowledge of women’s accomplishments that it provides for the youth of our country.

34

J. S. Tietjen Tietjen received her MBA from the University of North Carolina at Charlotte. She has been the recipient of many awards including the Distinguished Service Award from SWE (of which she has been named a Fellow and is a Society Past President), and the Distinguished Alumna Award from both the University of Virginia and the University of North Carolina at Charlotte. She has been inducted into the Colorado Women’s Hall of Fame, the Colorado Authors’ Hall of Fame, and the National Academy of Construction. Tietjen sits on the board of Georgia Transmission Corporation and spent eleven years on the board of Merrick & Company. Her publications include the bestselling and awardwinning books Her Story: A Timeline of the Women Who Changed America for which she received the Daughters of the American Revolution History Award Medal and Hollywood: Her Story, An Illustrated History of Women and the Movies which has received numerous awards.

Chapter 2

Twenty-First-Century Women in Telecommunications Margaret J. Lyons

2.1 Introduction The 25 women who have invested their time and talent to the creation of this volume have contributed to telecommunications and its underlying technology in many ways. They are a direct reflection and outgrowth of the pioneers celebrated in Chap. 1. One of the exciting aspects about studying, researching, or otherwise working in the field of Telecommunications is that it is a relatively new discipline. Some of these twenty-first century women may well be included in a future list of Telecommunications Pioneers. Some work with modern terrestrial radio. Others continue developing the signal processing and information theory methodologies that make modern telecommunications possible. One started a middle school curriculum to teach radio skills. You will note that many of the authors in this volume have studies and research or work experience in multiple countries on multiple continents. They participate and advocate for continued and deepening international sharing of the underlying concepts and technologies connecting our world. This chapter outlines the contributors’ wide-ranging career paths and illustrates how their talents and specialties within Telecommunications connect them to each other and to the pioneers of Chap. 1. In addition to technical academic pursuits and careers, Jill Tietjen and Carolina Botti work to preserve and promote people and places from history. This type of work provides us the ability make the connections and watch the threads of past provide the tapestry of today’s world.

M. J. Lyons () Freehold, NJ, USA e-mail: [email protected] © The Author(s), under exclusive license to Springer Nature Switzerland AG 2023 M. S. Greco et al. (eds.), Women in Telecommunications, Women in Engineering and Science, https://doi.org/10.1007/978-3-031-21975-7_2

35

36

M. J. Lyons

2.2 Maria Sabrina Greco, Full Professor at the DII1 , Università di Pisa, Pisa (Italy)

Maria Sabrina Greco graduated in Electronic Engineering in 1993 and received the Ph.D. degree in Telecommunication Engineering in 1998, from University of Pisa, Italy. From December 1997 to May 1998, she joined the Georgia Tech Research Institute, Atlanta, USA, as a visiting research scholar where she carried on research activity in the field of radar detection in non-Gaussian background. In 1993 she joined the Dept. of Information Engineering of the University of Pisa, where she is Full Professor since December 2016. She’s IEEE fellow since January 2011, and she was co-recipient of the 2001 and 2012 IEEE Aerospace and Electronic Systems Society’s Barry Carlton Awards for Best Paper and recipient of the 2008 Fred Nathanson Young Engineer of the Year award for contributions to signal processing, estimation, and detection theory. In May and June 2015, she visited as invited Professor the Université Paris-Sud, CentraleSupélec, Paris, France. She has been general chair, technical program chair, and organizing committee member of many international conferences over the last 10 years. She has been lead guest editor of the special issue on “Advanced Signal Processing for Radar Applications” of the IEEE Journal on Special Topics of Signal Processing, December 2015, guest co-editor of the special issue of the Journal of the IEEE Signal Processing Society on Special Topics in Signal Processing on “Adaptive Waveform Design for Agile Sensing and Communication,” published in June 2007 and lead guest editor of the special issue of International Journal of Navigation and Observation on” Modelling and Processing of Radar Signals for Earth Observation published in August 2008. She’s Associate Editor of IET Proceedings, Sonar, Radar and Navigation, Editorin-Chief of the IEEE Aerospace and Electronic Systems Magazine, member of the Editorial Board of the Springer Journal of Advances in Signal Processing (JASP), and Senior Editorial board member of IEEE Journal on Selected Topics of Signal Processing (J-STSP). She’s also member of the IEEE AES and IEEE SP Board of Governors and Past Chair of the IEEE AESS Radar Panel. She has been as well SP Distinguished Lecturer for the years 2014–2015, and now she’s AESS Distinguished Lecturer for the years 2015–2017 and member of the IEEE Fellow Committee. 1 Dipartimento

di Ingegneria dell’Informazione.

2 Twenty-First-Century Women in Telecommunications

37

She was elected President of the IEEE Aerospace Electronic Systems Society (AESS), an association of the IEEE (Institute of Electrical and Electronics Engineers) which has about 5000 members among Electronic Engineers, Telecommunications, Systems, Aerospace, and whose area of interest concerns complex space, maritime, and land-based systems, including radar, and its various applications. Professor Greco will hold the position of President-Elect for the next 2 years and will be the President of the Society for the other following 2 years. Her general interests are in the areas of statistical signal processing, estimation, and detection theory. In particular, her research interests include clutter models, coherent and incoherent detection in non-Gaussian clutter, CFAR techniques, radar waveform diversity, and bistatic/mustistatic active and passive radars and cognitive radars. She co-authored many book chapters and more than 190 journal and conference papers.

2.3 Dajana Cassioli, Associate Professor at the DISIM2 , Università degli Studi dell’Aquila, L’Aquila (Italy)

Dajana Cassioli is an Associate Professor of Telecommunications Engineering at the DISIM – University of L’Aquila, Italy. Her research interests span over wireless communications, 5G/B5G networks, and cybersecurity. She is the Coordinator of L’Aquila’s Node of the CINI National Laboratory on Cybersecurity. In 2010 and 2016, she was awarded the ERC-StG VISION and the ERC-PoCiCARE. She served as the Industry Co-Chair of PIMRC 2018, WIE Chair for RTSI 2018, RTSI 2019, RTSI 2020, MELECON 2020 and 2020 IEEE Int. Workshop for Industry 4.0 and IoT, and as TPC member of several International Conferences (ICC, PIMRC, VTC, GLOBECOM). She is an Editor of the IET Electronics Letters and IEEE Communications Letters and ITL and ETT John Wiley & Sons Ltd.

2 Dipartimento

di Ingegneria e Scienze dell’Informazione e Matematica.

38

M. J. Lyons

2.4 Silvia Liberata Ullo, Researcher at the DING3 , Università degli Studi del Sannio, Benevento (Italy)

Graduated with laude in Electronic Engineering at the Faculty of Engineering, of Federico II University, in Naples, Silivia Liberata Ullo received the M.Sc. in Management from the Massachusetts Institute of Technology (MIT) Sloan School of Boston, USA, in June 1992. Her research interests mainly deal with signal processing, remote sensing, image and satellite data analysis, machine learning applied to satellite images, quantum machine learning radar systems, sensor networks, and smart grids. She has been from 2019 to 2021 a member of the Academic Senate at University of Sannio and the National Referent for the FIDAPA BPW Italy Sciences and Technologies Task Force. Since 2004, she has been a researcher with the University of Sannio di Benevento, with interests in data analysis through satellite remote sensing (RS) for Earth observation, in communication networks with a particular reference to sensor networks and smart grids, in radar systems and radar detection in non-Gaussian environment, and in non-Gaussian models for the backscatter signal from natural surfaces. There, she teaches Signal Theory and Elaboration, and Telecommunication Networks, courses for the degree in Electronic Engineering and the Optical and radar remote sensing as Ph.D. course. In the last years, she is focusing on Machine Learning (ML) and Quantum Machine Learning (QML) applied to RS. She worked with ITALTEL S.p.A. for 8 years as Research Engineer and Production Manager and, for 4 years with the Benevento Municipality, as Officer at the Data Elaboration Center (CED). She has been a member of the Board of Directors for the Municipal Transport Society (AMTU) of Benevento for 3 years (1994–1997). She has been the promoter of the Agreement between the University of Sannio and the MAPSAT (Space agency in Benevento), the CSEO (Cyprus Space Exploration Organisation), and the Italian Atlantic Committee e ATA (Atlantic Treaty Association) and many other institutions and companies to work on research projects of joint interest that can involve Unisannio students for their bachelor

3 Dipartimento

di INGegneria (DING).

2 Twenty-First-Century Women in Telecommunications

39

and master program thesis. Moreover, she has promoted the (Memorandum Of Understanding) MOU with many Indian universities, such as the East Point College of Engineering and Technology in Bangalore, the GSSS Institute of Engineering Technology for Women in Mysuru, and many others. Silvia Ullo has been co-organizer of many national and international conferences as the MetroAeroSpace Workshop, since 2014, and the first Workshop on Networks and Cyber Security, in Benevento, 2017. She has also organized several special sessions in international conferences as, for instance, “Wireless Sensor Networks and Remote Sensing for Environmental Applications” within the first IEEE International Environmental Engineering Conference held in Milan, March 2018. She has received university and professional society recognitions. In 1990, she was awarded the “Golden Apple” from the “Marisa Bellisario” Foundation that every year recognizes this award to women who are excelling in typically male jobs. Moreover, she has devoted many years to spreading her experience and supporting young people to pursue scientific studies, above all young girls, by promoting several initiatives in which women can show themselves as role models.

2.5 Margaret J. Lyons, PE, RF/Communications Engineer, Retired

Margaret J. Lyons, PE, has more than 30 years of experience in wireless communications, including two-way radio, paging, and microwave radio systems engineering and consulting. She earned her Bachelor of Science in Computer and Electrical Engineering at Purdue University. She has been an active member of the Society of Women Engineers since 1984 including a term on the National Board of Directors. She has been a member of IEEE since 1986 and was a charter member of the NJ Coast Section Women in Engineering Affinity group (2009). At an early age, Margaret showed an affinity for puzzles and was much more interested in math and science studies. When she was a young girl at playtime with friends, her first action when playing “house” was to grab a piece of chalk and draw the floor plans. A later career in engineering was no surprise.

40

M. J. Lyons

She began her collegiate studies in 1982 at the University of Scranton, Scranton, Pennsylvania, USA, not certain if she would ultimately pursue Chemical Engineering or Electrical Engineering. Freshman Chemistry and Chemistry Lab experiences solidified for her that Electrical Engineering was the better path. A transfer to Purdue University, West Lafayette, Indiana, USA, brought with it the focus on Computer and Electrical Engineering at a time before those two disciplines were typically linked at the University level. Credit her father’s recognition that the future would include computers in everything for encouraging that dual degree pursuit. In 1986, her career started at RAM Communications Consultants, Inc., Avenel, New Jersey, USA (later RCC Consultants, Inc., of Woodbridge, New Jersey, USA), supporting the RF engineers as well as programming and IT support for the nascent computer systems in the engineering department. Over the course of 29 years at RAM/RCC, Margaret provided systems engineering design and implementation services for analog and digital paging, Itinerant Mobile Telephone Service (IMTS, the US precursor to cellular systems), cellular telephone/data systems (1G through 5G), conventional single frequency repeater systems, and complex private multichannel trunked radio systems across the continental USA and Hawaii. Her support to these industries continued through employment with V-COMM L.L.C., and Jacobs, until her retirement in 2021. Margaret became a licensed Professional Engineer in New Jersey in 1998 and subsequently registered in six additional states: Connecticut, Delaware, New York, Pennsylvania, Virginia, and Washington.

2.6 Jill S. Tietjen, P.E., CEO, Technically Speaking, Inc.

Jill S Tietjen, P.E., is an author, speaker, and electrical engineer. After more than 40 years in the electric utility industry, her focus is now on women’s advocacy worldwide. She restores the historical narrative by writing women back into history around the globe. Jill S. Tietjen, P.E., entered the University of Virginia in the fall of 1972 (the third year that women were admitted as undergraduates after a suit was filed in court by women seeking admission) intending to be a mathematics major. However, midway

2 Twenty-First-Century Women in Telecommunications

41

through her first semester, she found engineering and made all of the arrangements necessary to transfer. In 1976, she graduated with a B.S. in Applied Mathematics (minor in Electrical Engineering) (Tau Beta Pi, Virginia Alpha) and went to work in the electric utility industry. In 1982, she became licensed as a professional engineer in Colorado. Galvanized by the fact that no one, not even her Ph.D. engineer father, had encouraged her to pursue an engineering education and that only after her graduation did she discover that her degree was not ABET-accredited, she joined the Society of Women Engineers (SWE). For more than 40 years, she has worked to encourage young women to pursue science, technology, engineering, and mathematics (STEM) careers. Her efforts to nominate women for awards began in SWE and have progressed to her being known as one of the top nominators of women in the country. Her nominees have received the National Medal of Technology and the Kate Gleason Medal; they have been inducted into the National Women’s Hall of Fame and state Halls including Colorado, Maryland, and Delaware and have received university and professional society recognition. Tietjen believes that it is imperative to nominate women for awards – for the role modeling and knowledge of women’s accomplishments that it provides for the youth of our country. Her publications include the bestselling and award-winning books Her Story: A Timeline of the Women Who Changed America for which she received the Daughters of the American Revolution History Award Medal and Hollywood: Her Story, An Illustrated History of Women and the Movies, which has received numerous awards.

2.7 Carole Perry, WB2MGP, Director, Quarter Century Wireless Association; Retired, Industrial Arts and Technology Teacher

Carole Perry WB2MGP, worked as an executive secretary in an electronics manufacturing company, Rapid Circuit Inc. for 16 years. In 1980, when the company relocated she returned to Intermediate School 72 in Staten Island, NY where she worked until her retirement in 2004, teaching “Introduction to Amateur Radio” to

42

M. J. Lyons

sixth, seventh, and eighth graders for almost 30 years. Carole wrote the curriculum for “Introduction to Amateur Radio” a very successful program that had 950 students a year coming through it. In 2017, she was awarded the Brooklyn College Milton Fisher Second Harvest Award for her volunteer work with young people and technology, around the world. In February 2019, at Hamcation in Orlando Florida, USA, Carole became the first recipient of the newly created “Carole Perry Educator of the Year Award.” She also created the Radio Club of America (RCA) Young Achiever’s Award, given to students in grade 12 and below who have demonstrated excellence and creativity in wireless communications. So far, 141 youngsters have received this award. Under Carole’s leadership, the RCA Youth Activities Committee goes into schools across the country to set up radio/technology programs. Equipment, cash grants, books, and supplies are donated to the chosen schools or youth groups. Carole has moderated the Dayton Hamvention Youth Forum and Instructors’ Forum for 32 years. Carole Perry is the recipient of the prestigious 1987 Dayton Ham of The Year Award, the 1987 ARRL Instructor of The Year Award, the 1991 Marconi Wireless Memorial Award, the 1993 QCWA President’s Award, the 1996 Radio Club of America (RCA) Barry Goldwater Amateur Radio Award, the 2009 RCA President’s Award, the 2012 RCA President’s Award, and the 2015 Vivian Carr Award for Women in Radio. She is the winner of the 2016 SOAR (Sisterhood of Amateur Radio) Legacy award for Pioneering Women in Amateur Radio and the 2016 recipient of the YASME Foundation Award for Excellence. In 2017, she was the winner of the Brooklyn College Milton Fisher Second Harvest Award for her volunteer work with young people and technology, around the world. In February 2019, at Hamcation, Carole became the first recipient of the newly created “Carole Perry Educator of the Year Award.”

2.8 Katherine Grace August, PhD, Research Guest/Adjunct Professor, Stevens Institute of Technology, ECE Intelligent Networks

2 Twenty-First-Century Women in Telecommunications

43

Katherine Grace August, PhD (Kit), is a Research Guest/Adjunct Professor at Stevens Institute of Technology – ECE Intelligent Networks. Current research projects involve humanitarian activities following the UN Sustainable Development Goals and focus on employing low-cost mainstream technology to reduce inequity for those with differing abilities such as hearing loss and promoting improved opportunity for underrepresented, minorities, women, and girls through inventing. Research experience in neurorehabilitation with robots, haptics, augmented and virtual reality, functional brain imaging, signal processing, wireless, systems engineering, and the like. Kit received the PhD in Biomedical Engineering, NJIT; the MSCS-MIS, Marist College; and BFA Communications Design, Parsons The New School for Design. She worked at Bell Labs MTS New Service Concepts Systems Engineering 1991–2002. She is currently working with IEEE Standards Group P2933 TIPPSS for Connected Healthcare (Trust Identity Privacy Protection, Safety, Security). She leads IEEE SIGHT Project: “Do Good Things, Justice for All,” an experiential learning system to understand hearing loss and provide augmentative communications to reduce disparity. Kit has 18 US patents; 50 international patents; and citations 3091, h index 21, i10 index 24 and has won IEEE NJ Coast Section Volunteer and Region 1 Award 2020; “Hear, here!” Do Good Robotics Startup Competition Finalist University of Maryland, 2019; IEEE HAC “Justice for All” Event 2019. She is Chair of PACE SIGHT Group, IEEE NJ Coast Section History Chair, AP-VT-EMC Vice Chair; Whitaker Scholar 2009–2012 at ETH Zurich.

2.9 Afreen Siddiqi, a Research Scientist in the Department of Aeronautics and Astronautics at the Massachusetts Institute of Technology and an Adjunct Lecturer of Public Policy at Harvard Kennedy School

44

M. J. Lyons

Dr. Afreen Siddiqi is motivated about addressing urgent problems of sustainable resource use, development of sufficient and smart infrastructure, and equitable use of technology for improving human well-being and the natural environment. Dr. Siddiqi’s research develops systems-theoretic methods, with simulations, optimization, statistics, and decision analysis. Some of her recent work has been on earth observation systems, autonomous vehicles, energy, water, and agriculture systems. She has co-authored a book and over 100 publications in leading scientific and technical journals. She also engages internationally with policymakers on issues of development, planning, and technology. Dr. Siddiqi has an S.B. in Mechanical Engineering, an S.M. in Aeronautics and Astronautics, and a Ph.D. in Aerospace Systems, all from MIT. She has been a recipient of the Amelia Earhart Fellowship, Richard D. DuPont Fellowship, and the Rene H. Miller Prize in Systems Engineering.

2.10 Marina Ruggieri, Full Professor of Telecommunications Engineering at the University of Roma “Tor Vergata”

Marina Ruggieri is Full Professor of Telecommunications Engineering at the University of Roma “Tor Vergata” and therein co-founder and Chief Strategic Innovation Officer (CSIO) of CTIF – an Interdisciplinary Research Center on Information and Communications Technology and its verticals. Says Ruggieri: The pandemic has shown how connectivity can become ally of humanity in difficult times. In fact, as any technology, connectivity is neutral and we human beings have the capability of shaping it for either our benefit or our damage. Doing teaching and research in the connectivity field makes me feel the responsibility and the challenge of contributing to the future of humanity through both the infrastructure and the vertical application domains, also meeting the social goal of “connecting the unconnected”. In my research activities, cooperation and diversity are key-resources to aim at results being both valuable and lasting and human-centric. A major area of my research is, in fact, space connectivity systems and technologies that are intrinsically aligned with the need of broad teams.

2 Twenty-First-Century Women in Telecommunications

45

In light of the above, she is the Principal Investigator of the 40/50 GHz Communications Experiment of the TPD#5 payload on board the Alphasat satellite, launched on July 2013. Her research team believes that a deep understanding of the Q/V-band and beyond communications can contribute to the future deployment of connectivity infrastructures where the effective integration between terrestrial and space components and the sustainability of both Earth and space can be guaranteed. With the aim of contributing to the goal of “technology for the benefit of humanity,” she is also participating with different roles to IEEE activities. I am currently IEEE Aerospace and Electronic Systems Society (AESS) Vice President of Technical Operations and member of the Board of Governors (2019–2021). I am also an IEEE Technical Expert recognized in 2019 as “Impact Creator” in the area of Information and Communications Technology. I have been IEEE 2017 Vice President, Technical Activities; 2014–2015 Director of IEEE Division IX; and 2010–2011 AESS President. I have received some recognitions for my activity, like the 1990 Piero Fanti International Prize, the 2009 Pisa Donna Award for women in engineering, the 2013 Excellent Women in Roma Award, and the 2011 AESS Service Award. In 2015, she was inducted as a Professional into the IEEE Honor Society Eta Kappa Nu. She is also an IEEE Fellow “for contributions to millimeter-wave satellite communications.” She is author/co-author of 350 papers, 1 patent, and 12 books. Her writing is not only in the technical and scientific domain. In particular, I have written a novel (La finestra gialla) that received an award.

2.11 Asuncion (Beng) Connell, E.C.E., P.M.P., Wireless Discipline Specialist, Jacobs

Growing up in the Philippines, Beng started college studies in 1980 with the intention of becoming a Chemical Engineer. However, after 2 years of navigating through Math, Physics, and Chemistry, she realized that Chemistry was not her strongest point and opted to switch her major to Electronics Engineering instead.

46

M. J. Lyons

She subsequently graduated with a Bachelor of Science degree in Electronics and Communications Engineering (BS ECE) from the University of Santo Tomas, Manila, Philippines, in 1985 and became a registered Professional ECE the same year. Upon the recommendation by one of her college professors after graduating, she began her career with Miltech Industries, Inc., Philippines, in land mobile radio communications as Radio Frequency (RF) Engineering assistant. While under their employment, she enjoyed designing two-way radio systems; doing site surveys; site testing; demonstrating equipment capability to customers; and attending public conferences and bidding. In 1990 after 4 years, she shifted from the private land mobile radio industry to join Express Telecommunications, Inc., a pioneer cellular carrier in the Philippines. She remained with them for a short period of 4 months, at which time she returned to Miltech Industries. Working on two-way radio systems again and furthering her education toward a Master’s Degree in Public Management, she attained that goal in 1996 from the University of the Philippines just prior to immigrating to the United States. In 1997, she joined RCC Consultants, Inc., headquartered in Woodbridge, New Jersey, as RF Engineer. This position encompassed several northeast markets in their nationwide 900 MHz radio site acquisition and build-out project for BellSouth. She was with the Commercial Market Group until the 900 MHz project was completed and then joined the Public Safety Group of RCC, where her main assignments were various Port Authority of New York and New Jersey projects. She also assisted in other public safety projects for the City of Philadelphia, Pennsylvania, and the City of Norwalk, Connecticut, and in 2009 became a Project Management Professional. In 2015, RCC Consultants became a part of Black & Veatch. She remained with them in the same capacity until her most recent employment. In 2019, Beng accepted a position with Jacobs Engineering Group and is now involved in several Private Land Mobile Radio Services (PLMRS) in-building design projects for Grand Coulee Dam, Columbia Boulevard Wastewater Treatment Plant, Delaware River Port Authority, Metropolitan Transit Authority, and East and North River Tunnel Restoration for AMTRAK. She is currently on an on-call consultant assignment for the Port Authority of New York and New Jersey.

2 Twenty-First-Century Women in Telecommunications

47

2.12 Angela Sara Cacciapuoti, Associate Professor at the DIETI4 , Federico II University, Naples (Italy)

Angela Sara Cacciapuoti (M’10–SM’16) is a Professor at the University of Naples Federico II, Italy. In 2009, she received the Ph.D. degree in Electronic and Telecommunications Engineering and in 2005 a “Laurea” (integrated BS/MS) summa cum laude in Telecommunications Engineering, both from the University of Naples Federico II. She was a visiting researcher at Georgia Institute of Technology (USA) and at the NaNoNetworking Center in Catalunya (N3Cat), Universitat Politecnica de Catalunya (Spain). Since July 2018, she held the national habilitation as “Full Professor” in Telecommunications Engineering. Her current research interests are mainly in Quantum Communications, Quantum Networks and Quantum Information Processing. In December 2021, she was selected to be included in the annual list “N2Women: Stars in Networking and Communications” for her contribution in the quantum communication field.

2.13 Eleonora Losiouk, Postdoc Research Fellow, Dipartimento di Matematica, University of Padova

4 Dipartimento

di Ingegneria Elettrica e Tecnologie dell’Informazione.

48

M. J. Lyons

Eleonora Losiouk is a Postdoc Fellow working in the SPRITZ Group of the University of Padova, Italy. In 2018, she obtained her Ph.D. in Bioengineering and Bioinformatics from the University of Pavia, Italy. She has been a Visiting Fellow at the Ecole Polytechnique Federale de Lausanne in 2017. Since the Ph.D., her main research has targeted mobile OS, in particular the Android OS. She began by investigating telemedicine systems, with a particular focus on data synchronization between mobile devices (i.e., the Android ones) and wearable devices. During the last year of her Ph.D. studies, she visited EPFL to work on a project concerning mobile security. Afterwards, she relocated to Padua to join the Spritz Group as a Postdoc and work on cybersecurity topics, especially on Android security and future Internet security. During her first year of Postdoc, she was a researcher in the CISCO project on “Secure Mobility Management in ICN.” She had the chance to work on ICN (Information-Centric Networking), the paradigm of the future Internet architecture that should replace the current TPC/IP one. In 2019–2020 she has been a researcher in the EU funded project LOCARD – Lawful evidence cOllecting Continuity plAtfoRm Development, (SU-FCT02-20182019-2020), where she was involved in the design and development of three mobile forensics tools. Besides research activities, she is particularly proud of being the Instructor of the first course on “Mobile and IoT Security” for Master students at the University of Padua, which she set up from scratch. Moreover, she is Senior Instructor of the University of Padua team that takes part into the Italian CyberChallenge project (the UniPD team won the National Competition in the first edition). She also organized the “Wonder Women in CyberSecurity + Wonder Women in Computer Science” event and, in 2019, proposed a patent about a protection mechanism of the Bluetooth API implementation in the Android OS. Key achievements are represented by the following five publications. 1. The work in (Pham et al. 2019) focuses on a privacy issue of the Android OS, which refers to the capability of Android apps to get the list of other installed apps on the same mobile device either by querying specific APIs of the Android framework or by inspecting some “traces” left by other apps. The paper proposes a privacy-preserving solution based on the Android virtualization technique, which comprehensively hides the presence of an app. Because of this work, Google recently introduced a new Android permission, called “QUERY_ALL_PACKAGES,” which an app has to require to be granted to access the list of installed apps. 2. The paper in (Ruggia et al. 2021) is significant because of its contribution in the Android anti-repackaging research area. In fact, the solution presented is the first one aimed to prevent both traditional repackaging attacks and those based on the Android virtualization technique.

2 Twenty-First-Century Women in Telecommunications

49

3. The work in (Casagrande et al. 2021) is impactful because it is a contribution to the fight against the current SARS-CoV-2 pandemic. In particular, her team studied the design of contact-tracing apps, used to reduce the spread of the virus, and identified in many of them a vulnerability, which opens to relay attacks. They consequently designed a solution to detect a relay attack, thus preventing a healthcare systems breakdown. 4. The works in (El-Zawawy et al. 2021a, b) underline her contribution to the Android security research area. In particular, paper (El-Zawawy et al. 2021a) addresses vulnerabilities that affect the containers rendering web code in Android apps, while paper (El-Zawawy et al. 2021b) proposes a novel detection mechanism against possible uncontrolled data flow in Android apps.

2.14 Carolina Botti, P.E., Director of ALES the in House Company of the Italian Ministry of Culture

Carolina Botti, P.E., with a master degree in Electronic Engineering “cum laude” and a MBA from MIT, Sloan School of Management, is a nationally recognized professional engineer with 30 years, half in the private sector and half in the public sector. Since 2005, she is the Director of ALES the house company of the Italian Ministry for cultural heritage and tourism. She is responsible to lead the business unit aimed at designing and implementing strategic projects to enhance cultural heritage through the use of ICT. In her previous job experience, she was at the top of main consulting companies (Mac Group, Gemini Consulting, Booz Allen & Hamilton) managing many complex projects in the field of Telecommunications. She has been CEO of a Startup responsible for setting up the company and launching ITC services to manage and optimize the integrated supply chain in the healthcare industry. She received prestigious awards as Excellence award 2013 Manageritalia and the honor of “Cavaliere Ordine al Merito della Repubblica Italiana.”

50

M. J. Lyons

Carolina started out her career across different industries both in the public and in the private sector, after earning an Italian Master Degree with honors in Electronic Engineering, followed by an MSc in Management by the Sloan School of Management at MIT (Massachusetts Institute of Technology). She believes that an engineering background coupled with a proven strategic advisory experience provides a strong skill set to face all technical aspects and, above all, the challenges of change. Carolina spent over than 10 years in consulting, first with Gemini Consulting and later joining Booz & Company as Principal. The consulting experience proved to be very intense and professionally and personally successful. Carolina was responsible for account management and account development in Telecommunication, and, in particular, she led projects focused on strategy and marketing for major operators in fixed and mobile telephony. In the course of her assignments, Carolina had the opportunity of leading multicultural teams of brilliants consultants, while managing top manager clients. As a further managerial step forward in her career, Carolina later joined the Italian leading bank Monte dei Paschi di Siena, to become Managing Director of a controlled company in IT services. That professional experience proved to be most interesting and successful, and it reinforced Carolina’s determination to leverage and strengthen her professional background through managerial assignments. Carolina took up a new challenge when she became COO of the in-house company of the Italian Ministry of Culture (ARCUS, now incorporated in ALES as per the 2016 Italian budget law). ALES mission is to support the Ministry of Culture in managing and enhancing the Italian cultural heritage, as well as providing value-added services and innovative projects. Carolina leads the “Public-Private and Projects Funding Department,” focused on financing and monitoring strategic projects that use ITC to enhance the user experience of cultural heritage. This is a fascinating and challenging mission, carried out in a country (Italy) which has the majority of the world’s UNESCO cultural sites. As new investment strategies are being drawn and implemented under the post-Covid stimulus package, Carolina is moreover contributing to address the challenges ahead faced by the cultural sector. Carolina has been invested by a Knighthood for the Order of Merit of the Italian Republic. Moreover, she has received several national awards, including the “Excellence Award of Manageritalia.” She is an active member of no-profit organizations such as the “Marisa Bellisario Foundation” and the “Italian Parks and Gardens Association.”

2 Twenty-First-Century Women in Telecommunications

51

2.15 Monica Bugallo, Professor, Department of Electrical and Computer Engineering

Mónica F. Bugallo received her Ph.D in computer science and engineering from University of A Coruña, Spain. She is a Professor of Electrical and Computer Engineering and currently Vice Provost for Faculty Affairs and Diversity, Equity and Inclusion at Stony Brook University, NY, USA. Her research is focused on statistical signal processing, with emphasis on the theory of Monte Carlo methods and its application to different disciplines including biomedicine, ecology, sensor networks, and finance. She has also focused on STEM education and has initiated successful programs to engage students at all academic stages in the excitement of engineering and research, with focus on underrepresented groups. Bugallo has authored and co-authored two book chapters and more than 200 journal papers and refereed conference articles. She is a senior member of the IEEE, served as past chair of the IEEE Signal Processing Society Signal Processing Theory and Methods Technical Committee, and as past chair of the EURASIP Special Area Team on Theoretical and Methodological Trends in Signal Processing. She is also serving as Senior Associate Editor of the IEEE Signal Processing Letters and Associate Editor of the IEEE Transactions on Signal Processing and has been part of the technical committee and has organized various professional conferences and workshops. She has received several prestigious research and education awards including the State University of New York (SUNY) Chancellor’s Award for Excellence in Teaching (2017); the 2019 Ada Byron Award of the Galician Society of Computer Engineers (Spain) for a successful professional career path that inspires women to engineering study and careers; the Best Paper Award in the IEEE Signal Processing Magazine 2007 as co-author of a paper entitled Particle Filtering; the IEEE Outstanding Young Engineer Award (2009), for development and application of computational methods for sequential signal processing; the IEEE Athanasios Papoulis Award (2011), for innovative educational outreach that has inspired high school students and college level women to study engineering; the Higher Education Resource Services (HERS) Clare Boothe Luce (CBL) Scholarship Award (2017); and the Chair of Excellence by the Universidad Carlos III de Madrid-Banco de Santander (Spain) (2012).

52

M. J. Lyons

2.16 Jessica Illiano, Ph.D. Student in Information Technologies and Electrical Engineering at University of Naples Federico II

Jessica Illiano earned a Bachelor’s degree in 2018 and then in 2020 a Master’s degree in Telecommunications Engineering, both summa cum laude from University of Naples Federico II. Since 2018, she is a member of the www.QuantumInternet.it Research Group, FLY: Future Communications Laboratory. Currently, she is a Ph.D. student in Information Technologies and Electrical Engineering at University of Naples Federico II. She has always been curious about understanding the true functioning of the complex world of scientific phenomena and the new technologies that surrounded her. Her first approach with quantum technologies occurred in 2018 by attending a seminar organized by Prof. Angela Sara Cacciapuoti and Prof. Marcello Caleffi. Starting with that seminar, she understood that when it comes to quantum communications, a strong effort is needed to design a network that accounts for different underlying physical mechanisms. Jessica was deeply involved with the innovative perspective and the interdisciplinary open problems arising from this subject. Hence, she decided to continue her studies in that direction through her thesis at first and then through her PhD program by joining the quantum internet research group.

2 Twenty-First-Century Women in Telecommunications

53

2.17 Derya Malak, Assistant Professor, Communication Systems Department, EURECOM, Campus SophiaTech, Biot, France

Derya Malak is an Assistant Professor (Maître de Conférence) in Communication Systems at Eurecom, France. Previously, she was a tenure track Assistant Professor in the Department of ECSE at RPI between 2019 and 2021, and a Postdoctoral Associate at MIT between 2017 and 2019. She received her Ph.D. in ECE from the University of Texas at Austin in 2017, B.S. in Electrical and Electronics Engineering (EEE) with a minor in Physics from Middle East Technical University, in 2010, and M.S. in EEE from Koc University, in 2013. Dr. Malak has held visiting positions in INRIA and LINCS, Paris, and at Northeastern University, and summer internships at Huawei, Plano, TX, and Bell Labs, Murray Hill, NJ. Her expertise is in information theory, communication theory, and networking areas. She has developed novel distributed computation solutions, and wireless caching algorithms by capturing the confluence of storage, communication, and computation aspects. Dr. Malak was awarded the Graduate School fellowship by UT Austin between 2013 and 2017. She was selected to participate in the Rising Stars Workshop for women in EECS, MIT, in 2018. She received the best paper award in WiOpt 2022. Her research has been funded by the Huawei Chair program on Advanced Wireless Systems (lead 2022–), NSF, the Rensselaer-IBM AI Research Collaboration, and the DARPA Dispersive Computing Programs. She is the recipient of the ERC Starting Grant 2023–2028 on computing nonlinear functions over communication networks (SENSIBILITÉ).

54

M. J. Lyons

2.18 Rabia Yazicigil, ECE Department, Boston University, Boston, MA, USA

Rabia Tugce Yazicigil is an Assistant Professor of Electrical and Computer Engineering at Boston University and a Visiting Scholar at MIT. She received her Ph.D. degree in Electrical Engineering from Columbia University in 2016. She received the B.S. degree in Electronics Engineering from Sabanci University, Istanbul, Turkey in 2009, and the M.S. degree in Electrical and Electronics Engineering from École Polytechnique Fédérale de Lausanne (EPFL), Lausanne, Switzerland in 2011. Her research interests lie at the interface of integrated circuits, signal processing, security, bio-sensing, and wireless communications to innovate system-level solutions for future energy-constrained applications. She has been a recipient of a number of awards, including the “Electrical Engineering Collaborative Research Award” for her PhD research on Compressive Sampling Applications in Rapid RF Spectrum Sensing (2016), the second place at the Bell Labs Future X Days Student Research Competition (2015), Analog Devices Inc. outstanding student designer award (2015), and 2014 Millman Teaching Assistant Award of Columbia University. She was selected among the top 61 female graduate students and postdoctoral scholars invited to participate and present her research work in the 2015 MIT Rising Stars in Electrical Engineering Computer Science. She was selected as a semi-finalist for 2018, 35 Innovators Under 35 list sponsored by MIT Technology Review.

2 Twenty-First-Century Women in Telecommunications

55

2.19 Xing Zhang, a Postdoctoral Researcher with the Signal Acquisition Modeling and Processing Laboratory, Weizmann Institute of Science, Rehovot, Israel

Xing Zhang received the B.Eng. degree in information engineering in 2015 and the Ph.D. degree in information and communication engineering in 2021, both from the School of Information Science and Engineering, Southeast University, China. She is currently a Postdoctoral Researcher with the Signal Acquisition Modeling and Processing Laboratory, Weizmann Institute of Science, Rehovot, Israel. Her research interests include signal processing and its applications on communications.

2.20 Yonina Eldar, Weizmann University, Israel

Yonina C. Eldar received the B.Sc. degree in Physics in 1995 and the B.Sc. degree in Electrical Engineering in 1996 both from Tel-Aviv University (TAU), Tel-Aviv, Israel, and the Ph.D. degree in Electrical Engineering and Computer Science in 2002 from the Massachusetts Institute of Technology (MIT), Cambridge. From January 2002 to July 2002, she was a Postdoctoral Fellow at the Digital Signal Processing Group at MIT.

56

M. J. Lyons

She is currently a Professor in the Department of Mathematics and Computer Science, Weizmann Institute of Science, Rehovot, Israel. She was previously a Professor in the Department of Electrical Engineering at the Technion, where she held the Edwards Chair in Engineering. She is also a Visiting Professor at MIT, a Visiting Scientist at the Broad Institute, and an Adjunct Professor at Duke University and was a Visiting Professor at Stanford. She is a member of the Israel Academy of Sciences and Humanities (elected 2017), an IEEE Fellow, and a EURASIP Fellow. Dr. Eldar has received numerous awards for excellence in research and teaching, including the IEEE Signal Processing Society Technical Achievement Award (2013), the IEEE/AESS Fred Nathanson Memorial Radar Award (2014), and the IEEE Kiyo Tomiyasu Award (2016). She was a Horev Fellow of the Leaders in Science and Technology program at the Technion and an Alon Fellow. She received several best paper awards and best demo awards together with her research students and colleagues including the SIAM outstanding Paper Prize and the IET Circuits, Devices, and Systems Premium Award and was selected as one of the 50 most influential women in Israel. She was a member of the Young Israel Academy of Science and Humanities and the Israel Committee for Higher Education. She is the Editor in Chief of Foundations and Trends in Signal Processing, a member of the IEEE Sensor Array and Multichannel Technical Committee and serves on several other IEEE committees. She is author of the book “Sampling Theory: Beyond Bandlimited Systems” and co-author of the books “Information-Theoretic Methods in Data Science,” “Compressed Sensing in Radar Signal Processing,” “Compressed Sensing,” and “Convex Optimization Methods in Signal Processing and Communications,” all published by Cambridge University Press.

2.21 Ana I. Pérez-Neira, CTTC, Spain

Ana Pérez-Neira is full professor at Universitat Politècnica de Catalunya in the Signal Theory and Communication department since 2006 and was Vice rector

2 Twenty-First-Century Women in Telecommunications

57

for Research (2010–2014). Currently, she is the Director of Centre Tecnològic de Telecomunicacions de Catalunya, Spain. Her research is in signal processing for communications, focuses on satellite communications. She holds eight patents. She is the coordinator of the Networks of Excellence on satellite communications, financed by the European Space Agency: SatnexIV-V. She is IEEE Fellow and member of the Real Academy of Science and Arts of Barcelona (RACAB). She is recipient for the 2018 EURASIP Society Award and she has been the general chair of IEEE ICASSP’20 (the first big IEEE virtual conference held by IEEE with more than 15,000 attendees). In 2020, she was awarded the ICREA Academia distinction by the Catalan government.

2.22 Fabiola Colone, Associate Professor at the Faculty of Information Engineering, Informatics, and Statistics of Sapienza University of Rome

Fabiola Colone received the laurea degree (B.S. + M.S.) in Telecommunications Engineering and the Ph.D. degree in Remote Sensing from Sapienza University of Rome, Italy, in 2002 and 2006, respectively. She joined the DIET Dept. (formerly INFOCOM) of Sapienza University of Rome as a Research Associate in January 2006. From December 2006 to June 2007, she was a Visiting Scientist at the Electronic and Electrical Engineering Dept. of the University College London, London, UK. She is currently an Associate Professor at the Faculty of Information Engineering, Informatics, and Statistics of Sapienza University of Rome. The majority of Dr. Colone’s research activity is devoted to radar systems and signal processing. She has been involved, with scientific responsibility roles, in research projects funded by the European Commission, the European Defence Agency, the Italian Space Agency, the Italian Ministry of Research, and the radar industry. Her research has been reported in over 130 publications in international technical journals, book chapters, and conference proceedings. Since 2017, she is member of the Board of Governors of the IEEE Aerospace and Electronic System Society (AESS) in which she is currently serving as Vice-President for Member

58

M. J. Lyons

Services and Editor in Chief for the IEEE AESS QEB Newsletters. Dr. Colone has been co-recipient of the 2018 Premium Award for Best Paper in IET Radar, Sonar, & Navigation. She is Associate Editor for the IEEE Transactions on Signal Processing and member of the Editorial Board of the Int. Journal of Electronics and Communications (Elsevier).

2.23 Francesca Filippini, Postdoctoral Researcher with the Department of Information Engineering, Electronics and Telecommunications, Sapienza University of Rome

Francesca Filippini received the M.Sc. degree (cum Laude) in communication engineering and the Ph.D. degree (cum Laude) in radar and remote sensing from the Sapienza University of Rome, Rome, Italy, in 2016 and 2020, respectively. From January to May 2016, she has been working her Master Thesis project with the Passive Radar and Antijamming Techniques Department at Fraunhofer FHR, Wachtberg, Germany. She is currently a Postdoctoral Researcher with the Department of Information Engineering, Electronics and Telecommunications, Sapienza University of Rome. Dr. Filippini received the 2020 IEEE AESS Robert T. Hill Best Dissertation Award for her PhD thesis and the 2020 GTTI Best PhD Thesis Award defended at an Italian University in the areas of communications technology. She also received the 2018 Premium Award for the Best Paper in IET Radar, Sonar, & Navigation, the Best Paper Award at the 2019 Int. Radar Conference, the second Best Student Paper Award at the 2018 IEEE Radar Conference, and the Best Paper Award at the 2017 GTTI Workshop on Radar and Remote Sensing. She is a Member of the IEEE Aerospace and Electronic System Society (AESS) Board of Governors, where she is currently serving as Co- Editor in Chief for the IEEE AESS QEB Newsletters and Secretary.

2 Twenty-First-Century Women in Telecommunications

59

2.24 Sevgi Z. Gurbuz, Assistant Professor, The University of Alabama, Department of Electrical and Computer Engineering, Tuscaloosa, AL, USA

Dr. Sevgi Z. Gurbuz (nee Ertan) received the B.S. degree in Electrical Engineering with a minor in Mechanical Engineering as well as the M.E., degree in Electrical Engineering from the Massachusetts Institute of Technology, Cambridge, MA, in 1998 and 2000, respectively. As an undergraduate, she did research on the design of haptic interfaces for blind navigation under the supervision of Drs. Hong Tan and Prof. Alex Pentland in the Human Dynamics Laboratory of the MIT Media Laboratory. Upon graduation, she was also commissioned as a 2nd Lieutenant in the U.S. Air Force and was selected as of one of 13 Air Force Reserve Officer Training Corps (AFROTC) cadets to pursue graduate studies while on active duty. As a Charles Stark Draper Laboratory Graduate Fellow, she pursued research on the design of delta-sigma analog-to-digital converters. Upon completion of her graduate degree, she was assigned to the U.S. Air Force Research Laboratory (AFRL), Sensors Directorate, Radar Signal Processing Technology Branch, located in Rome, NY. Thus so, by order of the USAF, was she was converted from circuit designer to radar engineer, working in a small town in upstate New York. It turned into a life-changing assignment, where she was able to work side-by-side with brilliant experts in radar, such as Dr. Michael Wicks, Prof. Donald D. Weiner, Syracuse University, and Prof. Mehrdad Soumekh, SUNYBuffalo. In particular, Dr. Wicks was a wonderful mentor, who inspired her to pursue her PhD in radar signal processing, while a chance meeting with Prof. Mark Richards at an IEEE Radar Conference ultimately set her path towards studying at the Georgia Institute of Technology and finding Prof. Doug Williams and Dr. Bill Melvin (Georgia Tech Research Institute) as her Ph.D. advisors. Sevgi started her Ph.D. program by taking classes in radar and signal processing through Georgia Tech’s distance learning program. She took a break from classes in 2003, when she deployed for 6 months overseas as a Turkish linguist in Diyarbakir, Turkey. In 2004, she separated from active service to pursue her Ph.D. full-time at Georgia Tech. Her life in Atlanta would include many major events: meeting her future husband, Ali Cafer Gurbuz, in the very first class she took on campus, getting

60

M. J. Lyons

married within the same year, having her first child in January 2006, and losing her mother, Nazmiye Ertan, to brain cancer in February of 2007. Throughout all of this, she was fortunate and grateful to her advisor Dr. Melvin, who was not only supportive and provided great technical instruction in the art of radar but was also compassionate and offered wise advice about life. Sevgi defended her PhD thesis on the “Radar Detection and Identification of Human Signatures Using Moving Platforms” in August 2009, graduating in December 2009. Afterwards, as her mother’s only remaining family was in Turkey, she decided to take a job as a Senior Research Scientist in the Turkish Scientific and Technological Research Council (TUBITAK) Space Technologies Research Center (UZAY) in Ankara, Turkey. Later, in 2012, she stayed at TUBITAK part-time, while beginning a faculty position in the Dept. of Electrical-Electronics Engineering in the TOBB University of Economics and Technology, also in Ankara, Turkey. While in Turkey, she was blessed with the birth of two more children (both girls). Afterwards, in 2016, as a much larger family of five, Sevgi moved back to the USA, taking a visiting Assistant Professor position at Utah State University, Logan, UT. Currently, Dr. Gurbuz is an Assistant Professor in the Department of Electrical and Computer Engineering at the University of Alabama, where she directs the Laboratory for Computational Intelligence in Radar (CI4R). Her current research focuses on the advancement of RF-enabled Cyber-Physical Human Systems (CPHS), radar signal processing and machine learning algorithms to address the challenges of robust, accurate human micro-Doppler signature analysis, automatic target recognition (ATR), and control of CPHS for automotive, health, human computer-interaction, and defense applications. She has pioneered radar-based American Sign Language (ASL) recognition, for which she was awarded a patent in 2022, and is developing novel, interactive RF sensing paradigms built upon physics-aware machine learning and fully-adaptive (cognitive) radar that provide for unique AI/ML solutions to radar perception problems. She is editor of a recently published book “Deep Neural Networks Design for Radar Applications,” and is the author of over eight book chapters, over 25 journal, and 90 conference papers. She is the recipient of a 2023 NSF CAREER award, the 2022 American Association of University Women Research Publication Grant in Medicine and Biology, the 2021 Harry Rowe Mimno Award for Excellence in Technical Communications, the 2020 SPIE Rising Researcher Award, a 2010 Marie Curie Fellowship, and the 2010 IEEE Radar Conference Best Student Paper Award. She also serves as a member of the IEEE Radar Systems Panel and is an Associate Editor for the IEEE Transactions on Aerospace and Electronic Systems, IEEE Transactions on Radar Systems, and IET Radar, Sonar and Navigation. Dr. Gurbuz is a Senior Member of the IEEE, and a member of the SPIE and ACM.

2 Twenty-First-Century Women in Telecommunications

61

2.25 Dr. Kristine Bell, Senior Scientist, Metron, Inc., Reston, VA, USA

Kristine L. Bell is a Senior Fellow at Metron, Inc. and also holds an Affiliate Faculty position in the Statistics Department at George Mason University (GMU). Her technical expertise is in the area of statistical signal processing and multi-target tracking, and her current focus is on cognitive and fully adaptive radar and sonar systems. She received the B.S. in Electrical Engineering from Rice University in 1985, and the M.S. in Electrical Engineering, and Ph.D. in Information Technology from GMU in 1990 and 1995, respectively. From 1996 to 2009, Dr. Bell was an Associate/Assistant Professor in the Statistics Department and C4I Center at GMU. During this time, she was also a visiting researcher at the U.S. Naval Research Laboratory and the U.S. Army Research Laboratory, and served as a consultant to SAIC, Lockheed-Martin Orincon, Metron, ArgonST, and Signal Processing Consultants. Since 2013, Dr. Bell has been collaborating with colleagues at The Ohio State University (OSU) on a program of research to develop and analyze cognitive radar processing concepts analytically and experimentally. The Metron/OSU team has developed a theoretical framework for cognitive sensor/processor systems, a computational modeling and simulation architecture for algorithm implementation, and a cognitive radar experimental test bed for real-time operation of cognitive radar algorithms. As part of this effort, Dr. Bell has been a member of the NATO SET227 Panel on Cognitive Radar and has published numerous conference, journal, and magazine articles on cognitive radar. Another focus of her work for the last two decades has been developing the Maximum a Posteriori Penalty Function (MAPPF) multi-target tracking algorithm. The MAP-PF technique provides a tractable and practical multi-target tracking solution that eliminates the computational complexity of traditional data association techniques and provides a natural method for tracking target features for use in classification. Dr. Bell has served on the IEEE Dennis J. Picard Radar Technologies Medal Selection Committee, the IEEE Jack S. Kilby Signal Processing Medal Selection Committee, the IEEE Aerospace and Electronic Systems Society (AESS) Fellow Evaluation Committee, and the AESS Radar Systems Panel, where she was the chair of the Student Paper Competition Committee. She was the chair of the IEEE Signal

62

M. J. Lyons

Processing Society’s Sensor Array and Multichannel (SAM) Technical Committee. She is a co-author of the books Bayesian Multiple Target Tracking, 2nd Ed. (2014); Detection, Estimation, and Modulation Theory, Part I, 2nd Ed. (2013); and Bayesian Bounds for Parameter Estimation and Nonlinear Filtering/Tracking (2007). She received the George Mason University Volgenau School of IT and Engineering Outstanding Alumnus Award in 2009 and the IEEE AESS Harry Rowe Mimno Best Magazine Paper Award in 2021. She is a Fellow of the IEEE.

2.26 Ernestina Cianca, Assistant Professor at the Department of Electronic Engineering of the University of Rome “Tor Vergata” Ernestina Cianca is Assistant Professor at the Dept. of Electronic Engineering of the University of Rome “Tor Vergata,” where she teaches Digital Communications and ICT Infrastructure and Applications (WSN, Smart Grid, ITS etc.). She is the Director of the II Level Master in Engineering and International Space Law in Satellite systems for Communication, Navigation, and Sensing. She is vice-director of the interdepartmental Center CTIF-Italy.

Cianca graduated “cum laude” in Electronic Engineering at the University of L’Aquila in 1997 with a thesis on Power Control on CDMA-based Mobile Communications. At that time, there was an explosion of radio communications, and in Italy there were many work opportunity in this field and the main motivation to choose this field was related to that. As soon as she graduated, she went to work for one of the big telecom companies, Italtel-Siemens, working on SDH networks. However, she soon recognized a passion for research and continued academic education. After 1 year, she left the company to start a Ph.D. at the University of Rome Tor Vergata focused on satellite communications. The 2 years spent at Aalborg University as Research Engineer (2000–2001) and as Assistant Professor (2001–2003) in the Wireless Networking Groups (WING) coordinated by Prof. R. Prasad was an experience that completely changed her life. The professionally rich and hectic research environment gave her the opportunity to better understand the

2 Twenty-First-Century Women in Telecommunications

63

real impact of her area of research. Moreover, a research collaboration was initiated in that period that resulting in the foundation of the interdepartmental Center of Teleinfrastructure (CTIF) of the University of Rome Tor Vergata, in 2013, of which she is currently vice-director. CTIF is an interdisciplinary center focused on the use of ICT for vertical applications (biomedicine, health, food, energy, cultural heritage, economics, law) starting from the network (terrestrial, aerial, space) and service (communications, positioning, and sensing) integration, with the aim of improving the Quality of Life (QoL). The “mission” of the Center contains the word “Quality of Life,” but today “Sustainability” would have been written into the mission statement, which includes QoL. Using ICT technologies for improving the QoL/sustainability has always been the thread of Ernestina’s research activity. She has worked on wireless access technologies (CDMA, OFDM) and in particular, in the waveforms design, optimization, and performance analysis of radio interfaces both for terrestrial and satellite communications. An important part of her research has focused on the use of EHF bands (Q/V band, W band) for satellite communications and on the integration of satellite/terrestrial/HAP (High Altitude Platforms) systems. A real terrestrial/satellite/aerial integrated network will open many novel services and applications that could play a fundamental role for a more sustainable future. Currently her main research interests are in the use of radiofrequency signals (opportunistic signals such as Wi-Fi or specifically designed signals) for sensing purposes and in particular device-free RF-based activity recognition/crowd counting/density estimation and localization and UWB radar imaging (i.e., stroke detection). In the current pandemic situation, RF-sensing technology could provide interesting technological solutions to avoid crowded situations. She has been the coordinator of the activities of the Interdepartmental Center CTIF for the Italian Space Agency project “Sviluppo Terminale EGNSS multifunzionale e riconfigurabile (TESEI),” on the development of a GNSS multifunctional terminal. She has been principal investigator of the ASI project WAVE-A2, phase 2 study for two demonstrators and two pre-operative missions for satellite communications in W band.

References Casagrande M, Conti M, Losiouk E (2021) Contact tracing made un-relay-able. In: CODASPY 2021 El-Zawawy MA, Losiouk E, Conti M (2021a) Vulnerabilities in Android webview objects: still not the end! Comput Secur 109:102395 El-Zawawy MA, Losiouk E, Conti M (2021b) Do not let next-intent vulnerability be your next nightmare: type system-based approach to detect it in Android apps. Int J Inf Secur 20:39–58 Pham A, Dacosta I, Losiouk E, Stephan J, Huguenin K, Hubaux JP (2019) Hiding the presence of sensitive apps on android. In: USENIX security 2019 Ruggia A, Losiouk E, Verderame L, Conti M, Merlo A (2021) Repack me if you can: an antirepackaging solution based on android virtualization. In: ACSAC 2021

64

M. J. Lyons Margaret J. Lyons, PE, has more than 30 years’ experience in wireless communications, including: two-way radio, paging, and microwave radio systems engineering and consulting. She earned her Bachelor of Science in Computer and Electrical Engineering at Purdue University. She has been an active member of the Society of Women Engineers since 1984 including a term on the National Board of Directors. She has been a member of IEEE since 1986 and was a charter member of the NJ Coast Section Women in Engineering Affinity group (2009). At an early age, Margaret showed an affinity for puzzles and was much more interested in math and science studies. When she was a young girl at playtime with friends, her first action when playing “house” was to grab a piece of chalk and draw the floor plans. A later career in engineering was no surprise. She began her collegiate studies in 1982 at the University of Scranton, Scranton, Pennsylvania, USA, (just the 10th freshman class admitting women) not certain if she would ultimately pursue Chemical Engineering or Electrical Engineering. Freshman Chemistry and Chemistry Lab experiences solidified for her that Electrical Engineering was the better path. A transfer to Purdue University, West Lafayette, Indiana, USA, brought with it the focus on Computer and Electrical Engineering at a time before those two disciplines were typically linked at the University level. Credit her father’s recognition that the future would include computers in everything for encouraging that dual degree pursuit. In 1986, her career started at RAM Communications Consultants, Inc., Avenel, New Jersey, USA (later RCC Consultants, Inc., of Woodbridge, New Jersey, USA), supporting the RF engineers as well as programming and IT support for the nascent computer systems in the engineering department. Over the course of 29 years at RAM/RCC, Margaret provided systems engineering design and implementation services for analog and digital paging, Itinerant Mobile Telephone Service (IMTS, the US precursor to cellular systems), cellular telephone/data systems (1G through 5G), conventional single frequency repeater systems, and complex private multi-channel trunked radio systems across the continental US and Hawaii. Her support to these industries continued through employment with V-COMM L.L.C., and Jacobs, until her retirement in 2021. Margaret became a licensed Professional Engineer in New Jersey in 1998 and subsequently registered in six additional states: Connecticut, Delaware, New York, Pennsylvania, Virginia, and Washington.

Chapter 3

TLC Transversal and Strategic Role Carolina Botti

3.1 Introduction: How It All Began As a starting point for this brief account of my experience in the world of TLC, I chose a highly critical moment for an adolescent embarking on adult life: the choice of university. In the majority of cases, this choice has a crucial impact on the development of an individual’s professional skills and can influence their entire career and, consequently, the course of their life. When I had to make the decision, it was a landmark moment in history (the 1980s). For a woman, the “STEM” field of study (science, technology, engineering, and mathematics) was not an obvious choice inasmuch as the more natural option was a discipline that would make it possible to combine work and family life, such as teaching. At that time, I was extremely fortunate to have advice from someone who had my best interests at heart and who knew my potential (he later became my husband). He encouraged me to strike out in this new and challenging direction: I needed a little push for encouragement! And so I chose the engineering faculty in Naples, the city where I lived. It had an excellent reputation. Those university years were extremely intense (years after graduation I still had the recurring nightmare that I hadn’t finished my degree courses!), but at the same time, I fell in love at first sight with the scientific subjects and mathematics in particular: Analysis 1, Analysis 2, and so on. The bar was raised higher and higher, challenging me, but at the same time making the “victories” of passing exams and acquiring knowledge all the more gratifying. Having graduated in engineering with the highest possible grades, I had no problems facing the world of work as I was spoiled for choice with a number of high-profile offers. After a brief period of work (which began in the world of

C. Botti () Ales SpA, Rome, Italy e-mail: [email protected] © The Author(s), under exclusive license to Springer Nature Switzerland AG 2023 M. S. Greco et al. (eds.), Women in Telecommunications, Women in Engineering and Science, https://doi.org/10.1007/978-3-031-21975-7_3

65

66

C. Botti

Fig. 3.1 Marisa Bellisario

robotics), I began to have contact with the “real” world of TLC through Italtel, one of the leading telecommunications parastatal companies in Italy. It was 1989 and in order to commemorate the first anniversary of the untimely death of the CEO Marisa Bellisario, Italtel launched a national scholarship competition. It is interesting to provide a brief glimpse into the life of this woman, this icon for all female managers and particularly those working in the TLC sector (Fig. 3.1). Marisa Bellisario1 was born in 1935 in Ceva, a small town in the Province of Cuneo in Piedmont (Italy). She graduated in Turin in Economics and Commerce and her adventure in the world of new technologies began at Olivetti’s electronics division. In 1959, she started work in Milan in the world of computers, at the time an unexplored universe for the more intrepid. It was the gamble, the first in a long series of challenges and insights that this enterprising woman chose to tackle. In 1963, Olivetti merged with Bull, but in 1964, the winds of crisis were already blowing. The decision was made to sell the electronics division to General Electric. Marisa Bellisario became involved in the first international exchanges. In 1965, she went to New York for the first time, and very quickly her managerial skills were fully recognized in the United States. Her decision-making, her skills, and competence, coupled with her international experience, made her the undisputed protagonist of Honeywell. She rose so brilliantly in the ranks that, in January of 1979, this female manager was appointed President of the Olivetti Corporation of America, a position she held until 1981, when she returned to Italy to take over the reins of Italtel. In those years, the public company was going through a severe recession: the colossus had 30,000 employees, and the group consisted of 30 electromechanical companies, obsolete and in severe decline. As CEO, Marisa had to make courageous and farsighted choices. The unions were pitted against her, skeptical about her restructuring plan. The press wrote that a woman had been chosen to soften the blow of closing the entire complex. The only person to put his

1 https://www.fondazionebellisario.org/online/marisa-bellisario/Enciclopedia delle donne – Nicola

Misani. https://www.treccani.it/enciclopedia/marisa-bellisario_(Dizionario-Biografico)/Marisa Bellisario Donna e top manager –la mia storia Editore Rizzoli).

3 TLC Transversal and Strategic Role

67

trust in her was the Minister of State for Investment at that time, Gianni De Michelis. He was not wrong: Marisa succeeded in the miraculous task of transforming a conglomerate of outdated factories on the verge of falling into disuse into a modern, dynamic, cutting edge electronics company. She replaced 180 of the 300 managers, launched innovative projects that sparked interest in the United States as well, and in 3 years succeeded in the impossible mission of turning Italtel’s balance sheet around to show a profit, with a turnover of 1300 billion. It was an undisputed success, a case for the economics textbooks, an example of how to restructure a public company, and she earned the Manager of the Year Award in 1986. However, this victory did not make the road any easier for Marisa. She still had to fight against deeply rooted biases. A demonstration of this is the story of Telit, a large Italian telecommunications hub that should have been established with the merger of Italtel and Telettra, a Fiat company in the sector. The deal was killed due to Fiat’s obstinate refusal to appoint Marisa as CEO because she was a woman. It brought to an abrupt standstill the Italian telecommunications sector, which with Telit could have gained a prominent position on the international scene. It was also an exemplary obstacle to what Marisa Bellisario’s career was and represented. A career built alone, rejecting compromises and power games, easy accommodations and false solutions. In 1987, her position as CEO of Italtel was reconfirmed for another 3 years. She then tackled the restructuring of the company with the prospect of building a European telecommunications market, based on agreements between large companies in the sector; at the same time, Italtel was strengthening its relations with American companies (in 1987: the technological and marketing cooperation agreement with Apple and the agreement with Bell Atlantic). Hers was the first career, in our country, in the world of telecommunications and information technology, which she saw as the “future of nations”; hers was the first to be international, because she herself wrote of “having discovered twenty years earlier than economists and experts that a company must be international.” Marisa would certainly have risen to unthinkable heights for a woman if she had not been felled by an irreversible disease which took her life on August 4, 1988. This was the untimely ending to the story of a life, a woman, an entrepreneur, and an almost inimitable model even today. In fact, not only was Marisa the “tough but correct” manager, as the international press called her, she was also the woman who revolutionized the image of gray-suited CEOs with her explosive mix of firmness and sensitivity, coquetry, and managerial panache. Her face, her bold hairstyle, and her stylish clothes put her on front pages of newspapers all over the world, giving her the popularity of a diva. But above all, Marisa Bellisario showed how, with work, sacrifices and self-confidence, it is possible to get anywhere you want. She was candid and realistic, promising no easy roads. “It is harder for a woman to have a career but it is more fun,” she wrote. In her model of life and work, there was no place for gender differences, only values, and it was with this approach that she managed to sublimate the idea of equality. Going back again to 1989, it was the year in which Italtel launched a national competition in commemoration of Marisa Bellisario. This scholarship for women graduates in engineering would provide the winner with funding for a Master’s

68

C. Botti

degree at the Massachusetts Institute of Technology if they were accepted, and I won it! This is because Marisa Bellisario was firmly convinced that women needed to conquer new spaces in the world of work also through high-profile training. In fact, thanks to the scholarship, and after passing the selection process, I enrolled in the MBA program at the Sloan School at the Massachusetts Institute of Technology, where I was able to add managerial training to my technical training. It staggers the mind to think of the numerous advances made by the telecommunications sector in the years to follow. At that time the largest Italian TLC company gave me the opportunity to go to the United States to one of the most advanced universities of technology in the world. But making an intercontinental phone call still cost a fortune. Data transmission was not yet being used en masse; emails were virtually non-existent and every week I looked forward to the letters on paper that my loved ones sent me from Italy. Between that time and the next few years came the so-called technological convergence, in other words, the integration of services that provided telecommunications, Internet, devices, and applications. In particular, those who were voice TLC operators gradually became providers of integrated services based on data traffic.

3.2 The Professional World and the First Encounter with TLC On finishing my MBA, I wanted to go and work in TLC at Italtel, but unfortunately problems of a bureaucratic nature did not allow the company to make me as competitive an offer as the others I had received, and so I began my career as a consultant with the Gemini Consulting Group. Once again, apart from various projects in different sectors, I began to specialize in consulting within the TLC sector. During that period (the late 1990s), the world of TLC was about to open up to the imminent mass development of mobile telephony. It was the start of a revolution in communications that still has an influence on our private and professional lives and the entire global economic social system in general. During that time, I was in charge of leading a project to support the management of what was known then as the mobile radio transmitter division (today TIM) of the SIP state communications company (today Telecom Italia) in their commercial launch of a mobile telephony service in Italy. The challenge lay in creating a service for which very few points of reference or benchmarks existed. The sales structure, with a network of owned and franchised dealerships, created the “Telefonino (cell phone)” stores, and I can still remember the excitement when the management and I opened the first store in Rome on Via Minghetti! The great success of the mobile radio transmitter in Italy is due, in particular, to the engineer Mauro Sentinelli, with whom I had the pleasure of working in that period. Unfortunately, he passed away at the age of 74 in May of 2020. This person was an icon of the mobile radio transmitter, and for his role in the world of TLC, it is interesting to look at an excerpt from his life.

3 TLC Transversal and Strategic Role

69

Mauro Sentinelli2 was the “father” of mobile telephony in Italy. A competent, genial, and enterprising manager; with the launch of prepaid cards, he overcame the problematic impediment of the government concession fee, allowing mobile telephony to develop on a grand scale. Sentinelli understood the potential of the rechargeable subscription model which has become a large-scale business, making the mobile phone available to everyone and then spreading throughout the world. Even today, the majority of the customers of the consumer division of all mobile operators are made up of prepaid cards, because even today the government tax is disproportionate compared to the cost of the telephone service. This sales formula allowed TIM to become a European leader in subscribers. Engineers from all over the world listened to Sentinelli, and in those years he patented a number of innovations for the former TLC monopolist. At TIM, he was a charismatic leader, highly respected by his team and retailers. In 1974, after graduation, he joined what was then called SIP and left in 1998, soon after privatization, to pursue another dream: to launch mobile telephony via satellite with a start-up, Iridium. However, this did not achieve success as hoped. But in 1999, Marco De Benedetti, who left Omnitel for TIM, convinced him to return to TIM as the head of the mobile telephony strategies. In those years, mobile phones were becoming very popular not only as an instrument for business communication but also for personal use: TIM launched the claim “live without borders”, because, in fact, it provided the freedom to call from almost anywhere. The Sentinelli-De Benedetti leadership also managed to win a major battle on the international front: to avoid the sale of TIM Brasil, the Rio de Janeiro division that had become the largest and most promising outpost of Telecom’s foreign subsidiaries. Sentinelli apparently did not even know Portuguese, but when he went to Rio, he was still able to make himself understood by all the employees of TIM Brasil. “It was Sentinelli who convinced everyone, including De Benedetti, that Brazil was a resource that Telecom could not afford to lose,” recalls Stefano De Angelis, former CEO of TIM Brasil until 2018. “I remember once, at one of the conventions with dealers in Rome, that Sentinelli, to explain the new summer offer to the salespeople, said, ‘You haven’t understood. People at the seaside will take off their bathing suits, but they will always have their mobile phones in their hands.’ He was right.” “He was an engineer who not only had mastered the technology, but knew how to translate it into a marketing idea: he was the first to understand the importance of the fact that the mobile phone, an object originally created for an elite few, had to be everyone’s prerogative. In board meetings, when strategic decisions had to be made regarding a project or an idea, all eyes were on Sentinelli because his opinion made the difference. He was never frightened by finance. He maintained that investments in networks and new technologies had to be made, and then a solution would be found to market them: his charisma made Telecom an innovative company”. One of the founding fathers of the GSM standard (Global System for Mobile Com-

2 https://www.repubblica.it/economia/2020/05/17/news/addio_a_mauro_sentinelli_il_padre_ della_telefonia_mobile_di_TIM-256932674/

70

C. Botti

munications), with the advent of UMTS (European Telecommunications Standards Institute), he was always looking forward and imagining new technologies and new applications. Telecom rode the wave of the first of a series of revolutions related to 3G, focusing everything on video calling, which never took off until 4G, data usage and Internet connection. Sentinelli left Telecom in 2005 and returned in 2011, but this time as an independent director of Telecom Italia. In April of 1990, SIP became the European mobile radio operator with the largest number of subscribers, thanks to the introduction of the new analog ETACS (Extended Total Access Communication System, First Generation). In 1991, it activated its own digital telecommunications network (ISDN Integrated Services Digital Network), which made it possible to transmit voice and data on a single medium. On July 27, 1994, Telecom Italia was officially established following the 1993 reorganization plan of the telecommunications sector, with the consequent merger of SIP with Iritel, Telespazio, Italcable, and SIRM, other telecommunications companies that were already in operation. The following year, in 1995, Telecom Italia Mobile (TIM) was established, the first operator in the world to offer a tariff plan based on GSM technology (Second Generation), which ensured an exponential growth of mobile telephony.

3.3 The TLC Liberalization Process Living through the history of the launch of Telecom Italia and TIM first-hand allowed me to observe and understand the process of the liberalization of telecommunications,3 which took place in Europe in the early 1990s. The consequent demolition of public monopolies and elimination of reserved rights gradually opened the markets to new operators. Initially, Europe was concerned exclusively with liberalizing long-distance communications, mobile telephony, and market products, but later this extended to infrastructure and services. Italy was somewhat behind compared to other European countries. It was only after 1996 that all the directives were accepted and liberalization formally began in January of 1998, culminating in one of the major reforms in its economic history. The privatization of Telecom Italia, the issue of licenses for the mobile telephone service, the complete opening of the landline segment, and the establishment of an independent authority for the telecommunications, television, and press sectors were the crucial steps needed to open the market to competition.

3 http://www.proteo.rdbcub.it/article.php3?id_article=555#:~:text=In%20Italia%2C%20la%20 liberalizzazione%20nel,inizio%20degli%20anni%20’90. Giuseppina Galvano. Caruso, E. (2019). La liberalizzazione di poste e telecomunicazioni. Tratto da Impresa Oggi: http://www.impresaoggi.com/it2/1979-la_liberalizzazione_di_poste_e_telecomunicazioni/

3 TLC Transversal and Strategic Role

71

The liberalization process in Italy made it possible to establish one of the most open and competitive markets in Europe, with performances that were constantly close to those of the most industrialized countries. In fact, already in the early 2000s, over a hundred companies were offering alternative telephony services to Telecom, and, within 2 years, residential rates decreased by 40%, and business rates by 25%. Access to the market was limited only to obtaining licenses for both landlines and mobile telephony, as well as the acquisition of the spectrum of frequencies, which are considered a scarce resource.

3.4 From Liberalization to the Current Competitive Scenario With the liberalization4 of the markets, state monopolists saw their roles eroded and new companies tried to impose themselves on national and international scenarios to gain market share. Valid alternatives to Telecom Italia were springing up in Italy. One of the first companies to make a move in this competitive environment was Omnitel Sistemi Radiocellulari Italiani, founded in mid-1990 and supported by giants such as Lehman Brothers, Cellular Communications, and Bell Atlantic Corporation. During the early 1990s, Omnitel, along with the Unitel consortium led by the Berlusconi and Agnelli families, exerted considerable pressure on the antitrust to move beyond the outdated and backward jurisdiction by requesting the liberalization of the mobile telephony market. In 1994, the participants in the tender for the assignment of the second operator license were Unitel, Omnitel, and Pronto Italia (led by the German Mannesmann and the American Air Touch). Shortly before the tender, the latter two joined forces, creating the single Omnitel Pronto Italia consortium that won the tender and launched the commercial service in December 1995, becoming the second telephone operator in Italy. In 2000, the British company Vodafone acquired Mannesmann, entering into definitive control of Omnitel. It was the beginning of a series of turnovers that brought the name of the English company to the fore, first with the name change to Omnitel Vodafone, until the definitive adoption of the name Vodafone Italia in 2003. The third major operator, Wind Telecomunicazioni, was launched at the end of 1997, a joint venture between ENEL, France Telecom and Deutsche Telekom. After 4 Cisotti,

E. (2019, 03 25). La storia dell’operatore Blu (che a qualcuno ha ricordato Iliad). Taken from Mobileworld.it: https://www.mobileworld.it/2018/09/11/storia-blu-operatore-190994/ Nicola De Mori (a.a 2018/2019) Tesi Corso di Laurea Magistrale in Ingegneria Gestionale “Diversificazione di offerta per affrontare la concorrenza: il caso Vodafone”. Martucci, G. (2019, Agosto 6). Omnitel, la storia della prima alternativa telefonica che ha inaugurato la concorrenza in Italia. Tratto da Mondomobileweb.it: https://www.mondomobileweb. it/73192-storia-omnitel-prima-alternativa-telefonica-italia/ Vodafone Italia. (2018). La nostra storia. Tratto da Vodafone.it: http://www.vodafone.it/portal/ Vodafone-Italia/Chi-siamo/Vodafone-Italia/La-nostra-storia Wind Group. (2016). La nostra storia. Tratto da Windgroup.it: http://www.windgroup.it/it/ azienda/storia.phtml

72

C. Botti

a year, Wind received the authorization for landline services and won the tender to obtain the GSM (Global System for Mobile communications, Second Generation) license for mobile telephony in Italy: in 1999, just 6 months after the commercial launch, it reached its first million customers. From the very start, Wind had been aiming at a business customer acquisition strategy with greater profitability, also turning later to private ones. In October 2000, Wind acquired Infostrada, a telephone company founded in 1996 to compete with Telecom Italia in the landline sector and achieved complete coverage of the national territory at the end of the same year, successfully responding to the increase in demand for Internet and mobile services. The early 2000s saw the development of the Third Generation cellular mobile phone standard, called UMTS (Universal Mobile Telecommunications System), an evolution of the GSM system. Compared to the Second Generation standard, this had the advantage of being based on the higher data transmission speed, which enabled a much wider range of services compared to GSM. UMTS made new services possible, including video calling, packet data transmission such as MMS and digital images, Internet access, and web browsing. Between 1999 and 2000, other companies were ready to compete for UMTS licenses, but for some it did not go well. The first was Blu, a telephone operator founded in 1999 with the support of important market players such as Mediaset, British Telecom, and Benetton. The main objective of the telephone company was to lay the foundations on the Italian telecommunications market and then bet everything on the upcoming 3G technology that would radically change the way the mobile phone was used. Blu was also part of my consultancy work and had a very interesting management support project for the business plan that was to be the basis of the offer for the auction of UMTS frequencies. However, close to the allocation date, Blu withdrew in favor of the competition from the other participants. Blu’s turnaround was probably attributable to a change of heart by one of its main shareholders. In 2002, it officially ceased its activities after being dismantled and sold in pieces to the other existing operators. It was in this context that the Andala UMTS company was founded in 1999. The following year, it managed to win UMTS licenses and then be acquired by Hutchison Whampoa, a Chinese conglomerate also operating in the telecommunications sector, which became the largest shareholder. The Chinese company, also known by the acronym H3G, markets mobile telephony services under the 3 brand: this was behind the creation of one the mobile operators destined to contribute to the history of Italian telephony, 3 (Tre) Italia, which in 2004 had three million customers. On December 31, 2016, the merger of the operators Wind and 3 Italia was completed, and the new company Wind Tre S.p.A. was founded, when, after various negotiations, the companies Veon and C.K. Hutchison, respectively, owners of Wind (Wind Telecomunicazioni S.p.A.) and 3 Italia (H3G S.p.A.), completed their merger process.

3 TLC Transversal and Strategic Role

73

3.5 From Direct Commitment in the TLC Sector to Their Leverage in Other Sectors and Particularly the Tourism/Cultural Sector After a managerial experience in the position of CEO for a start-up in the pharmaceutical ICT world, in 2005, I decided to move from the private world to the public world and was hired as Central Director at Arcus (today named Ales following a merger in 2016), a start-up fully owned by the Ministry of Culture. Despite having a variety of tasks related to the enhancement of cultural heritage, I was faced with a new TLC challenge: satellite. In fact, at the helm of a consortium made up of 13 companies and together with Next S.p.A., Arcus won a European tender with the Galileo-CUSPIS project (Cultural heritage space identification system). This is an initiative aimed at defining the operational standards associated with the use of the Galileo satellite system in the field of cultural heritage, run cooperatively with the European Union and the ESA, the European Space Agency. Galileo5 is Europe’s Global Navigation Satellite System (GNSS), providing improved positioning and timing information with significant positive implications for many European services and users. Galileo is the system that provides users with a reliable alternative to the non-civilian American GPS or Russian GLONASS, but at the same time is designed to be compatible and interoperable with them. The innovative idea behind CUSPIS was to explore potential applications and market value of this important European service infrastructure for the cultural sector. The protection, fruition, and support of the cultural heritage are becoming more important in European society, which is particularly rich in cultural assets. The European Commission has given great importance to the subject, promoting actions for protection, improving understanding and dissemination of the culture and history of European citizens, and making cultural heritage increasingly available and accessible. The cultural assets can be seen, as a consequence, as one of the most valuable infrastructures owned by European citizens. Galileo represents a unique opportunity to implement both institutional and commercial applications for the cultural assets protection, enhancement, and fruition with value-for-money solutions under the appropriate regulation umbrella. In this context, the advantages of Galileo and EGNOS, with respect to a solution based on GPS, are maximized together with the valuable role that a deployed solution can play for this user community. The CUSPIS project, realized by the consortium led by Next S.p.A. and Arcus (society for the development of art, culture and entertainment established in 2004 by the Italian Ministry for Culture), deployed operative solutions for the transportation and fruition of secure assets in open archeological areas using EGNOS and Galileo. CUSPIS is unique within the cultural heritage user community in merging GNSS aspects, authentication capabilities, and infomobility services in a single solution.

5 https://www.euspa.europa.eu/cultural-heritage-space-identification-system

74

C. Botti

CUSPIS aimed to demonstrate the benefits that EGNOS and Galileo can bring to the cultural heritage user community, with a strong accent on validation of results, economical and technical viability of systems, and its dissemination. The goal of the project was to study the situation in the cultural heritage sector, identifying the areas where innovative solutions, based on satellite navigation services, can prove to be beneficial. The CUSPIS project established proper user community groups, involved significant actors from the relevant community and institutional users, transferred to the cultural heritage world a new awareness of the innovative applications offered by GALILEO, and put an accent on authentication aspects, IPR, and Digital Rights Management and European multilingualism. The key to CUSPIS was the use of the so-called Geo-Time Authentication (GTA) for the unique cataloguing, authentication, and determination of the precise location of indoor and outdoor cultural assets. The concept was demonstrated for secure management, tracking, and transportation of cultural assets between various sites (cultural asset management) and provision of information to both tourists and experts in the cultural heritage field (cultural asset fruition). The CUSPIS work tasks were the following: • Establish proper user community groups in the cultural asset context, enhancing their awareness of the potential of satellite navigation. • Provide deep analysis on the issues related mainly to security, management, and fruition. • Identify all potential applications enabled by Galileo services in the cultural asset sector. • Analyze the potential applications, examining all the elements that can affect the final services involving the key actors of the value chain, including the user community-related ones, and assessing the associated market potentiality and building the service models in order to complete a relevant application scenario. • Design and develop major demonstrators and a key technological proof-ofconcept related to the secure cultural asset management, services for the tourist and authentication. • Assess and evaluate the results available from the demonstration campaigns. • Perform a business and cost benefit analysis. • Disseminate the project results via a dedicated CUSPIS portal, organizing workshops, participating in conferences and selected national and international events. Subsequently, as part of the dissemination activity related to the project in question, a mission was carried out in Beijing. This mission was inspired by a Memorandum of Understanding signed between the pro-tempore Minister of Culture and the Minister for Cultural Heritage of Beijing containing instructions regarding the parties’ wish to strengthen the technological exchange and cooperation to prevent theft, illicit excavations, and illegal imports/exports of cultural goods, with a special focus on facilitating the application of the technologies of the European global positioning system Galileo for the prevention mentioned above.

3 TLC Transversal and Strategic Role

75

Increasingly, within the sphere of my work with Ales even today, I have been dealing with the key role that telecommunications play in the development of services, and in general, its role as an enabling factor for industry 4.0. characterized by the growing digitalization and interconnection of products, services, value chains, and business models through Cloud Computing, the Internet of Things, Artificial Intelligence, and Augmented & Virtual Reality (AR & VR). Technology applied to the field of cultural heritage is especially interesting in the archeological sector, where it becomes a fundamental research tool and because the developments can be interesting in regard to the use of objects that are often understood only by scholars because they are ruins from the past and all aspects of monitoring and control as well. As the CUSPIS project also revealed, for things to function, the sites where the operations take place need adequate TLC network coverage which is not always present in these places but increasingly strategic. I am currently working on a very interesting project that aims to demonstrate how AR and VR can and should be used for research and the subsequent development of services accessible to end users. The use of Virtual Reality in Archeology dates back to the late 1980s and the first person to merge these two worlds was Paul Reilly who introduced the term “Virtual Archeology” in 1990 to describe simulation models of cultural heritage (Reilly 1990, p.133). Since the first experiments there has been a gradual interest in combining these two realities with the aim of achieving increasingly accurate scientific visualizations of the past. Based on the studies, virtual archeology could be defined as a study and interpretation process aimed at reconstructing and simulating the past using digital technologies and a theoretical and multidisciplinary scientific approach. Continuous experimentation, on the one hand, has opened up perspectives and research problems that were inconceivable until now (detection, simulation, analysis, and conservation), and on the other hand, it has led to the design and development of technologies specifically for the cultural sector or for the assimilation and adaptation of paradigms and technologies developed in other spheres. The goal of the project, managed by Ales on behalf of the Ministry of Culture (MIC), is to employ these technologies so that the archaeological cultural heritage may be used through virtual reconstruction products created with a scientifically rigorous and traceable methodological approach. In particular, the focus was on eight of the Ministry of Culture’s archeological sites that still had undeveloped potential. The outputs from the project will not only provide an enriched offer for public use and scientific knowledge of the sites, by publishing the scientific work behind the virtual reconstructions so that this could be understood at a hypothetical level and enjoyed by the entire community, but also an important tool to promote our cultural heritage and the surrounding socioeconomic fabric, to be disseminated in Italy and abroad.

76

C. Botti

3.6 Theoretical Approach and Reference Guidelines Unfortunately, we often see works that strive only for the “Wow” factor, but which lack a rigorous interpretation of the study sources, or works that are the domain of a rare few and cannot be reused. Therefore, the most interesting aspect to the work – which involved ten research institutes with scientific coordination by the ISPC-CNR (Istituto di Scienze del Patrimonio Culturale- Centro Ricerche), five companies specialized in 3D reconstruction and rendering as well as other companies that handled aspects related to the website, filming and bringing in cabling to the sites – was implementing a rigorous approach to virtual reconstructions based on the 2008 London Charter (whose foundations had already been laid in 2003 in the UNESCO Charter on the Preservation of Digital Heritage, which governs the principles of visualization in Cultural Heritage). After a shaky beginning due to the lack of widespread and adequate technological solutions and an information science still unprepared to formulate communication languages appropriate to this new reality, we are witnessing an increasingly prudent use of virtual reconstructions of Cultural Heritage. Over the years, in fact, the problems relating to the scientific approach in the creation of simulation models of the past and their communication have been discussed in order to ensure that the methods of visualization are applied with academic rigor. The London Charter aims to define the objectives and basic principles relating to the use of three-dimensional visualization methods in relation to intellectual integrity, reliability, transparency, documentation, standards, sustainability, and access.

3.6.1 Objectives • Provide a foundation with broad recognition among its stakeholders. • Promote technical and intellectual rigor in the digital visualization of cultural heritage. Ensure that digital visualization processes and results can be understood and evaluated by users. • Provide digital visualization with scientific authority in the study, interpretation, and management of cultural heritage. • Ensure that accessibility and sustainability strategies are established and applied. • Provide a solid foundation upon which communities engaged in the sector can build more detailed guidelines for the implementation of the London Charter.

3.6.2 Principles Principle 1: Implementation

3 TLC Transversal and Strategic Role

77

The principles of the London Charter apply whenever digital visualization is applied to cultural heritage research and dissemination. Principle 2: Aims and methods A digital display method should normally only be used when it is the most appropriate available method for that purpose. Principle 3: Research sources In order to ensure the intellectual integrity of digital visualization methods and results, the relevant sources must be identified and evaluated in a way that is documented and structured. Principle 4: Documentation Sufficient information should be provided in order that digital visualization methods and results be understood and evaluated appropriately with respect to the contexts and purposes in which and for which they are disseminated. Principle 5: Sustainability Strategies should be planned and implemented to ensure the long-term sustainability of digital visualization documentation and results regarding cultural heritage to prevent the loss of this growing part of humanity’s cultural, economic, social, and intellectual heritage. Principle 6: Accessibility In the creation and dissemination of digital visualizations, the ways in which the results of the work can contribute to the study, knowledge, interpretation, and management of cultural heritage should be kept in consideration. The Seville Principles of 2009 mandate the need to implement and apply the contents of the London Charter to the integrated management of Archeological Heritage. Eight principles bring together the guidelines of a vast scientific community that believes in the affirmation of virtual archeology as an independent discipline based on its scientifically valid and widely shared regulations and methodologies. The following are some of the fundamental principles: • Interdisciplinarity (Principle 1): “Any project involving the use of new technologies, linked to computer-based visualization in the field of archeological heritage, whether for research, documentation, conservation or dissemination, must be supported by a team of professionals from different branches of knowledge.” • The purpose of the work (Principle 2): “Prior to the development of any computer-based visualization, the ultimate purpose or goal of our work must always be completely clear. Therefore, different levels of detail, resolutions and accuracies might be required.”

78

C. Botti

• Authenticity (Principle 4): “Computer-based visualization normally reconstructs or recreates historical buildings, artifacts and environments as we believe they were in the past. For that reason, it should always be possible to distinguish what is real, genuine or authentic from what is not. In this sense, authenticity must be a permanent operational concept in any virtual archaeology project.” • Historical rigor (Principle 5): “To achieve optimum levels of historical rigor and veracity, any form of computer-based visualization of the past must be supported by solid research, and historical and archaeological documentation.” • Scientific transparency (Principle 7): “All computer-based visualization must be essentially transparent, i.e. testable by other researchers or professionals, since the validity, and therefore the scope, of the conclusions produced by such visualization will depend largely on the ability of others to confirm or refute the results obtained.” The application of these reference principles aims to make the scientific process of a virtual reconstruction (1) Documented, (2) Explicit, (3) Testable, and (4) Replicable.

3.7 Methodological and Operational Approach The scientific community agrees that the creation of semantic mapping that can indicate the levels of reliability, the sources used, the comparative cases, and the methods/interpretative processes (i.e., how the sources were used and integrated) is crucial to documenting the reconstructive processes (Pietroni and Ferdani 2021). From a general point of view, the reconstructions can be based on: • Elements that are still visible or documented as visible in the past and which can be placed with certainty in situ • Architectural forms and the theory of proportions • Figurative deductions • Comparative cases and cultural patterns For practical purposes, an overall image of the virtual reconstruction process can be divided into five main phases: (1) collecting sources (including the possible 3D survey of monumental or archeological remains), (2) stratigraphic reading of the elements (useful for decomposing 3D models into semantic units), (3) creating a reconstructive hypothesis represented by a volumetric model with a reliability index, (4) creating a representation model complete with materials that are as photorealistic as possible and, finally, (5) publishing the reconstruction on one or more dissemination channels (online, in a museum installation, in a video, etc.). With the CNR-ISP researchers, who worked alongside me in the scientific coordination of the project, the following guidelines were provided to the research bodies and companies involved in the virtual reconstructions.

3 TLC Transversal and Strategic Role

79

3.8 Guidelines Provided to the Research Bodies Involved With regard to the mapping of the scientific back end (levels of reliability, sources and interpretative processes) on the 3D reconstruction models, universities were urged to follow, in the first phase, a “simplified” approach, that at the same time was compatible with the abovementioned five-phase process, in order to retrieve accurate information regarding the sources and processes used. To this end, the proposal was to map this information on 2D images of the virtual reconstruction (renderings that were static from various points of view) in the minimum number necessary to have a visual coverage of all the significant elements of the model (front, side, rear, zenithal, exterior, and interior). On each image, the chromatic backgrounds corresponding to the levels of reliability and the relative sources must then be defined as follows: Red to identify the archeological elements documented in situ (present or past) Blue to identify the virtual restoration of damaged areas derived from physical evidence (e.g., partial collapse of a wall) Yellow for anastylosis Dark yellow for the repositioning of pieces that exist but were placed elsewhere (e.g., museum) Light yellow for the reconstruction of a missing element Green to identify elements reconstructed in the absence of concrete evidence, for deductive processes: literary sources, architectural modules, comparative cases, and typological similarities A practical example is given below to highlight the parts that make up the documentation of the reconstructive process, the Forum of Augustus, in which there is a digital replica and a reconstructive model (Fig. 3.2). 1. Select relevant viewpoints that show all the reconstructive elements, and save them in image format. 2. Source mapping. The second stage in the simplified approach is to prepare colored backgrounds corresponding to levels of reliability, according to the color scheme above. On the backgrounds of the various elements, there must be indications of the numbers associated with a legend which show the following: interpretative sources used, comparative cases, methods/interpretative processes (how the sources were used and integrated) (Fig. 3.3). For each number, there must be indications of (a) the sources used and (b) the reconstructive process.

3.8.1 The Sources (a) For bibliographic and literary sources, cite the complete bibliographic reference (possibly with a link to access it) + passage of writing that speaks of that specific element (Fig. 3.4) (Ferdani et al. 2020).

80

C. Botti

Fig. 3.2 Forum of Augustus: virtual reconstruction, by CNR ISPC in collaboration with the Capitoline Superintendence

Fig. 3.3 Forum of Augustus mapping of sources: by CNR ISPC

Fig. 3.4 Forum of Augustus source mapping: associating interpretative sources with the various elements of the architectural model through a simple legend system, elaborated by CNR ISPC

For iconographic sources show images of: • Comparative cases with caption • Plans of architectural modules (from ancient treatises etc.)

3 TLC Transversal and Strategic Role

81

Extant structures (archaeological evidences in situ)

Fragmental virtual anastylosis

Structural virtual reconstruction

(repositioning of collapsed architectural fragments which were found out of context)

(based on archaeological evidences in situ)

Fragmental virtual reconstruction

Non-structural virtual reconstruction

(completion of repositioned fragments)

(based on sources and comparisons)

Fig. 3.5 Forum of Augustus mapping of reliability levels, by CNR ISPC in collaboration with the Capitoline Superintendence

3.8.2 The Process (b) The process should be explained in three or four lines and, in particular, how the sources were integrated. This short text should consist of two parts: (b1) how the source was interpreted (e.g., which passage of a text was used or which part of an image) and (b2) the reasoning or how the source was used for reconstruction purposes. Based on all the preliminary work carried out by research institutions on 2D renderings of virtual reconstructions, companies specialized in 3D reconstruction and rendering will be required to translate all the previously mentioned information, relative to the scientific back end, to the 3D models. A semantic mapping is then applied to the 3D models or to their interactive 360◦ panoramic renderings, making it possible to superimpose the colors of the reliability levels onto the real materials, select the single element, and open the corresponding information on the side (sources, interpretative processes, as described above) (Fig. 3.5).

3.9 Conclusions Despite having dealt professionally in a variety of sectors and in different roles, my experience in TLC and my general familiarity with technology has allowed me to work on innovative projects that were abreast or even ahead of the times. More particularly, I learned to evaluate and apply the technology to the cultural sector by borrowing methodological experiences from research utilized in other sectors and evaluating new applications and consequently, new markets. I hope that the best practices and approach taken in the latest working experiences may inspire an increasingly greater use of scientific working models that are also open to reuse.

82

C. Botti

The expectation is also that the “digital divide” will grow increasingly smaller and allow for the development and use of new services. Acknowledgments A special thanks to the group of researchers from ISPC – CNR for their contribution to the virtual archeology project that is underway: Eva Pietroni (Project scientific coordinator) Alberto Bucciero Emanuel Demetrescu Bruno Fanini Daniele Ferdani

References Ferdani D, Fanini B, Piccioli MC, Carboni F, Vigliarolo P (2020) 3D reconstruction and validation of historical background for immersive VR applications and games: the case study of the Forum of Augustus in Rome. J Cult Herit 43:129 Pietroni E, Ferdani D (2021) Virtual restoration and virtual reconstruction in cultural heritage: terminology, methodologies, visual representation techniques and cognitive models. J Inf 12:167, ISSN 2078-2489, MDPI, special issue on “Virtual reality technologies and applications for cultural heritage”, guest editor Juan Carlo Torres. https://www.mdpi.com/2078-2489/12/4/ 167/pdf Reilly P (1990) Towards a virtual archaeology. In: Computer applications in archaeology. British Archaeological Reports, Oxford, pp 133–139

Carolina Botti started out her career across different industries both in the public and in the private sector, after earning an Italian Master Degree with honors in Electronic Engineering, followed by a MSc in Management by the Sloan School of Management at MIT (Massachusetts Institute of Technology). She believes that an engineering background coupled with a proven strategic advisory experience provides a strong skill set to face all technical aspects and, above all, the challenges of change. Carolina spent over than 10 years in consulting, first with Gemini Consulting and later joining Booz & Company as Principal. The consulting experience proved to be very intense and professionally and personally successful. Carolina was responsible for account management and account development in Telecommunication and, in particular, she led projects focused on strategy and marketing for major operators in fixed and mobile telephony. In the course of her assignments, Carolina had the opportunity of leading multicultural teams of brilliants consultants, while managing top manager clients. As a further managerial step forward in her career, Carolina later joined the Italian leading bank Monte dei Paschi di Siena, to become Managing Director of a controlled company in IT services. That professional experience proved to be most interesting and successful, and it reinforced Carolina’s determination to leverage and strengthen her professional background through managerial assignments. Carolina took up a new challenge when she became COO of the in-house company of the Italian Ministry of Culture (ARCUS,

3 TLC Transversal and Strategic Role

83

now incorporated in ALES as per the 2016 Italian budget law). ALES mission is to support the Ministry of Culture in managing and enhancing the Italian cultural heritage, as well as providing value-added services and innovative projects. Carolina leads the “Public-Private and Projects Funding Department”, focused on financing and monitoring strategic projects that use ITC to enhance the user experience of cultural heritage. This is a fascinating and challenging mission, carried out in a country (Italy) which has the majority of the world’s UNESCO cultural sites. As new investment strategies are being drawn and implemented under the post-Covid stimulus package, Carolina is moreover contributing to address the challenges ahead faced by the cultural sector. Carolina has been invested by a Knighthood for the Order of Merit of the Italian Republic. Moreover, she has received several national awards, including the “Excellence Award of Manageritalia.” She is an active member of no-profit organizations such as the “Marisa Bellisario Foundation and the “Italian Parks and Gardens Association.” Carolina routinely contributes opinions and editorial pages for Italian newspapers. She lectures at several master classes and is often invited as a keynote speaker to relevant conferences and events related to her professional interests.

Chapter 4

Recent Advances in Bayesian Inference for Complex Systems Mónica F. Bugallo

4.1 Introduction The world is challenged by complexity. At the root of the challenges is the remarkable growth in population, from today’s 7.5 billion to the expected 9 billion by the middle of the century. There are many serious questions and threats that arise in many domains including the sustainability of the planet, climate change, or global health (Hutt 2016). Computational and mathematical models and methods have become critical tools to investigate many of the underlying problems related to these complex challenges. In recent years, the analysis of complex systems has become a popular area of research within the signal processing community. These types of systems are prevalent in fields as diverse as artificial intelligence (Binford et al. 2013), astronomy (Jenkins and Peacock 2011), econometrics (Lopes and Tsay 2011), physics (von Toussaint 2011), or medicine (Hinne et al. 2013). In a typical application, one uses a collection of observations to recover and understand the underlying system for a larger purpose, such as informed decision-making, model selection or prediction (see Fig. 4.1). Regardless of the application, a common challenge shared by all complex systems is the large number of unknowns that characterize them and need to be estimated (high-dimensionality) and/or the large amounts of data available for the inference (big data). If one can address these challenges in a principled way, then there will be a critical gain in understanding real-world phenomena. A popular approach to inferring the unknowns of a system is Bayesian inference. The goal is to find the posterior distribution, i.e., the distribution of the unknowns of the system given the noisy observations. In many real-world problems, these

M. F. Bugallo () Department of Electrical and Computer Engineering, Stony Brook University, Stony Brook, NY, USA e-mail: [email protected] © The Author(s), under exclusive license to Springer Nature Switzerland AG 2023 M. S. Greco et al. (eds.), Women in Telecommunications, Women in Engineering and Science, https://doi.org/10.1007/978-3-031-21975-7_4

85

86

M. F. Bugallo

Fig. 4.1 Complex system components

distributions are impossible to obtain in analytical form and Monte Carlo (MC) methods using random samples allow for their approximation. However, complex systems pose tremendous challenges for existing approaches (Fearnhead 2008). In some settings, they are required to handle all the unknowns and/or data points as well as the possibly non-Gaussian, nonlinear and multimodal nature of the distributions of interest. This questions the suitability of Bayesian inference for these types of systems. The main focus of this chapter is on reviewing the Bayesian inference framework for models with large numbers of unknowns and/or observed data described by distributions characterized by nonlinearities, non-Gaussianities and multi-modalities. Four interconnected topics are addressed: Bayesian structures that can handle the high dimensionality of the vector of unknowns (and possibly the data), learning approaches (e.g., initialization, proposal learning, model uncertainty, or distributed learning) for high-dimensional vectors of unknowns, fusion strategies of random measures for big data and model selection techniques for complex systems. The chapter also includes two case study problems, one in the field of ecology on inference of demographic rates of penguin populations in the Antarctic and the other in the field of genetics on inferring the topology of a gene network.

4.2 Background Statistical inference deals with the estimation of a set of unknowns given a collection of noisy observed data (Casella and Berger 2002). In the most general scenario, this amounts to inferring both dynamic and static parameters1 of interest from the observations (Kay 1993; Scharf 1991; Van Trees 2004). For instance, in denoising applications, the aim is to reconstruct the original signal (e.g., an audio recording or an image) from noisy data (Godsill and Rayner 1998). An extended version of this problem occurs in blind deconvolution, where a noisy filtered signal is available and the goal is to recover both the unknown filter and the input (Haykin 1994). Finally, as a third application, target tracking in wireless sensor networks requires estimating the position of the target over time (maybe jointly with some parameters

1 Dynamic parameters are often referred to as state or hidden state and static parameters as model parameters (Adali and Haykin 2010; Liu and West 2001).

4 Recent Advances in Bayesian Inference for Complex Systems

87

of the system, such as the noise variance or even the position of the sensors) from sensor measurements (Wang et al. 2017). In the Bayesian framework, all the aforementioned problems are addressed by formulating a prior distribution, which should gather all the available information about the set of unknowns of interest before the data are observed, and by assuming a model (the likelihood), that incorporates a belief on the way the observed data relate to the unknown parameters (Bernardo and Smith 2001). Then, Bayes theorem is used to obtain the posterior distribution, which takes into account both the effect of the prior information and the observed data in an optimal way. Finally, the desired Bayesian estimators are obtained by minimizing a pre-defined cost function (such as mean-square error) that can typically be expressed either as some integral measure with respect to the posterior or as some optimization problem. Unfortunately, obtaining closed-form expressions for any of these estimators is usually impossible in real-world applications. This issue can be circumvented by using approximate estimators (e.g., heuristic estimators in the frequentist context or variational Bayesian approximations) or by restricting the class of models that are considered (e.g., using only conjugate priors). With the increase in computational power and the extensive development of MC tools, Bayesian inference has been freed from the use of a restricted class of models and much more complicated problems can now be tackled in a more realistic way. Still, many challenges must be addressed to turn Bayesian inference methods into competitive standard tools for complex systems.2

4.2.1 The Mathematical Problem The systems of interest are described by state-space models that relate the data to an evolving hidden state through a set of unknown model parameters, i.e., xt ∼ p(xt |xt−1 , θx ), t = 1, . . . , T, .

(4.1)

yt ∼ p(yt |xt , θy ),

(4.2)

.

t = 1, . . . , T,

where .xt ∈ Rdx is the hidden state at time .t with .x0 ∼ p(x0 ), .yt ∈ Rdy is the observation at time .t, .p(x0 ) is the initial distribution of the hidden state, .p(xt |xt−1 , θx ) is the state transition distribution parameterized by the vector .θx , .p(yt |xt , θy ) is the observation distribution parameterized by the vector .θy , and .T is the time horizon. The vector of unknown model parameters is given by the stacked   dθ vector .θ = [θ x , θy ] ∈ R (see Fig. 4.2).

2 Asymptotic methods, multiple quadrature approaches, and subregion adaptive integration (Evans and Swartz 1995) are also used for approximation of integrals but cannot be applied in highdimensional setups and only MC approaches become feasible in many practical applications.

88

M. F. Bugallo

Fig. 4.2 Considered system

We can better understand the system at hand by learning the joint posterior distribution of the hidden states and the model parameters given the observations, i.e., p(x0:T , θ|y1:T ) = p(x0:T |θ, y1:T )p(θ|y1:T ).

.

(4.3)

Obtaining an approximation to (4.3) is challenging due to the lack of efficient inference algorithms that can handle both hidden states and constant unknown parameters. It is worth noting that the posterior in (4.3) is also required for computation of predictions and system diagnostics, which is a task of upmost importance for most real-world systems, including those pertaining to the case studies addressed by this chapter in Sect. 4.2.3.

4.2.2 Bayesian Inference Schemes MC methods include a class of stochastic simulation techniques that proceed by obtaining a large set of samples (i.e., potential values of the desired unknowns) and substituting integrals with averages of those samples. In practice, these values are obtained either by physically replicating the desired experiment or by characterizing it probabilistically and generating a set of random realizations. The most well-known MC technique employed for Bayesian inference is Markov chain Monte Carlo (MCMC) sampling (Brooks et al. 2011). MCMC methods build a numerical approximation of the posterior distribution by constructing a Markov chain whose stationary distribution is the posterior. This is achieved through a random walk over the possible values of the unknowns, accepting samples that most likely represent the system parameters, and rejecting all others. This methodology has important theoretical implications, in that the approximating distribution converges to the true posterior in the limit of infinite sampling iterations (Brooks et al. 2011). The problem is that complex systems usually can only be represented by models with many unknowns. In these high-dimensional scenarios, MCMC is notoriously computationally intensive due to the low accept nature of the algorithm, making inference infeasible.

4 Recent Advances in Bayesian Inference for Complex Systems

89

Fig. 4.3 PF and AIS flowchart

An alternative to MCMC sampling, is another family of MC methods known as importance sampling (IS) and its adaptive versions (AIS, see right flowchart in Fig. 4.3) (Bugallo et al. 2017). AIS iteratively approximates the posterior distribution by using a set of samples and weights. At each iteration, the weights provide information about which samples are most representative of the posterior, allowing for adaptation of the sampling space to improve computational efficiency. AIS has some advantages over traditional MCMC schemes. First, in AIS, all proposed samples contribute to the approximation of the posterior—eliminating concern that any computation is wasted. Also, the simplicity of AIS allows for parallel implementations of the algorithm, a feature that is of utmost interest in big data applications (Bugallo et al. 2017). Finally, AIS methods allow for a straightforward computation of the Bayesian model evidence, which is a necessary quantity for resolving model selection tasks. Numerous AIS implementations have been proposed in the literature. Advances include the development of alternative weighting schemes and effective parameter updates of mixture proposal distributions, which are used to obtain samples (Cappé et al. 2008a, 2004; Cornuet et al. 2012; El-Laham et al. 2018; Martino et al. 2015). Some hybrid methods have been proposed combining MCMC and AIS (Martino et al. 2017). Other alternatives incorporate stochastic optimization techniques within AIS. For example, in Ryu (2016), Ryu and Boyd (2014) the parameters of a proposal from the exponential family are adapted to minimize the per-sample variance of the IS estimator. However, there are still many challenges in the AIS framework related to complex systems. In particular, state-of-the-art methods face problems scaling to highdimensional systems and dealing with distributions that are multimodal. Little work has been done to address the challenges posed by problems with large data-sets

90

M. F. Bugallo

and regarding how to adapt proposal distributions to efficiently sample when the geometric shape of the posterior distribution is complex. Particle filtering (PF, see left flowchart in Fig. 4.3) methods are an alternative to the Kalman filter for the estimation of dynamic parameters (Doucet and Johansen 2009; Gordon et al. 1993). These methods are also based on the IS methodology, with weights that are sequentially updated as new observations become available. To avoid the problem of “sample degeneracy” (Li et al. 2015), sequential Monte Carlo (SMC) methods were proposed as an improved PF using an additional algorithmic step referred to as resampling (Doucet and Johansen 2009). Despite the success of PF in many applications (Andrieu et al. 2001; Karlsson 2005; Orchard and Vachtsevanos 2007), its performance heavily depends on a number of factors. For instance, the presence of unknown static parameters could introduce some difficulties, since the framework for PF methods is traditionally setup for dynamic state estimation (Liu and West 2001). Moreover, PF methods suffer from the curse of dimensionality, which causes the performance of the method to drastically deteriorate and in many cases to completely fail (Adali and Haykin 2010; Bengtsson et al. 2003). Finally, there are alternatives to MC methods for Bayesian estimation, such as variational inference (VI) (Jordan and Kleinberg 2006), which has become popular in the machine learning community because of its scalability to models with large numbers of unknowns. This family of techniques takes an optimization approach to approximating the posterior distribution. In particular, whereas MC techniques provide a numerical approximation to the exact posterior using a set of samples, VI provides a locally-optimal, analytical solution as an approximation of the posterior. However, deriving the set of equations used to iteratively update the parameters often requires a large amount of computations. This is the case even for many models that are conceptually quite simple. Furthermore, unlike many MC schemes, the approximating distribution in VI is not guaranteed to converge to the posterior and typically underestimates the variance of the true posterior distribution.

4.2.2.1

Algorithmic Structures

The first step in the Bayesian inference framework when dealing with complex systems is to develop algorithmic structures that can handle the high dimensionality of such systems. The task of jointly estimating the hidden states .x0:T and the model parameters .θ has undergone intense study in the past several decades. A variety of approaches have been proposed in the literature for both online estimation (sequential processing) and offline estimation (batch processing). For online estimation, PF has become the method of choice when dealing with nonlinear and non-Gaussian systems. At time instant .t, PF methods generate samples (i.e., particles) from a proposal distribution and assign them weights that measure their adequacy to represent the unknowns. They then build an approximation to the joint posterior .p(x0:t , θ|y1:t ) via a random measure composed

4 Recent Advances in Bayesian Inference for Complex Systems (n)

91 (n)

(n)

of the samples and the weights, i.e., .Xt = {(x0:t , θ(n) ), wt }N n=1 , where .x0:t denotes the .nth particle stream at time instant .t, .θ(n) denotes the .nth sampled model (n) parameter vector, and .wt is the corresponding weight. This process continues as more data become available, i.e., as one progresses to the next time instant. Note that this traditional PF framework does not benefit from possible iterations over a fixed time instant to obtain improved approximation of the target distribution at that time. Moreover, even state-of-the-art PF implementations struggle in handling both hidden states and unknown model parameters (and even more so for highdimensional scenarios) since the PF framework is traditionally applied to dynamic state estimation. For high-dimensional scenarios, multiple PF (MPF) has been proposed (Djuri´c et al. 2007). MPF methods utilize the divide-and-conquer approach to address the curse of dimensionality and the vector of unknowns is partitioned into subvectors, each dealt with by a separate filter. Weighting is carried out via the exchange of information between filters. However, for most models, the weighting is improper and therefore, no theoretical results regarding convergence of MPF exist in the literature. Additional open issues include how to systematically partition the vector of unknowns, how to theoretically determine the tradeoff between amount of information exchanged among individual filters and desired accuracy, or how to determine the number of samples needed by each filter. Particle MCMC (PMCMC) (Andrieu et al. 2010) is considered to be the stateof-the-art approach for building an offline approximation to the joint distribution .p(x0:T , θ|y1:T ). Unlike standard PF methods, PMCMC uses a Markov chain to approximate the joint posterior by processing the entire set of observations .y1:T together, making the methodology unsuitable for online settings. At the .ith iteration, a sample of the model parameter vector .θ() is proposed. PF is then utilized to () obtain a sample of the hidden states .x0:T (conditioned on the sampled parameter vector .θ() ) and an unbiased estimate of the marginal likelihood .pˆ (y1:T |θ() ). The marginal likelihood estimate is used to compute the probability of accepting the () proposed sample .(x0:T , θ() ) as the next state in the approximating Markov chain. Unfortunately, methods like PMCMC, which rely on additional MCMC sampling (Andrieu et al. 2010; Berzuini et al. 1997; Gilks and Berzuini 2001), suffer from possibly poor mixing of the used chain. There is also difficulty in developing parallel implementations of PMCMC since samples are proposed one at a time.

4.2.2.2

Learning Strategies

Several important issues need to be addressed for application of the algorithmic structures to complex systems. Among them, learning strategies for initialization, proposal learning, model uncertainty, and distributed processing of the algorithms are necessary for practical implementation. For instance, initialization is a crucial step for the efficiency of MC methods. If initialization happens far away from areas of high probability density, the algorithm will need many iterations traversing the space in search of areas of high probability density, reducing the efficiency of the

92

M. F. Bugallo

algorithm. Moreover, the performance of adaptive MC methods highly depends on proposal learning strategy inherit to the algorithm. Learning Strategies for Algorithm Initialization Recently, there has been a push in combining MC methods and VI to exploit their strengths and resolve some issues regarding parameter initialization (Naesseth et al. 2017). The overarching idea of these hybrid methodologies is that VI can be used to find the relevant regions of the sampling space, while MC schemes can be used to obtain the final posterior approximation. In VI, an approximation to the posterior distribution is formed by maximizing    p(y1:T |θ)p(θ) dθ  log p(y1:T |θ)p(θ)dθ , .L(λ)  q(θ; λ) log q(θ; λ) Rdθ Rdθ (4.4) where .q(θ; λ) denotes the approximation to the posterior, and the lower-bound .L(λ) is commonly referred to as the evidence lower bound (ELBO) (Murphy 2012).3 In modern implementations, the maximization of .L(λ) is carried out via stochastic optimization, which has enabled the scalability of VI to complex probabilistic models (Boyd and Vandenberghe 2004). Unfortunately, to preserve scalability, most VI schemes construct Gaussian approximations to the posterior, rather than approximating with a more complex distribution, such as a mixture distribution. This assumption can be dangerous and lead to poor performance in Bayesian models with highly non-Gaussian posteriors (e.g., heavy-tailed or multimodal). Note that mixture distributions are not typically used in the VI framework due to the difficulty of applying a reparameterization of the ELBO. Reparametrization of the ELBO is necessary for efficient computation of stochastic gradients with respect to the parameters of the approximating distribution .λ, which is crucial for implementation of VI methods within the Bayesian framework for complex models (Graves 2016). 



Learning Strategies for Constructing Proposals For any adaptive MC method, the choice of the proposal distribution is critical to guarantee performance. This is particularly sensitive for complex systems, since with large numbers of unknowns, the resulting target posterior distributions are extremely peaky and thus incredibly difficult to approximate. Recent advances have been produced in the adaptive MC literature for constructing suitable proposal distributions. We provide some context regarding AIS methods. Generally, an AIS algorithm adapts the proposal distribution over iterations by solving some optimality criterion to improve the performance of the sampler. For instance, in the mixture population MC algorithm, a mixture proposal is adapted to minimize the Kullback-Leibler divergence (KLD) between the target and the proposal (Cappé et al. 2008b). In some implementations, the proposal is adapted to directly minimize the variance of IS estimators (El-Laham et al.

3 Note that here we discuss initialization strategies for Bayesian models with only unknown static parameters for simplicity in the notation and presentation.

4 Recent Advances in Bayesian Inference for Complex Systems

93

Fig. 4.4 Model selection

2019; Ryu 2016; Ryu and Boyd 2014). Unfortunately, even though AIS algorithms obtain better proposals with increasing iterations, the overall performance can be compromised by samples drawn from poorly fit proposals in the early iterations of the algorithm. Some solutions have been proposed to tackle this problem based on re-weighting previously drawn samples (Cornuet et al. 2012; Portier and Delyon 2018). More recently, there have been several pushes to improve the performance of adaptive Monte Carlo methods using ideas from optimal control (Bojesen 2018; Heng et al. 2017). Some of these methods need deeper analyses and require high computation for even simple models. Learning Strategies for Model Selection Model selection is a classical signal processing problem that amounts to choosing the model (among several candidates) that better describes the system of interest using the available observations (see Fig. 4.4). While model selection has been studied in great detail in the literature (Claeskens and Hjort 2008; Ding et al. 2018), complex systems pose new challenges which require novel solutions beyond stateof-the-art techniques. Joint model selection and parameter estimation have been addressed using AIS (Hong et al. 2010; Kilbinger et al. 2010). In these methods, an AIS scheme is run for each of the considered models. Each AIS scheme provides an MC approximation to the corresponding model likelihood, which is then utilized to solve the maximum a posteriori optimization problem to obtain the most suitable model. Regrettably, this approach demands that the number of models is defined a priori and can be infeasible when the number of considered models is large. These computational limitations also plague PF methods, as most model selection schemes in the PF literature require that a separate filter is run for each model (Adali and Haykin 2010; Djuri´c et al. 2007). A more favored and general approach to joint parameter estimation and model selection considers the number of examined models to be unknown. A method from the MCMC literature that is based on this idea is the reversible-jump MCMC (RJMCMC) (Green 1995; Marin and Robert 2010), which allows for simulation of the posterior distribution on spaces of varying dimensions, meaning that the number of unknowns is learned through the algorithm. However,

94

M. F. Bugallo

this method suffers from poor mixing of the Markov chain due to the low-acceptance probability of the proposed samples in high-dimensional systems and from lowacceptance probabilities for jumping to alternative models due to the lack of a notion of closeness in model space. This raises the question of how to incorporate methods like RJMCMC in the next generation of Bayesian methods for complex systems.

4.2.2.3

Smart Distributed Processing

Although MC methods are considered to be the state-of-the-art methods for approximating Bayesian posteriors, there are a multitude of issues that arise when dealing with applications with large sets of observations. Traditional implementations of MC schemes cannot easily scale to large datasets, mainly because the evaluation of the likelihood becomes too computationally expensive. Therefore, novel distributed implementations must be developed in order to accommodate for the additional challenges that arise when dealing with complex systems. Over the past several years, many advances have been made in an attempt to develop distributed implementations of Bayesian methods. One popular approach is the expectation propagation (EP) algorithm (Minka 2001), which approximates the posterior distribution by assuming that it comes from a parameterized family of probability distributions (e.g., exponential family) and then optimizes over the parameters of that family using a message passing scheme. Recent advances based on stochastic optimization have allowed for the development of efficient EP schemes that provide excellent predictive performance in practice (Bui et al. 2016; Chen et al. 2009). Unfortunately, similar to VI, the EP approximation of the full data posterior is biased, and provides a poor estimate of the posterior’s variance. Another class of popular methods again rely on partitioning the full dataset into subsets and then obtaining a local posterior approximation over each subset in the form of a discrete random measure (unweighted or weighted samples). The local approximations are then fused to approximate the full posterior (Minsker et al. 2014; Neiswanger and Xing 2013; Wang and Dunson 2013). For example, in Neiswanger and Xing (2013), the outputs of MCMC samplers are first summarized using kernel density estimates (KDEs), and the KDEs are then fused to form an approximation to the full data posterior. Another approach called WASP (Srivastava et al. 2015) directly fuses the samples from a set of MCMC samplers by formulating the fusion problem as an optimal transport problem. In this approach, the local posteriors are combined by determining their Wasserstein barycenter (Rabin et al. 2011). While WASP outperforms competing algorithms in many scenarios, the method’s main convergence result requires that the data are conditionally independent within each sampler, complicating its applicability to complex systems, which include both hidden states and unknown model parameters.

4 Recent Advances in Bayesian Inference for Complex Systems

95

4.2.3 Case Studies Two case studies are here presented and serve as examples of the applicability and power of the methodologies described in real scenarios.

4.2.3.1

Case Study 1: Evolution of Penguin Population Dynamics

An important problem in ecology is related to modeling the evolution of the population of a species. For instance, investigating penguin populations in the Antarctic is critical for improving ecological forecasting (predicting penguin abundance) and for building effective Antarctic conservation policy and management for protected area design and tourism management. Traditionally, the systems describing such problems are learned through the statistical analysis of data collected in a markrecapture studies, where banded animals are studied over time (Williams et al. 2002). While this approach is reliable and widely accepted, there is growing interest in finding alternative methods, as banding of animals has raised important ethical concerns (Putman 1995). The use of advanced computational models and methods based on simple point counts of breeding animals to represent the ecological system can enable non-invasive learning of the unknown demographic rates and related parameters. However, these approaches involve resolving a number of theoretical and practical challenges including the uncertainty of the dynamical model describing the problem (model uncertainty); the large number of parameters, size of species and system rates (static and dynamic) to be estimated (highdimensionality); the processing of all time-series observations with likely missing data points (distributed processing); and the uncertainty and unknown nature of many of the parameters and factors involved in the underlying model (e.g., the age distribution within the population or the noise distribution of the observations). This example focuses on the representation of Adélie penguin dynamics using simple-point counts of breeding animals (possibly at several Antarctic sites) and learning the Bayesian posterior distributions of the unknown demographic rates as well as possible prediction of the populations (see Fig. 4.5). Dynamics of penguin populations can be described using a simple life-cycle diagram as in Fig. 4.6 for .J = 4 age classes. At the .th Antarctic site at time instant .t, observations (i.e., .y() t ) of the total number of adult breeding penguins () () and chicks are collected each year, denoted .S˜ b,t and .C˜ t , respectively. A possible ()

mathematical model is shown below to the right, with the hidden states being .xt = () () () () () () ()  () [S() 1,t , . . . , SJ,t , Sb1,t , . . . , Sb(J−2),t , C1,t , . . . , CJ−2,t , Sb,t , Ct ] and .Sj,t denoting ()

()

the number of age .j adults, .Sbj,t the number of age .j + 2 breeders, .Cj,t the number of chicks from age .j + 2 breeders, and .J the total number of age classes.

96

M. F. Bugallo

Fig. 4.5 Penguin map Fig. 4.6 Life cycle

 () S1,t

∼ Binomial



()

Ct−1 2

, ψjuv ,

  () () Sj,t ∼ Binomial Sj−1,t−1 , ψadu , j = 2, . . . , J − 1   () () () SJ,t ∼ Binomial SJ−1,t−1 + SJ,t−1 , ψadu ,   () () Sbj,t ∼ Binomial SJ+2,t , pbj , j = 1, . . . , J − 2 ⎛ J−2 () ⎞ j=1 Sbj,t ⎠ , j = 1, . . . , J − 2 prj = pr0j ⎝1 − k   () () Cj,t ∼ Binomial Sbj,t , prj ,

j = 1, . . . , J − 2

  2  ˜S() ∼ N S() , σs S() , b,t b,t b,t

Sb,t =

()

J−2  j=1

()

Sbj,t ,

4 Recent Advances in Bayesian Inference for Complex Systems

97

Fig. 4.7 Combined AIS/PF-processing for the penguin population dynamics

    () () () 2 ˜ Ct ∼ N Ct , σc Ct ,

() Ct

=

J−2 

()

Cj,t .

j=1

For all the .L Antarctic sites, the model parameters are .θ = [ψjuv , ψadu , pb1 , . . . , pb(J−2) , pr01 , . . . , .pr0(J−2) , κ1 , . . . , κL , σc , σs ] , with .ψjuv being the juvenile survivorship, .ψadu the adult survivorship, .pbj the breeding propensity of the .(j+2)th age class, and .pr0j and .prj the natural and effective reproductive rates of the .(j + 2)th age class, respectively. The parameter .κ represents the carrying capacity of the site, and .σs and .σc the noise parameters. The goal is in estimating the joint posterior distribution of the demographic rate parameters .θ and the latent populations .x0:T = (1) (L) {x0:T , . . . , x0:T } given all observations collected across all Antarctic sites .y1:T = (1) (L) {y1:T , . . . , y1:T }. Obtaining the posterior distribution will provide insight about the overall ecological system. Computed parameter estimates and penguin population projections could see future use in decision support application, and could shape formulation of Antarctic conservation policy. We applied a simple version of a combined AIS/PF framework to inference of the demographic rates of penguins for a single Antarctic site (i.e., a single time-series) (El-Laham et al. 2021). Results are displayed in Fig. 4.7, which shows the posterior samples of two model parameters (adult and juvenile penguin survivorship) along with the evolution of effective sample size (ESS), a typical diagnostic utilized for sampling methods (Adali and Haykin 2010; Morita et al. 2008). The framework utilized a simple proposal to sample model parameters and the bootstrap filter (Adali and Haykin 2010) to sample hidden states and to obtain marginal likelihood estimates. The performance compares open-source Bayesian inference software like JAGS (Plummer 2003) and shows that the AIS/PF approach obtains the approximated posterior at a much faster rate.

.

98

M. F. Bugallo

Fig. 4.8 Gene regulatory network

Coefficient Matrix Gene 1

Gene 2

0 -0.1 -4.4 3.0 0 0 0 0.8 0 Adjacency Matrix

Gene 3

4.2.3.2

0 1 0

1 0 1

1 0 0

Case Study 2: Reconstruction of Gene Regulatory Networks

A problem related to complex systems in life sciences addresses the reconstruction of gene regulatory networks from various data sources (see Fig. 4.8 left). Learning genetic interaction networks, which reflect dependencies among genes is extremely relevant to gain knowledge about the complex metabolic processes and the functional organization of cells (Barabasi and Oltvai 2004). In recent years, there has been a large push towards developing mathematical models to learn gene interactions (Bugallo et al. 2015; Chang et al. 2008; Ortega et al. 2018; Ta¸sdemir et al. 2017). A promising direction for researching gene network topology is based on the analysis of next generation sequencing (RNASeq) and/or microarray data containing the expression of genes at a large-scale (thousands of genes at the same time). The observed gene expressions are noisy and subject to various systematic errors, and therefore, it is difficult to straightforwardly use them to learn about the gene-to-gene interactions in the network. There are many challenges associated with learning the topology of gene regulatory network including the large number of genes that characterize the dynamical network (high-dimensionality); the uncertainty about the underlying dynamical model describing the gene expressions (model uncertainty); the required joint processing of gene expression data that needs to be handled by the inference machine (distributed processing); and the nature of parameters representing the system including the unknown noise distributions of the gene expressions or the sparsity of the gene interactions. This case study shows how to infer the interaction among genes in a regulatory network given a time series of gene expressions, and learn the topology of the network. A gene regulatory network can be graphically represented with nodes (genes) and edges (interactions) as in Fig. 4.8. Mathematically, the system is characterized by a coefficient matrix .C, containing interaction information among the genes, and the adjacency matrix .A, which indicates the structure of the coefficient matrix, i.e., whether two particular genes in the network are connected. At each time instant, observations (i.e., .yt ) of the noisy gene expressions are collected in a RNA-Seq and/or microarray database. The system can be formulated as:

4 Recent Advances in Bayesian Inference for Complex Systems

99

Fig. 4.9 Model selection for gene regulatory networks

xt = Cg(xt−1 ) + ut , .

yt = xt + vt

(4.5)

where the time propagation of the gene expressions is modeled using the nonlinear transformation .g(·), scaled by the coefficient matrix .C, and the uncertainty of the model is reflected by the process and observation noise vectors, .ut and .vt , respectively. For a given model, the unknowns are the hidden states .x0:T , and the model parameters .θ = [C, A]. Moreover, the function .g(·) that describes the model is also unknown. Preliminary results for joint model selection and parameter estimation are shown for gene regulatory network reconstruction based on synthetic data (Iloska et al. 2020). A particle Gibbs sampler was derived to approximate the joint posterior over the unknown gene expressions, the unknowns models, and the unknown parameters of those models. Here, each model corresponded to a different nonlinear function that describes the evolution of the gene expression. Figure 4.9 shows a histogram of the sampled models, i.e., the approximated posterior distribution .pˆ (M|y1:T ). Based on this approximation, the selected model would be .M1 , which was the true model used to generate the data.

References Adali T, Haykin S (2010) Adaptive signal processing: next generation solutions, vol 55. Wiley Andrieu C, Davy M, Doucet A (2001) Improved auxiliary particle filtering: Applications to timevarying spectral analysis. In: Proceedings of the 11th IEEE signal processing workshop on statistical signal processing (Cat. No. 01TH8563). IEEE, pp 309–312 Andrieu C, Doucet A, Holenstein R (2010) Particle Markov chain Monte Carlo methods. J Roy Stat Soc B 72(3):269–342 Barabasi AL, Oltvai ZN (2004) Network biology: Understanding the cell’s functional organization. Nat Rev Genet 5:101–113

100

M. F. Bugallo

Bengtsson T, Snyder C, Nychka D (2003) Toward a nonlinear ensemble filter for high dimensional systems. J Geophys Res 108 Bernardo JM, Smith AF (2001) Bayesian theory. Wiley Berzuini C, Best NG, Gilks WR, Larizza C (1997) Dynamic conditional independence models and Markov Chain Monte Carlo methods. J Am Stat Assoc 92:1403–1412 Binford TO, Levitt TS, Mann WB (2013) Bayesian inference in model-based machine vision. Preprint.‘arXiv:13042720 Bojesen TA (2018) Policy-guided Monte Carlo: Reinforcement-learning Markov chain dynamics. Phys Rev E 98(6):063303 Boyd S, Vandenberghe L (2004) Convex optimization. Cambridge University Press Brooks S, Gelman A, Jones GL, Meng XL (eds) (2011) Handbook of Markov Chain Monte Carlo. CRC Press Bugallo MF, Elvira V, Martino L, Luengo D, Míguez J, Djuri´c PM (2017) Adaptive importance sampling: The past, the present, and the future. Signal Process Mag 34:60–79 Bugallo MF, Ta¸sdemir Ç, Djuri´c PM (2015) Estimation of gene expression by a bank of particle filters. In: 2015 23rd European signal processing conference (EUSIPCO). IEEE, pp 494–498 Bui T, Hernández-Lobato D, Hernandez-Lobato J, Li Y, Turner R (2016) Deep gaussian processes for regression using approximate expectation propagation. In: International conference on machine learning, pp 1472–1481 Cappé O, Douc R, Guillin A, Marin J, Robert CP (2008a) Adaptive importance sampling in general mixture classes. Stat Comput 18(4):447–459 Cappé O, Douc R, Guillin A, Marin JM, Robert CP (2008b) Adaptive importance sampling in general mixture classes. Stat Comput 18:447–459 Cappé O, Guillin A, Marin J, Robert CP (2004) Population Monte Carlo. J Comput Graph Stat 13(4):907–929 Casella G, Berger RL (2002) Statistical inference. Duxbury, Pacific Grove, CA (USA) Chang C, Ding Z, Hung YS, Fung PCW (2008) Fast network component analysis (fastnca) for gene regulatory network reconstruction from microarray data. Bioinformatics 24(11):1349–1358 Chen H, Tiˇno P, Yao X (2009) Predictive ensemble pruning by expectation propagation. IEEE Trans Knowl Data Eng (7):999–1013 Claeskens G, Hjort NL (2008) Model selection and model averaging. Cambridge University Press Cornuet J, Marin J, Mira A, Robert CP (2012) Adaptive multiple importance sampling. Scand J Stat 39(4):798–812 Ding J, Tarokh V, Yang Y (2018) Model selection techniques: An overview. IEEE Signal Process Mag 35(6):16–34 Djuri´c PM, Lu T, Bugallo MF (2007) Multiple particle filtering. In: 2007 IEEE international conference on acoustics, speech and signal processing-ICASSP’07, vol 3. IEEE, pp III–1181 Doucet A, Johansen AM (2009) A tutorial on particle filtering and smoothing: Fifteen years later. Handbook of nonlinear filtering, vol 12(656-704), p 3 El-Laham Y, Djuri´c PM, Bugallo MF (2019) A variational adaptive population importance sampler. In: ICASSP 2019-2019 IEEE international conference on acoustics, speech and signal processing (ICASSP). IEEE, pp 5052–5056 El-Laham Y, Elvira V, Bugallo MF (2018) Robust covariance adaptation in adaptive importance sampling. IEEE Signal Process Lett 25 El-Laham Y, Yang L, Lynch HJ, Djuri´c PM, Bugallo MF (2021) Particle Gibbs sampling for regime-switching state-space models. In: IEEE international conference on acoustics, speech and signal processing (ICASSP) Evans M, Swartz T (1995) Methods for approximating integrals in statistics with special emphasis on Bayesian integration problems. Stat Sci, 254–272 Fearnhead P (2008) Computational methods for complex stochastic systems: A review of some alternatives to MCMC. Stat Comput 18(2):151–171 Gilks WR, Berzuini C (2001) Following a moving target - Monte Carlo inference for dynamic Bayesian model. J Roy Stat Soc B 63:127–146 Godsill SJ, Rayner PJW (1998) Digital audio restoration, 1st edn. Springer, London (UK)

4 Recent Advances in Bayesian Inference for Complex Systems

101

Gordon NJ, Salmond DJ, Smith AFM (1993) Novel approach to nonlinear/non-Gaussian Bayesian state estimation. In: IEE proceedings F (radar and signal processing), vol 140, pp 107–113 Graves A (2016) Stochastic backpropagation through mixture density distributions. Preprint. arXiv:160705690 Green PJ (1995) Reversible jump Markov chain Monte Carlo computation and Bayesian model determination. Biometrika 4:711–732 Haykin SS (1994) Blind deconvolution. Prentice Hall, Englewood Cliffs, NJ (USA) Heng J, Bishop AN, Deligiannidis G, Doucet A (2017) Controlled sequential Monte Carlo. Preprint. arXiv:170808396 Hinne M, Heskes T, Beckmann CF, van Gerven MAJ (2013) Bayesian inference of structural brain networks. NeuroImage 66:543–552 Hong M, Bugallo MF, Djuri´c PM (2010) Joint model selection and parameter estimation by population Monte Carlo simulation. IEEE J Sel Top Signal Process 4:526–539 Hutt R (2016) What are the 10 biggest global challenges? In: World economic forum annual meeting. https://www.weforum.org/agenda/2016/01/what-are-the-10-biggest-global-challenges/ Iloska M, El-Laham Y, Bugallo MF (2020) A particle Gibbs sampling approach to topology inference in gene regulatory networks. In: IEEE international conference on acoustics, speech and signal processing (ICASSP) Jenkins CR, Peacock JA (2011) The power of Bayesian evidence in astronomy. Mon Not Roy Astron Soc 413:2895–2905 Jordan M, Kleinberg J (2006) Information science and statistics, vol 4 Karlsson R (2005) Particle filtering for positioning and tracking applications. PhD thesis, Linköping University Electronic Press Kay SM (1993) Fundamentals of statistical signal processing: estimation theory. Prentice Hall, Upper Saddle River, NJ (USA) Kilbinger M, Wraith D, Robert CP, Benabed K, Cappé O, Cardoso JF, Fort G, Prunet S, Bouchet FR (2010) Bayesian model comparison in cosmology with population Monte Carlo. Mon Not Roy Astron Soc 405:2381–2390 Li T, Bolic M, Djuric PM (2015) Resampling methods for particle filtering: classification, implementation, and strategies. IEEE Signal Process Mag 32(3):70–86 Liu J, West M (2001) Combined parameter and state estimation in simulation-based filtering. In: Sequential Monte Carlo methods in practice. Springer, pp 197–223 Lopes HF, Tsay RS (2011) Particle filters and Bayesian inference in financial econometrics. J Forecast 30:168–209 Marin JM, Robert CP (2010) A practical approach to computational Bayesian statistics. Springer Martino L, Elvira V, Luengo D, Corander J (2015) An adaptive population importance sampler: Learning from uncertainty. IEEE Trans Signal Process 63(16):4422–4437 Martino L, Elvira V, Luengo D, Corander J (2017) Layered adaptive importance sampling. Stat Comput 27(3):599–623 Minka TP (2001) Expectation propagation for approximate Bayesian inference. In: Proceedings of the seventeenth conference on uncertainty in artificial intelligence. Morgan Kaufmann Publishers Inc., pp 362–369 Minsker S, Srivastava S, Lin L, Dunson D (2014) Scalable and robust Bayesian inference via the median posterior. In: International conference on machine learning, pp 1656–1664 Morita S, Thall PF, Müller P (2008) Determining the effective sample size of a parametric prior. Biometrics 64(2):595–602 Murphy KP (2012) Machine learning: A probabilistic perspective. The MIT Press Naesseth CA, Linderman SW, Ranganath R, Blei DM (2017) Variational sequential Monte Carlo. Preprint. arXiv:170511140 Neiswanger C Wand Wang, Xing E (2013) Asymptotically exact, embarrassingly parallel MCMC. Preprint. arXiv:13114780 Orchard ME, Vachtsevanos GJ (2007) A particle filtering-based framework for real-time fault diagnosis and failure prognosis in a turbine engine. In: 2007 mediterranean conference on control & automation. IEEE, pp 1–6

102

M. F. Bugallo

Ortega A, Frossard P, Kovaˇcevi´c J, Moura JMF, Vandergheynst P (2018) Graph signal processing: Overview, challenges, and applications. Proc IEEE 106(5):808–828 Plummer M (2003) JAGS: A program for analysis of Bayesian graphical models using Gibbs sampling. In: Proceedings of the 3rd international workshop on distributed statistical computing, Vienna, Austria, vol 124, p 10 Portier F, Delyon B (2018) Asymptotic optimality of adaptive importance sampling. In: Advances in neural information processing systems, pp 3134–3144 Putman RJ (1995) Ethical considerations and animal welfare in ecological field studies. Biodivers Conserv 4(8):903–915 Rabin J, Peyré G, Delon J, Bernot M (2011) Wasserstein barycenter and its application to texture mixing. In: International conference on scale space and variational methods in computer vision. Springer, pp 435–446 Ryu EK (2016) Convex optimization for monte carlo: Stochastic optimization for importance sampling. PhD thesis, Stanford University Ryu EK, Boyd SP (2014) Adaptive importance sampling via stochastic convex programming. Preprint. arXiv:14124845 Scharf LL (1991) Statistical signal processing. Addison-Wesley, Reading, MA (USA) Srivastava S, Cevher V, Dinh Q, Dunson D (2015) WASP: Scalable Bayes via barycenters of subset posteriors. In: Artificial intelligence and statistics, pp 912–920 Ta¸sdemir Ç, Bugallo MF, Djuri´c PM (2017) A particle-based approach for topology estimation of gene networks. In: 2017 IEEE 7th International workshop on computational advances in multi-sensor adaptive processing (CAMSAP). IEEE, pp 1–5 Van Trees HL (2004) Detection, estimation, and modulation theory, part I: detection, estimation, and linear modulation theory. Wiley von Toussaint U (2011) Bayesian inference in physics. Rev Mod Phys 83:943–999 Wang X, Dunson DB (2013) Parallel MCMC via Weierstrass sampler, vol 24. Preprint. arXiv:13124605 Wang X, Li T, Sun S, Corchado J (2017) A survey of recent advances in particle filters and remaining challenges for multitarget tracking. Sensors 17(12):2707 Williams BK, Nichols JD, Conroy MJ (2002) Analysis and management of animal populations. Academic Press

Mónica F. Bugallo received her Ph.D. in computer science and engineering from University of A Coruña, Spain. She is a Professor of Electrical and Computer Engineering and currently Vice Provost for Faculty Affairs and Diversity, Equity, and Inclusion at Stony Brook University, NY, USA. Her research is focusing on statistical signal processing, with emphasis on the theory of Monte Carlo methods and its application to different disciplines including biomedicine, ecology, sensor networks, and finance. She has also focused on STEM education and has initiated successful programs to engage students at all academic stages in the excitement of engineering and research, with focus on underrepresented groups. Bugallo has authored and coauthored two book chapters and more than 230 journal papers and refereed conference articles. She is a senior member of

4 Recent Advances in Bayesian Inference for Complex Systems

103

the IEEE, served as past chair of the IEEE Signal Processing Society Signal Processing Theory and Methods Technical Committed and as past chair of the EURASIP Special Area Team on Theoretical and Methodological Trends in Signal Processing. She is also serving as Senior Associate Editor of the IEEE Signal Processing Letters and Associate Editor of the IEEE Transactions on Signal Processing and has been part of the technical committee and has organized various professional conferences and workshops. She has received several prestigious research and education awards including the State University of New York (SUNY) Chancellor’s Award for Excellence in Teaching (2017), the 2019 Ada Byron Award of the Galician Society of Computer Engineers (Spain) for a successful professional career path that inspires women to engineering study and careers, the Best Paper Award in the IEEE Signal Processing Magazine 2007 as coauthor of a paper entitled Particle Filtering, the IEEE Outstanding Young Engineer Award (2009), for development and application of computational methods for sequential signal processing, the IEEE Athanasios Papoulis Award (2011), for innovative educational outreach that has inspired high school students and college level women to study engineering, the Higher Education Resource Services (HERS) Clare Boothe Luce (CBL) Scholarship Award (2017), and the Chair of Excellence by the Universidad Carlos III de Madrid-Banco de Santander (Spain) (2012).

Chapter 5

Hardware-Limited Task-Based Quantization in Systems Derya Malak, Rabia Yazicigil, Muriel Médard, Xing Zhang, and Yonina C. Eldar

5.1 Motivation and Background The problem of representing data has been the subject of extensive study in information theory and signal processing. Starting from the seminal work of Claude Shannon (1948) and its many extensions, information theory is concerned with how to encode and decode information sources with dependent data. In compression, the goal of this encoding is to provide parsimony, i.e., to encode information sources such that they take the minimal amount of space, or equivalently, such that the representation is shortest in terms of the number of bits to be communicated over a network or exchanged in a system. This traditional objective of representing the data itself is not tailored for modeling a task derived from the data. In this chapter, we review approaches that concern the design and implementation of distributed task-based compression and quantization for systems. In this chapter, we present the theoretical and practical design efforts for developing a principled framework for the efficient representation of a large amount of data for the purposes of realizing tasks, which is a challenge in various systems

D. Malak () Communication Systems Department, EURECOM, Campus SophiaTech, Biot, France e-mail: [email protected] R. Yazicigil ECE Department, Boston University, Boston, MA, USA e-mail: [email protected] M. Médard EECS Department, MIT, Cambridge, MA, USA e-mail: [email protected] X. Zhang · Y. C. Eldar Faculty of Math & CS, Weizmann Institute of Science, Rehovot, Israel e-mail: [email protected]; [email protected] © The Author(s), under exclusive license to Springer Nature Switzerland AG 2023 M. S. Greco et al. (eds.), Women in Telecommunications, Women in Engineering and Science, https://doi.org/10.1007/978-3-031-21975-7_5

105

106

D. Malak et al.

and network applications. We review various information-theoretic and signal processing approaches to distributed function-aware quantization schemes for taskbased compression. These schemes capture the correlation between the sources and the task structure as a means of dimensionality reduction through joint compression of data as well as quantization for approximating functions. We also focus on techniques for mitigating practical constraints in quantization, such as hardware and power limitations, for devising hardware-limited task-based quantization models. We commence the chapter by outlining distributed coding techniques for compression and the fundamentals of distributed quantization over a communication network, which we detail in Sect. 5.2. More specifically, we consider the distributed quantization of functions. This objective is motivated by finding a parsimonious representation for describing a function, which is a high-level abstraction of a task across a network. To tackle this problem, motivated by the theorem of Slepian and Wolf (1973) that provides the information-theoretic lower bounds on the lossless coding rate for distributed sources to recover them at a destination with an arbitrarily small error probability, distributed source compression, and distributed functional compression, techniques have been studied under various forms. The primary focus of these approaches is on the post-quantization phase that concerns how to perform distributed function computation and find the minimum entropy colorings of source graphs. In complementary to these techniques, we describe an approach that provides vector quantized functional representations in systems. More specifically, this approach, incorporating a linear hyperplane-based functional compression, generalizes the orthogonal binning of source sequences in Slepian– Wolf using hyper bins. The idea relies on exploiting linear discriminant analysis (LDA) to distinguish the source data that yield the same outcome. By carefully adjusting the hyperplane parameters, it is possible to design achievable encodings for quantizing functions. In Sect. 5.3, we consider hardware-limited task-based quantization in systems, including the general system model, the specific system design for linear and quadratic tasks, followed by numerical examples. In practice, signals are usually acquired in order to extract some underlying information, i.e., for a specific task. By exploiting this fact, at the receiver, analog and digital signal processing modules can be jointly designed for the recovery of the task rather than the original received signal using a small number of bits in a hardware-efficient way. Specifically, task-based quantization first introduces an analog combiner that reduces the dimensionality of the input to the quantizers, and then scalar quantizers are employed considering practical hardware limitations, followed by a digital domain processing module. By properly designing the overall quantization system based on the task, satisfying performance can be achieved with fewer quantization bits compared to the conventional scheme that is power-consuming with high-resolution quantizers. In Sect. 5.4, we present a power-efficient hybrid multiple-input multiple-output (MIMO) receiver design utilizing the task-specific quantization via the use of low-resolution ADCs (Zirtiloglu et al. 2022). An optimal task recovery accuracy with suppression of undesired interferers can be obtained at significantly lower power consumption compared to task-agnostic MIMO receivers by developing a

5 Hardware-Limited Task-Based Quantization in Systems

107

joint optimization framework for analog pre-processing hardware with embedded beamforming and task-specific digital signal processing (Zirtiloglu et al. 2022). We consider optimizing task-based quantization in more general settings in Sect. 5.5 via incorporating network sources. More specifically, this requires capturing the critical aspects of data and function, e.g., Shannon entropy and surjectivity, network, such as congestion and latency, for task distribution via combining the tools of queueing theory, optimization, and entropy coding. By leveraging queueing theory’s Little’s law, a relation between how to distribute tasks and their completion time can be established. Melding Little’s law with the techniques from Section 1.2 for distributed quantization, and by incorporating a routing policy, the reduction of flow caused by compression at each node can be captured, which paves the way for optimizing the assignment and completion time of the tasks. The rest of the chapter is organized as follows. In Sect. 5.2, we describe an approach called hyper binning to jointly quantize and compress distributed data for obtaining a vector quantized functional representation. In Sect. 5.3, we detail the concept of task-based quantization considering practical hardware constraints in order to leverage that signals in practice are acquired to extract the underlying information for a specific task. In Sect. 5.4, we focus on the recovery of signals under bit constraints by using power-efficient hybrid MIMO communication systems. In Sect. 5.5, we consider task-based quantization in networks, to capture the tradeoff space between communications and computation of various task models. Finally, Sect. 5.6 summarizes our main contributions and points out future directions.

5.2 Theory and Principles In networked environments, one does not compute a function of distributed sources and then quantize it directly. Instead, the sources can jointly perform compression to quantize the function of interest. This scenario is motivated by the fundamental limits for the lossless coding rate for distributed coding of two statistically dependent sources where the sources are not allowed to communicate with each other. For this setting, the Slepian–Wolf theorem gives a theoretical bound on compression rate for i.i.d. finite alphabet source sequences (Slepian and Wolf 1973). The extensions of Slepian and Wolf (1973) to functional compression in Feizi and Médard (2014) and Basu et al. (2020) are also on the already post-quantized data streams. However, the separation-based approach, which first quantizes and then compresses the data, may be suboptimal. A strategy that employs compression on the functional representation of the vector quantized data can outperform separation. The focus of this section is on the problem of distributed vector quantization of sources for the purposes of describing a function, which was proposed in Malak and Médard (2020, 2023). This technique generalizes traditional vector quantization for data compression and brings together techniques from information theory, such as distributed source and functional compression, to the area of signal processing via function quantization inspired from hyperplane-based vector quantizers. The

108

D. Malak et al.

objective of distributed function-aware quantization is to minimize the entropy of joint source partitioning. To do so, the proposed model in Malak and Médard (2020, 2023) extends the Slepian–Wolf binning model to a linear hyperplane-based scheme for encoding continuous functions. Rather than determining a quantized representation of the dispersed source data, it finds a partitioning of the sources decided by the hyperplane arrangement to allow describing the function up to some quantization distortion in a high-dimensional codebook space. Such a scheme needs fewer dimensions than the codeword size and captures the dependence of the function on the distributed data. We next describe the fundamentals of distributed function-aware quantization. Section 5.2.1 provides a background on coding for distributed compression of sources and functions. Section 5.2.2 motivates a linear hyperplane-based function encoding approach called hyper binning and details the necessary conditions for encoding. Section 5.2.3 focuses on the analytical details of hyper binning for encoding functions to determine the optimal hyperplane allocation that maximizes a notion of mutual information between the function and the partitions. Section 5.2.4 provides a discussion on the connections between hyper binning and coloring-based coding schemes, and Sect. 5.2.5 draws some conclusions.

5.2.1 Background on Coding for Distributed Compression Starting from the distributed source coding scheme of Slepian–Wolf in Slepian and Wolf (1973), practical distributed encoding schemes have been developed, including coset codes (Pradhan and Ramchandran 2003), trellis codes (Wang and Orchard 2001), and turbo codes (Bajcsy and Mitran 2001). Other examples include rate region characterization using graph-entropy-based approaches, such as Körner (1973), Alon and Orlitsky (1996), Orlitsky and Roche (2001), Doshi et al. (2010), Feizi and Médard (2014), Feng et al. (2004), Gallager (1988), and coding for computation with communication constraints (Li et al. 2018; Kamran et al. 2019; Yu et al. 2018). While some approaches focus on network coding for computing linear functions, such as Kowshik and Kumar (2012), Huang et al. (2018), Appuswamy and Franceschetti (2014), Koetter and Médard (2003), Li et al. (2003), Ho et al. (2006), there exist works exploiting functions with special structures, e.g., in Shen et al. (2019), Gorodilova (2019), and Giridhar and Kumar (2005), and compression of sparse graphical data (Delgosha and Anantharam 2019). Another perspective to efficient representation is coding for functional compression. In Basu et al. (2020), the authors proposed a hypergraph-based coloring scheme whose rate lies between the Berger–Tung inner and outer bounds and showed that for independent sources, their scheme is optimal for general functions. In Servetto (2005), the author derived inner and outer bounds for multiterminal source coding. The author showed that for scalar codes, i.e., scalar quantizers followed by block entropy coders, the two bounds converge. In Misra et al. (2011), the authors considered the distributed functional source coding problem, in which

5 Hardware-Limited Task-Based Quantization in Systems

109

the sink node computes an estimate of the function under the mean-square error (MSE) distortion criterion. In complement to compression methods, the focus of the signal processing domain is on vector quantization and distributed estimation-based models. A vector quantization technique was proposed in Padmanabhan et al. (1999), to partition the feature space such that the average entropy of the class distribution in the partitioned regions is minimized. In Ribeiro and Giannakis (2006a), conditions for efficiently quantizing scalar parameters were characterized, and estimators that require transmitting just one bit per source that exhibits variance almost equal to the minimum variance estimator based on unquantized observations were proposed. Max–Lloyd algorithm, which is a Voronoi iteration method, was applied to vector quantization and pulse-code modulation (Max 1960). Vector quantization using linear hyperplanes was applied to distributed estimation in sensor networks in the presence of additive Gaussian noise (Fang and Li 2009) and with resource constraints (Ribeiro and Giannakis 2006b).

5.2.2 Toward Vector Quantized Functional Representations A common assumption in a point-to-point model of Shannon (1948) or communication systems is that signal is in general a discrete-time sequence .X(l), .l = 1, 2, . . . . The goal is to design a coding design for the best possible reconstruction given a distortion criterion. To that end, we first focus on a simplified model for discretevalued sources. Consider the distributed encoding scenario in Slepian and Wolf (1973) with two sources .X1 and .X2 of finite alphabets .X1 and .X2 , respectively, one destination that aims to compute the outcomes of the function .f (X1 , X2 ), and f is known both at the sources and the decoder. For a given blocklength n, we let .Xni = {Xi (l)}nl=1 = Xi (1), Xi (2), . . . , Xi (n), where .xi (l) corresponds to the .l th entry of n n source .i ∈ {1, n sequences drawn i.i.d. according to n2}. Let .(X1 , X2 ) be length n n n ∈ Xn . We assume that there is no feedback .p(x , x ) = p(x (l), x (l)) for . x 1 2 l=1 i 1 2 i from the decoder to the sources. In functional compression, the work (Feizi and Médard 2014) has shown that instead of sending source variables, it is optimal to send coloring variables. The destination then uses a look-up table to compute the desired function value by using the received colorings. More specifically, sending colorings of sufficiently large power graphs of characteristic graphs followed by source coding, e.g., Slepian–Wolf compression (Slepian and Wolf 1973), leads to an achievable encoding scheme provided that the functions satisfy some necessary conditions (Feizi and Médard 2014). While in some cases, the coloring problem is not NP-hard, in general, finding this coloring is an NP-complete problem (Cardinal et al. 2004). A special case of this scenario is when f is the identity function, and it has been studied by Slepian–Wolf in their landmark paper (Slepian and Wolf 1973). Indeed, Cover developed an asymptotically optimal encoding scheme using

110

D. Malak et al.

orthogonal binning (Cover and Thomas 2012). The orthogonal binning is such that the codewords are selected uniformly at random from each bin, and the bins are equally likely. The work in Malak and Médard (2020, 2023) provides a natural generalization of the approach in Slepian and Wolf (1973), to the case where f is a general continuous function that satisfies several additional properties. The idea behind hyper binning is to partition a high-dimensional codebook space into closed convex regions called hyper bins that capture the correlations between .X1 and .X2 as well as the dependency between the function f and .(X1 , X2 ). The key intuition is that closed convex sets have dual representations as an intersection of half-spaces. Using a finite set of hyperplanes, their intersection determines the hyper bins, i.e., the quantized outcomes of f . Via hyper binning, it is possible to accurately represent f up to a distortion. The quantization error can vanish by optimizing the number, parameters, and dimensions of the hyperplanes employed. As in Slepian–Wolf encoding (Slepian and Wolf 1973), in hyper binning (Malak and Médard 2020, 2023), each bin represents a typical sequence of function f ’s outcomes and is a collection of infinite length sequences. Hyper binning does not rely on NP-hard concepts such as finding the minimum entropy coloring of the characteristic graph of f . Unlike graph coloring, hyper binning with a sufficient number of hyperplanes in GP jointly partitions the source random variables in a way to achieve the desired quantization error at the destination for a given computation task. Given an entropy-based distortion measure as in, e.g., Courtade and Wesel (2011), we require the following condition on the number of hyperplanes J : 2n    J = min k : h(qj ) ≤  ,

.

k

(5.1)

j =k+1

J to ensure that .R1 + R2 = j =1 h(qj ) ≥ 1 − , where .h(p) is the binary entropy function, which satisfies .h(p) = −p log2 p − (1 − p) log2 (1 − p). Hyper binning naturally allows (a conditionally) independent encoding across the sources via an ordering of hyperplanes at each source prior to transmission and their joint decoding at the destination. This is possible with a helper mechanism that ensures the communication of the common randomness, i.e., the Gács–Körner common information (CI) (Gács and Körner 1973), characterized via hyperplanes. The CI measures provide alternate ways of compression for computing when there is common randomness between two jointly distributed source variables .X1 and .X2 (Yu et al. 2016b; Gács and Körner 1973). In distributed CI extraction, to the best of our knowledge, the Gács–Körner CI is the only CI that exploits the combinatorial structure of the distribution .pX1 ,X2 (x1 , x2 ) to decompose the sources into disjoint components of a bipartite graph before compression. The Gács–Körner CI extracts the same random variable from each source and is the maximum information that can be extracted from each source. We detail this CI scheme in Sect. 5.2.4.2. We next give the necessary conditions for distributed functional quantization of real-valued source random variables .X1 and .X2 via hyper binning. This binning scheme yields a partitioning of the joint sources’ data into convex sets .Pk , where

5 Hardware-Limited Task-Based Quantization in Systems

111

k ∈ {1, 2, . . . , M} such that M is the total number of partitions obtained from a set of hyperplanes in GP. Given a number of hyperplanes in GP, the hyperplanes attain the maximum number of partitions M. The function .f : (X1 , X2 ) → Z is such that the mapping .Z → {1, 2, . . . , M} is a bijection. Our model is restricted to a class of functions f satisfying that if .P(f ∈ Pk ) = 1, then .E[f ] ∈ Pk for each .Pk (Iosif et al. 2012). The function f has to be continuous at .(x1 , x2 ) ∈ (X1 , X2 ) since −1 (P ) is a neighborhood of .(x , x ) for every neighborhood .P of .f (x , x ) in .f k 1 2 k 1 2 .Z. However, the domain of f can be discrete (Feizi and Médard 2014). .

5.2.3 Designing Hyper Bins The hyper binning approach in Malak and Médard (2020, 2023) generalizes the classical compression algorithms, such as Slepian and Wolf (1973), Feizi and Médard (2014), that work on the quantized single-letter representation of data. Its primary focus is on continuous-valued sources. In this subsection, we first present the basic properties of hyperplanes and next discuss hyper binning, which is a linear hyperplane-based function encoding approach. We detail the coding rate of hyper binning, which can be characterized using the binary entropy function.

5.2.3.1

Data and Hyperplane Arrangement

Every linear hyperplane .η is an affine set parallel to an .(n−1)-dimensional subspace of .Rn (Dattorro 2005). Let .Hn be the space of hyperplanes in .Rn . A hyperplane n n .η ∈ H ⊂ R is characterized by the linear relationship given as follows: η(a, b) = {y ∈ Rn : a y = b}, a ∈ S n−1 , b ∈ R,

.

(5.2)

where .a is the nonzero normal and .S n−1 is the unit sphere. The design of hyper binning for distributed functional quantization builds on representing the source data by feature vectors. The feature vectors .{xt } lie in an n-dimensional space and are a mixture of Gaussian random variables. We assume that .{xt } are independent and belong to either of the sources .X1 and .X2 . We employ a linear hyperplane arrangement for classifying .{xt }. The model includes a total number of classes of M, where the number of feature vectors of class k is .nk , the total number of feature vectors is N , and .nk /N is the relative count of class k data. To describe the hyperplane arrangement, we need .J (n + 1) parameters, where J is the number of hyperplanes. The hyperplane parameters .{(aj , bj )}Jj=1 capture the characteristics of the joint source distribution and its relation with f . To achieve the desired distortion for a given function f , we shall choose J following (5.3) to represent or distinguish the desired number of distinct outcomes M of f . The orientations of hyperplanes will depend on the correlations between .X1 and .X2 as well as the correlations between .(X1 , X2 ) and f .

112

D. Malak et al.

Projection of .x ∈ Rn onto .η (Dattorro 2005, Ch. E.5) is given as P x = arg min x − y2 = x − a(a a)−1 (a x − b).

.

y∈H

We shall let s and J denote the number of sources and hyperplanes, respectively. A hyperplanearrangement of size J in an s-dimensional source space creates at most J  s J regions. Hyperplanes in general position (GP) divide the ≤ 2 .r(s, J ) = k=0 k space into .r(s, J ) regions (Rini and Chataignon 2019). For the sake of presentation, we focus on the case .s = 2.

5.2.3.2

Linear Discriminant Analysis for Classifying Source Features

We represent the source data by feature vectors .{xt } ∈ Rn that are mixtures of Gaussian variables. We use linear discriminant analysis (LDA) to distinguish different classes of feature vector combinations yielding the same function outcome. LDA is a classification technique for the separation of multiple classes of variables using the linear combinations of features, where the classes are known a priori. LDA works when the measurements on independent variables for each observation are continuous quantities. The set of features .{xt } for each sample of an event has a known class y. The classification problem is to find a good predictor for the class y of any sample of the same distribution given only an observation .xt . In LDA, the conditional probability density functions (pdfs) .p(xt |y = 0) and .p(x  t |y = 1) are both the normal distribution with mean and covariance parameters . μ0 ,  and .(μ1 , ), respectively. Hence, the Bayes optimal solution is to predict points as being from the second class if the log-likelihood ratio is bigger than some threshold, so that the decision criterion of .xt being in a class y is a threshold on the dot product .w · xt > c, i.e., a function of the linear combination of the observations, for some threshold c, where .w =  −1 (μ1 − μ0 ) and .c = w · 12 (μ1 + μ0 ). The observation belongs to y if the corresponding .xt is located on a certain side of a hyperplane perpendicular to .w. The location of the plane is defined by c. For multiple classes, the analysis can be extended to find a subspace that contains the class variability. The conditional pdfs .p(xt |y = k), .k = 1, . . . , M, for the feature vectors .{xt } of a given class k are independent Gaussian variables with a mean vector .μk and a covariance matrix .. The scatterbetween class variability M 1 T is the sample covariance of the class means .b = M l=1 (μl − μ)(μl − μ) , where .μ is the mean of the class means. The class separation in a direction .w is T T −1 . .S = w b w(w w) For each hyperplane .η(a, b), there are .n + 1 unknowns .a, b to be determined; hence, there are .(n + 1)J unknown hyperplane parameters in total. A given number of hyperplanes J in GP can support a feature vector in an n-dimensional space where the dimension is upper bounded as nmax = max [n | (n + 1)J ≤ r(s, J )].

.

n≥1

(5.3)

5 Hardware-Limited Task-Based Quantization in Systems Fig. 5.1 Maximum n vs. J in GP for a different number of sources s

10

113

6

10 4

10 2

10

0

0

20

40

60

80

100

This inequality is because the required number of hyperplane parameters, .(n + 1)J unknowns, should be smaller than the number of regions denoting the quantized function outcomes where each output is an equation represented by intersections of half-spaces. This result gives a necessary condition (a lower bound on J ) to support a feature vector in .Rn . In Fig. 5.1, we sketch the relation between .nmax and J . We observe that the number of dimensions a hyperplane arrangement can capture scales exponentially in the number of sources s (for .s > 2) vs. orthogonal binning that provides linear scaling of the total number of dimensions in s as J increases. For

the sake of presentation, .s = 2, and it follows from (5.3) that .nmax = J2 + O J1 , i.e., J linearly scales with n as .J ≈ 2nmax as n tends to infinity. Hyper binning embeds the quantization phase and is discrete, i.e., it does not require further post-quantization prior to compression. Hence, we can apply the scheme of Berger–Tung, detailed in Tung (1978) and Berger (1978), or the coding scheme in Feizi and Médard (2014) and its generalization in Basu et al. (2020) on the quantized representation. Since the compression gain of hyper binning over random binning lies in the pre-quantization aspect, any compression scheme, such as Tung (1978); Berger (1978); Feizi and Médard (2014); Basu et al. (2020), can be implemented on top of hyper binning to recover the function.

5.2.3.3

Optimizing Hyperplane Arrangement

To provide a joint characterization of sources by capturing their correlation as well as the features of the function f , we exploit LDA. In LDA, the encoded data are obtained by projecting the source data on a hyperplane arrangement and by looking at on which side of each hyperplane the vector lies. The criterion of a vector being in a class y is purely a function of this linear combination of the known observations. The observation belongs to y if the corresponding vector is located on a certain side of a hyperplane. We independently design each hyperplane. The hyperplane

114

D. Malak et al.

arrangement, i.e., the collection of hyperplane parameters .(a, b), depends on the particular function f and the distribution of the vectors .{xt }. For a hyperplane .η(a, b) described by vectors .a ∈ Rn and .b ∈ R as in (5.2), the projected feature vector .ut = a xt lies on one side of .η(a, b) if .ut ≤ b, and on the other side if .ut ≥ b. Mapping .xt to the .ut space is equivalent to computing the inner product of the feature vector and .a. As a result of this linear mapping of a high-dimensional Gaussian (with independent coordinates) to one-dimensional space, the distribution for the one-dimensional mapping outcome that models class k is also Gaussian distributed and has the following mean and variance, respectively: 

mk = μk a,

.

σ 2 = a a,

k = 1, . . . , M.

(5.4)

In our setup, the feature vectors lie in a high-dimensional space and form an independent set of Gaussian random variables. Because their linear projections onto hyperplanes are also Gaussian and independent, the notions of the set of hyperplanes and the feature vector classes are exchangeable. With a careful choice of the parameters .{mk }M k=1 and .σ , we can observe that: (i) projecting multiple vector classes onto a single hyperplane is equivalent to (ii) projecting a set of feature vectors .{xt } onto M hyperplanes to generate a total number of classes M where each class index k can be considered as a mapping from .{xt } to a hyper bin index. Using this analogy, we represent our model in (ii) via the multi-class interpretation in (i). In the multi-class interpretation, the number of feature vectors of class k that lie to the right of .b = a μ for a hyperplane characterized by .η(a, b) is given by .nk,r = nk pk , where the probability that a feature vector belongs to partition k, or equivalently it lies to the right of b, is given by pk = Q

.

|b − m |

k , σ

k = 1, . . . , M,

(5.5)

where .mk is the one-dimensional mapped mean  of .xt to the .ut space given that it belongs to class k, and .Q(z) = 12 erfc √z is the complementary cumulative 2 distribution function (CDF) of the standard Gaussian distribution such that .Q(z) → 0 as .z → ∞ monotonically. An observation is that as .σ increases, .pk becomes higher due to (5.5). As .σ increases, since .pk ’s also increase, .pM+1 increases. Furthermore, .pk increases in .mk given that .b ≥ mk . We assume that .pk is a fixed constant, and the function f determines the distribution .{pk }M k=1 .  In the multi-hyperplane interpretation, let .qj = P(aj xt ≥ bj ) be the probability that a feature vector lies to the right of .bj for a hyperplane .j = 1, . . . , J characterized by .η(aj , bj ), i.e., the tail probability of one-dimensional Gaussian J variable. Hence, the relation between .{pk }M k=1 and .{qj }j =1 satisfies pk =



.

j ∈Sk

qj

 j ∈S / k

(1 − qj ),

k = 1, . . . , M,

(5.6)

5 Hardware-Limited Task-Based Quantization in Systems

115

where .Sk is the set of the hyperplanes j for which the hyper bin k lies to the right of bj . Our goal is to determine the class of hyperplanes with optimal set of parameters J .{(aj , bj )} j =1 for each hyperplane such that if we assign the feature vectors at each source to one of two partitions based on whether .ut ≤ b (or equivalently .x a ≤ b), then the average of the entropy of the class distribution in the partitions is minimized (Padmanabhan et al. 1999). Minimizing the entropy of partitioning is equivalent to maximizing the mutual information associated with the partitioning, i.e., the difference between the entropy of function f and the average of the entropy of the partitions. Our goal is to minimize the entropy of the partitioning. To that end, we choose the following mutual information metric associated with hyper binning: .

I (M) = h(pM+1 ) −

.

M  nk k=1

N

h(pk ).

(5.7)

This metric captures the accuracy of classifying the function outcomes. The higher the entropy for the classification of M partitions, i.e., .{h(pk )}M k=1 , the lower .I (M) is. The trend of .I (M) in (5.7) depends on the distributions of the feature vectors. To maximize .I (M) via hyper binning, it is intuitive that .(a, b) should be such that 1 M .pk ’s are close to 0 or 1 to minimize .h(pk )’s, and .pM+1 = k=1 nk,r is close N to .0.5, i.e., there are an approximately equal number of feature vectors in the two partitions. Assume for optimal .I (M) that each .pk is approximately 0 or 1. As .σ increases, .h(pM+1 ) decreases, and the entropy .h(pk ) increases. For the asymmetric case where .ni is proportional to .pk , i.e., .nk ∝ pk , incrementing M will improve .I (M) because each added hyperplane will provide more information to distinguish the function outcomes. However, for the symmetric case where .nk = N/M, incrementing M beyond a certain number of hyperplanes will not help.

5.2.3.4

Binning for Distributed Source Coding

In this part, we first detail a fundamental limit for the asymptotic compression of distributed sources followed by an achievable random binning. We then contrast baselines that rely on random binning with hyper binning. Slepian–Wolf Compression This scheme is the distributed lossless compression setting with source variables .X1 and .X2 jointly distributed according to .pX1 , X2 , where the function .f (X1 , X2 ) is the identity function. The Slepian–Wolf theorem gives a theoretical bound for the lossless coding rate for distributed coding of the two statistically dependent i.i.d. finite alphabet source sequences .X1 and .X2 as (Slepian and Wolf 1973) RX1 ≥ H (X1 |X2 ), RX2 ≥ H (X2 |X1 ),

.

116

D. Malak et al.

RX1 + RX2 ≥ H (X1 , X2 ),

(5.8)

where .H (X) = E[− log2 (X)] is the entropy of X in bits. Similarly, .H (X1 , X2 ) is the joint entropy of .X1 and .X2 , and .H (X1 |X2 ) is entropy of .X1 conditioned on .X2 . Equation (5.8) implies that .X1 can be asymptotically compressed up to the rate .H (X1 |X2 ) (Slepian and Wolf 1973). This theorem states that making use of the correlation allows a much better compression rate to jointly recover .(X1 , X2 ) at a receiver at the expense of vanishing error probability for long sequences; it is both necessary and sufficient to separately encode .(X1 , X2 ) at rates satisfying (5.8). The codebook design is done in a distributed way, i.e., no communication is necessary between the encoders. Distributed codebook design for computing functions f on the data .(X1 , X2 ) at the receiver sites is challenging, irrespective of whether or not .X1 and .X2 are correlated. A random code construction for source compression that achieves this fundamental limit, i.e., the Slepian–Wolf rate region for distributed sources given in Slepian and Wolf (1973), has been provided by Cover in Cover (1975). This type of random binning is equivalent to orthogonal quantization of typical source sequences, as we will describe in Proposition 1. Proposition 1 (Cover’s Random Binning (Cover 1975)) Binning asymptotically achieves zero error for the identity function .f (X1 , X2 ) = (X1 , X2 ) when the encoders assign sufficiently large codeword lengths .nR1 and .nR2 in bits to each source sequence where .R1 > H (X1 ) and .R2 > H (X2 |X1 ). To showcase the compression gains of an optimally designed hyper bins versus other well-known binning methods, we next devise an example. Our goal is to explore how informative different types of partitioning can be for quantifying a function. Example 1 (Different Binning Methods for Distributed Functional Compression) Consider a functional compression problem where the sources .X1 and .X2 to be encoded are continuous-valued. We consider three ways of compressing the sources to recover an approximate representation of a function at the decoder. While random binning is asymptotically optimal, for ease of exposition, we first assume that the blocklength satisfies .n = 1. To indicate their main features, we illustrate the encoding for different binning schemes in Fig. 5.2, where .X1 ∈ [0, 1] and .X2 ∈ [0, 1] lie on the y- and x-axes, respectively. For example, in Slepian–Wolf encoding (left), each source independently and uniformly partitions the source outcome into 4 bins. Hence, there are .4 × 4 = 16 bins in total. The block binning scheme (middle) trims some of the bins in the encoding scheme of Slepian–Wolf because the function is piecewise constant or block, and there is no correlation across bins. This approach modularizes the encoding into uniform quantization and compression (bin trimming). In this example, there are 4 blocks, and each .Bk can be obtained via aggregating the bins of Slepian–Wolf. If the function is more general than a block function, orthogonal trimming may not work. Instead, hyper binning can leverage the function and its dependency on the jointly distributed

5 Hardware-Limited Task-Based Quantization in Systems

X2

1

Identity function (X1, X2) X1

X2

1

Block function fB(X1, X2) X1

SW(i, j)

117

B1

X2

General function f(X1, X2) X1

B2

P1 P2

0.27 B3

0

0 0

B4

0

1 Slepian-Wolf binning

P3 1

Block binning

-0.13

P4 0.67 0.09 Hyper binning

Fig. 5.2 Hyperplane organization. (Left) Slepian–Wolf binning (Slepian and Wolf 1973). (Middle) Function-sensitive, correlation-insensitive partitioning. (Right) Function- and correlationsensitive partitioning Xn1

Quantization (Orthogonal Binning) X1n → unif{1, . . . , 2nR1 }

Compression

R1

Rate

Encoder 1 Xn2

Quantization (Orthogonal Binning) X2n → unif{1, . . . , 2nR2 }

Decoder Compression

R2

ˆ n, X ˆ n) (X 1 2 Estimate of message

Rate

Encoder 2

Fig. 5.3 Distributed compression scheme of Slepian–Wolf (Slepian and Wolf 1973) via binning constructed using the asymptotically optimal approach in Cover and Thomas (2012)

sources via the regions created from the intersections of linear hyperplanes and can make the quantization phase function-oriented, where the hyperplane parameters J .{(aj , bj )} j =1 are adjusted according to the function .f (X1 , X2 ). As a result, this reduces the redundancy in compression because the quantization is tailored for recovering the intended function and is more effective. We next detail each binning scheme separately. We emphasize that for illustration purposes, we chose .n = 1. Binning Approach of Slepian–Wolf (Slepian and Wolf 1973) In this scenario, the sources first uniformly (scalar) quantize .xn1 ∈ [0, 1]n and .xn2 ∈ [0, 1]n into a discrete set using 2 bits each. The bin assignments .(m1 (xn1 ), m1 (xn2 )) ∈ [1 : 4] × [1 : 4] for the source pair .(Xn1 , Xn2 ) take .M = 16 possible outcomes that are equally likely. The Slepian–Wolf encoding scheme distinguishes all possible jointly typical outcomes. However, the binning scheme does not capture the function’s structure, i.e., it does not distinguish .f (Xn1 , Xn2 ) and .(Xn1 , Xn2 ) from each other. In this case with .M = 16 equally likely partitions (bins), .P((Xn1 , Xn2 ) = (i1 , i2 )) = 1/16, the entropy of the partitions equals .H (Xn1 , Xn2 ) = log2 (16) = 4. Then, .ISW = H (Xn1 , Xn2 ) − H (Xn1 , Xn2 ) = 0. We show the block diagram for independent encoding and joint decoding of two correlated data streams .Xn1 and .Xn2 in Fig. 5.3. Orthogonal Trimming of the Binning-Based Codebook When the function (on [0, 1]2 ) is piecewise constant in the blocks domain, then the uniform (scalar) quantization followed by trimming achieves an optimal encoding rate. The block binning or generalized orthogonal binning scheme can capture functions with the pair .(Xn1 , Xn2 ) having a blockwise dependence. In this example, there are 4 blocks

.

118

Xn1

Xn2

D. Malak et al.

Quantization

Quantization

Orthogonal Trimming Encoder 1 Orthogonal Trimming Encoder 2

Compression

R1

Rate Decoder

Compression

R2

fˆ(X1n , X2n ) Estimate of the function

Rate

Fig. 5.4 Orthogonal trimming of the random-binning-based (uniformly quantized bins) codebook

Bk , with indices .k = 1, . . . , 4, corresponding to different function outcomes. Hence, fB (Xn1 , Xn2 ) and .(Xn1 , Xn2 ) can be distinguished under this blockwise partitioning. This encoding scheme is easy to implement by combining some of the blocks prior to implementing the Slepian–Wolf encoding scheme in each .Bk . Clearly, this is more efficient than completely ignoring the function’s structure and directly implementing the Slepian–Wolf encoding. Hence, for sources sharing blockwise dependency, i.e., .H (fB (Xn1 , Xn2 )) < H (Xn1 , Xn2 ). In this example with 4 blocks, we use 3 hyperplanes, as shown inFig. 5.2 (middle). Hence, for block binning, n n .P(Bk ) = P(fB (X , X ) = k) = i1 ,i2 : fB =k pi1 i2 . The colored region .B2 has a 1 2 probability .P(B2 ) = 9/16. Similarly, .P(B1 ) = 3/16, .P(B3 ) = P(B4 ) = 2/16. This implies that the entropy of the partitions equals .H (fB (Xn1 , Xn2 )) = 1.67. In this case, block binning yields .IB = H (Xn1 , Xn2 ) − H (fB (Xn1 , Xn2 )) = 2.33. We show the block diagram for orthogonal trimming for piecewise constant functions in Fig. 5.4. . .

Hyper-Binning-Based Codebook If the function is not piecewise constant, then quantizing and then compressing may not be as good. The hyper binning scheme can capture the dependencies in the pair .(Xn1 , Xn2 ) and .f (Xn1 , Xn2 ), unlike the block binning scheme. In this scheme, we cannot consider the partitions .Pk , with indices .k = 1, . . . , 4, corresponding to function outcomes independently since each partition shares a non-orthogonal boundary to capture the dependency across the sources. With hyper binning, it is possible to jointly encode correlated sources as well as the function up to some distortion, determined by the hyperplane arrangement. As a result, for sources with dependency (more general than blockwise dependency), we can achieve .H (f (Xn1 , Xn2 )) < H (fB (Xn1 , Xn2 )). We partition the region using 2 hyperplanes in GP by incorporating the correlation structure between the function and the sources. In this case, .P(P1 ) = 0.375, .P(P2 ) = 0.531, .P(P3 ) = 0.031, .P(P4 ) = 0.063, and the entropy of the partitions satisfies n n .H (f (X , X )) = 1.42 for each k. Hence, the hyper binning model yields .I (M) = 1 2 n n H (X1 , X2 ) − H (f (Xn1 , Xn2 )) = 2.58. 5.2.3.5

Compression at Finite Blocklengths and Hyper Binning

For finite blocklengths, the rate limits in (5.8) do not hold. In that case, we can exploit the notion of Kolmogorov complexity .K(x n ), which is the minimum

5 Hardware-Limited Task-Based Quantization in Systems

119

description length of a string .x n . Let .Xn be i.i.d. integer-valued

variables with K(Xn ) be the average entropy .H (X), where .X is their finite alphabet, and .E n shortest description length of length-n sequence .Xn . Then, there is a constant c such that the relation of Kolmogorov complexity and entropy for all n satisfies (Cover and Thomas 2012, Ch. 7.3)  H (X) ≤ E

.

 |X| log n c K(Xn ) ≤ H (X) + + . n n n

(5.9)

In random binning, the .J = − log() bit quantization of .Xn1 has an entropy of approximately .h(Xn1 ) + J , where the quantization bin length . satisfies . = 2−J . For the J -bit quantization of a string .xn1 , we obtain the average description length via the addition of . Jn bits on both sides of (5.9) as H (X ) ≤ E

.

K({X (l)}n ) c |X | log n  l=1 + , ≤ H (X ) + n n n

where .X is the alphabet for the quantized variable .X with .|X | = 2J , and 1 J n .H (X ) ≈ n h(x1 ) + n bits. The finite length n description of J -bit quantization n of .Xi , for .i ∈ {1, 2}, requires an additional . |X |nlog n bits on top of quantization.

From (5.3), we have .n ≤ J2 + O J1 . Combining this with (5.9), the representation complexity of random binning due to separation of quantization and compression phases is approximately 2 bits higher than that of hyper binning. Hyper binning, unlike orthogonal binning, eliminates the need for post-quantization. The J -bit vector quantization is tailored for the functions, and each function outcome relies on a collection of binary decisions. This process does not involve quantization of continuous variables, i.e., approximating the differential entropy via the addition of J bits. The complexity is solely determined based on the binary entropy function. We show the diagram of hyper binning for compression in Fig. 5.5. Sampling is a suboptimal single-letter approach. In information theory, coding and compression typically follow signal processing. Hyper binning does the compression step after signal processing and before coding. It captures the entire data vector instead of a single-letter representation, giving a functional equivalence of vector quantization.

Xn1

Hyperbinning {η(aj , bj )}Jj=1

1{aj X1n >bj }

J j=1

Compression

R1

Rate

Encoder 1 Xn2

Hyperbinning {η(aj , bj )}Jj=1

1{aj X2n >bj }

Decoder J j=1

Compression

R2 Rate

fˆ(X1n , X2n ) Estimate of the function

Encoder 2

Fig. 5.5 Hyper binning, which generalizes the orthogonal compression to convex regions determined by the intersections of hyperplanes

120

D. Malak et al.

5.2.4 A Discussion on Computational Information Theory In this section, to devise a new perspective to computational information theory, we provide connections between our distributed computationally aware quantization scheme that relies on hyper binning and the coloring-based coding models for distributed functional compression. First, in Sect. 5.2.4.1, we describe coloring-based modular coding models that decouple coloring from Slepian–Wolf compression. Next, in Sect. 5.2.4.2, we shift our focus to describe an achievable encoding for hyper binning and detail the encoding implementation in 3 steps.

5.2.4.1

Coloring-Based Coding Schemes Versus Hyper Binning

Since the sources cannot communicate with each other, the only way to rate reduction is through a source defining its equivalence class for functional compression. We next give a block function example for which codebook trimming followed by the Slepian–Wolf encoding is asymptotically optimal. Example 2 (A Trimmable Codebook) Assume that sources .X1 ⊥ ⊥ X2 and are uniformly distributed over the same alphabet .X = {0, 1, 2, 3}. The function is .f (X1 , X2 ) = X1 ⊕ X2 . This function exhibits the piecewise block behavior. Given the function, source 1 can determine an equivalence class .[x1 ] that is mapped to .f (x1 , X2 ). Similarly, source 2 can determine an equivalence class .[x2 ] mapped to .f (X1 , x2 ). For this example, .[0] = [2] and .[1] = [3] for both .X1 and .X2 , i.e., each source needs 1 bit to identify themselves since the data distributions are uniform. However, the entropy of the function is 1 bit as there are only 2 equally likely classes. To compute the function, each source specifies its equivalence class without any help from the other source. To specify its equivalence class .[x1 ], source 1 transmits .R1 = 1 bit. Similar arguments follow for source 2 and .R2 = 1. Hence, .R1 +R2 = 2. In this example, each equivalence class is equiprobable and has the same size, which is 2 per source, making the setup more tractable. While for a specific class of functions, the random binning approach works, or we can orthogonally trim the binning-based codebook, we conjecture that orthogonal binning or trimming may not optimize the rate region for general functions (even without correlations). However, as authors have shown in Basu et al. (2020) that for independent sources, the Berger–Tung inner and outer bounds converge, and hence, their hypergraph-based coloring scheme rate lies between the bounds of Tung (1978) and Berger (1978) and is optimal for general functions. For functions with particular structures, e.g., a piecewise block function, we can trim the binning-based codebook, as we detailed in Example 1. In general, trimming may not work, e.g., a smooth function. We next provide an example where orthogonal binning of a codebook is suboptimal for distributed functional compression.

5 Hardware-Limited Task-Based Quantization in Systems Fig. 5.6 Source combinations for computing .f (X1 , X2 ) in Example 3 for which trimming of orthogonal codebook does not hold. The source pairs with different outputs have different patterns

121

X2 (ˆ xn1 , x ˆn2 )(˜ xn1 , x ˆn2 )

X1 (ˆ xn1 , x ˜n2 ) (˜ ˜n2 ) xn1 , x

Example 3 Let .f (X1 , X2 ) = (X1 · X2 ) mod 2 with discrete alphabets .X1 = {1, 2, 3, 4} and .X2 = {0, 1}. We infer that f = 0 ⇒ x˜1 ∈ X1 , x˜2 = 0,

.

or

xˆ1 ∈ {2, 4}, xˆ2 = 1.

f = 1 ⇒ x˜1 ∈ {1, 3}, xˆ2 = 1, but .f (3, 1) = f (2, 1). We illustrate the sources with distinct outcomes in Fig. 5.6, indicating that orthogonal trimming may not work even if sources have no correlation. From Example 3, we conjecture that orthogonal binning is in general not efficient when computing general functions and/or with correlated sources. To see that when the decoder observes .f (xˆ1n , xˆ2n ), it is possible that .f (xˆ1n , xˆ2n ) = f (x˜1n , x˜2n ) for some source pair .(x˜1n , x˜2n ) = (xˆ1n , xˆ2n ). In this case, the bins cannot be combined since .f (x ˆ1n , x˜2n ) = f (x˜1n , xˆ2n ) in general. Hence, orthogonal binning is clearly suboptimal. Exploiting the notion of characteristic graphs introduced by Körner in Körner (1973), authors in Alon and Orlitsky (1996); Orlitsky and Roche (2001) have recently devised coloring-based approaches and used them in characterizing rate bounds in various functional compression setups. We use the notation .HGX1 (X1 ) to represent the graph entropy for the characteristic graph .GX1 that captures the equivalence relation source .X1 builds for a given function f on the source random variables .(X1 , . . . , Xs ). Similarly, the other sources have characteristic graphs .GX1 , . . . , GXs , respectively. Definition 1 ((Feizi and Médard 2014, Defn. 19)) A joint-coloring family .VC = {vc1 , . . . , vcl } for .Xi with any valid colorings .cGXi for .i = 1, . . . , s is such that

each .vci , called a joint-coloring class, is the set of points .(x1i1 , x2i2 , . . . , xsis ) whose coordinates have the same color, i.e., .vci = {(x1i1 , x2i2 , . . . , xsis ), (x1l1 , x2l2 , . . . , xsls ) : cGX1 (x1i1 ) = cGX1 (x1l1 ), . . . , cGXs (xsis ) = cGXs (xsls )}, for any valid .i1 , . . . , is and i i .l1 , . . . , ls . .vc is connected if between any two points in .vc , there exists a path that i lies in .vc .

For any achievable coloring-based coding scheme, authors in Doshi et al. (2010) have provided a sufficient condition called the zig-zag condition, and authors in Feizi and Médard (2014) both a necessary and sufficient condition called the

122

D. Malak et al.

coloring connectivity condition (CCC). These are modular schemes that decouple coloring from Slepian–Wolf compression. We next state the condition in Feizi and Médard (2014). Definition 2 ((Feizi and Médard 2014, Defn. 20)) Let .Xi be random variables with any valid colorings .cGXi for .i = 1, . . . , s. A joint-coloring class .vci ∈ VC satisfies CCC when it is connected, or its disconnected parts have the same function values. Colorings .cGX1 , . . . , cGXs satisfy CCC when all joint-coloring classes satisfy CCC. CCC Versus Orthogonal Binning CCC ensures the conditions for orthogonal binning, i.e., codebook trimming. A coloring-based encoding that satisfies CCC is applicable to Example 2. However, it may be suboptimal for functions not allowing for trimming, see Example 3. Let .x˜1n ∈ {1, 3} and .xˆ1n ∈ {2, 4} and .x˜2n = 0 and .x ˆ2 = 1. Note that .(xˆ1n , xˆ2n ) ∼ (xˆ1n , x˜2n ) and .(xˆ1n , x˜2n ) ∼ (x˜1n , x˜2n ) (CCC preserved). However, .(x˜1n , x˜2n ) ∼ (x˜1n , xˆ2n ) (CCC not preserved). Hence, CCC is necessary for trimming. This function also explains the suboptimality of coloring-based coding in general.

5.2.4.2

An Achievable Encoding Scheme for Hyper Binning

We next provide a high-level abstraction for an achievable encoding of hyper binning with .s = 2 sources. For a function .f (X1 , X2 ) known both at the sources and at the destination, let .{η1 , η2 , . . . , ηJ } ∈ H2 ⊂ R2 be the hyperplane arrangement of size J in GP that divides .R2 into exactly .M = r(2, J ) regions and is designed to sufficiently quantize .f (X1 , X2 ). The goal is to predetermine J .{(aj , bj )} j =1 that maximizes .I (M). These parameters are known at both sources and sent to the destination only once. The Gács–Körner Common Information Carried via Hyperplanes To enable the distributed computation for non-decomposable functions, we envision a helperbased distributed functional compression approach. Hyper binning requires the transmission of common randomness between the source data and across the data and its function, captured through the hyperplanes. The common information (CI) measures provide alternate ways of compression for computing when there is common randomness of the jointly distributed sources (Yu et al. 2016b; Gács and Körner 1973). Among these measures, the Gács–Körner CI has applications in the private constrained synthesis of sources and secrecy (Salamatian et al. 2016) and is relevant here because it can be separately extracted from either marginal of .X1 and .X2 (Gács and Körner 1973). In our distributed quantization setting, the helper should communicate in a prescribed order with the hyperplane parameters that are .J (n + 1) in total. The rate of CI is the rate of compressing the parameters J . While these parameters are real-valued, they have approximate .{(aj , bj )} j =1 floating-point representations. Furthermore, while they might need to be updated

5 Hardware-Limited Task-Based Quantization in Systems

123

with n, from (5.3), the update rates of J and hence of the hyperplane parameters are logarithmic in n. Encoding Each source .Xi , .i = 1, 2 independently determines an ordering of hyperplanes to compress .Xi . Let these orderings be .OXi ⊆ πXi ({η1 , η2 , . . . , ηJ }), where .πXi is the permutation of the hyperplane arrangement from the perspective of source i. Note that .πXi1 = πXi2 for .i1 = i2 because sources might build different characteristic graphs. Source i determines an ordering .OXi , which is from the most informative, i.e., decisive in classifying the source data, to the least such that the first bit provides the maximum reduction in the entropy of the function outcome. Transmission Because each source has the knowledge of .{(aj , bj )}Jj=1 , it does the comparisons .aj xt ≥ bj for hyperplane j and sends the binary outcomes of these comparisons. Hence, each source needs to send at most J bits (1 bit per hyperplane) to indicate the region representing the outcome of f . There are at J

most .2J possible codewords, among which nearly .|C|HP = 2 j =1 h(qj ) are typical. Source i transmits a codeword that represents a particular ordering .πXi . Hence, in the proposed scheme with J hyperplanes, we require up to 2J bits to describe a function with .M = r(2, J ) outcomes. This is unlike the Slepian–Wolf setting, where source i has approximately .|C|SW = 2nH (Xi ) codewords to represent the typical sequences with blocklength n as n goes to infinity (Slepian and Wolf 1973). Hence, an advantage of the hyper binning scheme over the scheme of Slepian– Wolf is that it can capture the growing blocklength n with J hyperplanes without exceeding an expected distortion. Note that as hyper binning captures the correlation between the sources as well as between the sources and the function, it provides a representation with a reduced codebook size .|C|HP < |C|SW for distributed functional compression. If using .J  n hyperplanes ensures that the majority of .qj is in .{0, 1}, then the efficiency of the function representation is obvious. However, if J linearly scales with n, since .HGXi (Xi ) is the entropy of the characteristic graph that source i  builds to distinguish the outcomes of f (Körner 1973), a sufficient condition for . Jj=1 h(qj ) ≈ nHGXi (Xi ) is that .h(qj ) ≈ Jn HGXi (Xi ), .∀ j . Reception At the destination, each codeword pair received from the sources yields a distinct function output that can be determined by the specific order of the received bits in the codebooks designed for evaluating the outcome of f along with the CI carried via the hyperplanes. We note that the achievable schemes in Sects. 5.2.4.1 and 5.2.4.2 are suboptimal in some cases. However, hyper binning is not modular, unlike the coloring-based approaches, e.g., graph coloring followed by Slepian–Wolf compression in Feizi and Médard (2014) or its hypergraph-based extension in Basu et al. (2020). Hyper binning does not involve a coloring step or a separate quantization phase prior to the compression phase. Instead, it jointly performs the quantization and compression phases. This joint design is possible through the knowledge of the hyperplane parameters at the source sites.

124

D. Malak et al.

5.2.5 Conclusions We detailed various approaches to distributed functional compression, including distributed source compression algorithms that in general focus on quantizing continuous variables and then compressing them (Slepian and Wolf 1973; Feizi and Médard 2014), and hyper binning (Malak and Médard 2020, 2023), which is a function-aware quantization scheme for distributed functional compression. Hyper binning does the compression step on the functional representation, providing a natural generalization of orthogonal binning to computation and a fresh perspective to vector quantization for computing. Optimizing the tradeoff between the number of hyperplanes and the blocklength is crucial in exploiting the high-dimensional data, especially in a finite blocklength setting. This model can also be enhanced to adapt to the changes and learn from data by successively fine-tuning the hyperplane parameters with the growing data size. Due to Kolmogorov complexity (Cover and Thomas 2012), for finite blocklengths, hyper binning can be iteratively refined to capture the function accurately at a lower cost than random binning. In this section, we did not explicitly account for the constraints of analog-to-digital converters in quantization for computing functions. To that end, in the next section, we review how quantization can be tailored for specific tasks under storage and power constraints in hardware.

5.3 Hardware-Limited Task-Based Quantization in Systems In practical systems, the received continuous-time analog signals are first sampled and quantized by analog-to-digital converters (ADCs) to obtain a set of discrete samples, based on which the desired information can be extracted using digital signal processing algorithms (Eldar 2015). Conventional ADCs are designed to facilitate recovery of the received signals by sampling at the Nyquist rate of the received signals and using high-resolution quantizers. However, to meet the ever-increasing demand for higher data rates nowadays, the dimensionality and bandwidth of the received signals can be extremely high. For example, in beyond 5G wireless communication systems, a large number of antennas, i.e., massive multiple-input multiple-output (MIMO), and large bandwidths in the millimeter wave (mmWave) bands are employed (Heath et al. 2016). Consequently, using conventional ADCs and the associated information extraction schemes can be inefficient since the power consumption of ADCs and the required storage memory grow with the sampling rate and quantization resolution. To address this issue, in Shlezinger et al. (2019c), a hardware-limited taskbased quantization scheme was proposed by exploiting the fact that, in practice, signals are usually acquired in order to extract some underlying information, i.e., for a specific task. The proposed quantization system first introduces an analog combiner that reduces the dimensionality of the input to the quantizers, and then

5 Hardware-Limited Task-Based Quantization in Systems

125

scalar quantizers are employed considering practical hardware limitations, followed by a digital domain processing module. Under the constraint that the total number of quantization bits is fixed and finite, the reduction of signal dimensionality allows more bits for each sample, mitigating the effect of the structure imposed by hardware limitations. Experiment results show that properly designed hardwarelimited systems can approach the optimal performance achievable with vector quantizers, and that by taking the underlying task into account, the quantization error can be negligible with a relatively small number of bits, addressing the power consumption and storage issues of conventional quantization schemes. In this section, we introduce the basics of task-based quantization. Section 5.3.1 introduces the general system model of hardware-limited task-based quantizers. Sections 5.3.2 and 5.3.3, respectively, specialize the general system model to linear and quadratic tasks. Section 5.3.4 provides numerical results, followed by some conclusions in Sect. 5.3.5.

5.3.1 System Model We begin by introducing the general model of the hardware-limited task-based quantization system proposed in Shlezinger et al. (2019c). In contrast with conventional quantization systems where ADCs are fixed and the task is performed separately in the digital domain, task-based quantization systems jointly design the analog and digital parts as well as the bridges in between, i.e., the ADCs. In particular, as shown in Fig. 5.7, the aim is to recover the task vector .s ∈ Rk , which is statistically related to the observed signal .x ∈ Rn with function .fx|s . This formulation accommodates a broad range of applications, for example, in radar systems, .s can represent the unknown target parameters, while .x is the received signal. To recover .s, the observed signal .x is first projected into .Rp , .p ≤ n, by using an analog combiner .ha (·). Then, each entry of .ha (x) is quantized using the same scalar quantizer with resolution .M˜ p  M 1/p , whose operation is denoted as .Q1˜ (·). The symbol M is the overall number of quantization levels, which Mp

represents the memory requirement of the system and is also directly related to the

( h ( x)) a

s

f x|s

x

1

Scalar quantizer

QM1 p (◊ )

Post-quantization

Pre-quantization

ha :

n

hd :

p

Scalar quantizer

( h ( x)) a

p

QM1 p ( ◊)

Fig. 5.7 Hardware-limited task-based quantizer (Shlezinger et al. 2019c)

p

k

sˆ

126

D. Malak et al.

ADC power consumption. In the following, the value of M is kept fixed and finite; thus different selections of p will result in different quantization resolutions since ˜ p )p ≤ M. In the digital domain, the representation of .s, denoted as .sˆ, is obtained .(M as the output of the post-quantization processing module .hd (·): .Rp −→ Rk . The task-based quantization framework can be summarized as  

sˆ = hd Q1M˜ ((ha (x))1 ) , · · · , Q1M˜ (ha (x))p .

.

p

p

(5.10)

The novelty of the framework, compared to previous works on quantization for specific tasks with serial scalar ADCs (Li et al. 2017a; Choi et al. 2016), is in the introduction of the analog combiner. The role of the analog combiner is to reduce the dimensionality of the input to the quantizers, i.e., .p ≤ n in an optimal way so that the required information about .s is preserved even after low-bit simple scalar quantization. In practical applications, e.g., in MIMO communications, using such a combiner reduces the number of radio-frequency (RF) chains and thus facilitates low-cost hardware implementation. In addition, the mixing of the combiner helps increase the information content about .s after scalar quantization. The problem in task-based quantization is to jointly design the analog combiner 1 (·), and the digital processing part .h (·) .ha (·), the quantization rule of .Q d ˜ Mp

according to an appropriate metric. However, explicitly characterizing the general quantization system is difficult. Therefore, Shlezinger et al. (2019c) and Salamatian et al. (2019a) focus on scenarios in which the stochastic relationship between .s and .x is linear or quadratic, as summarized in the following subsections.

5.3.2 Linear Estimation Tasks To design the task-based quantizer, we first consider scenarios in which the stochastic relationship between the task vector .s and the observations .x is linear. One example can be the channel estimation problem of wireless communication systems, where the task vector .s represents the unknown channel. By defining the pilot matrix as .D, the received signal can be expressed as a linear function of .s, that is .x = Ds + w, with .w denoting the additive noise. In such cases, the optimal linear estimation of .s from .x is given by .s˜ = x, where . is the linear minimum mean-square error (MMSE) estimation matrix. Accordingly, the analog combiner and the digital processing module are restricted to be linear, namely, .ha (x) = Ax, .A ∈ Rp×n , and .hd (u) = Bu, .B ∈ Rk×p . Furthermore, to formulate the input–output relationship of the scalar quantization 1 (·), in Shlezinger et al. (2019c), the analysis is carried out by assuming dithered .Q ˜ Mp

quantization (Gray and Stockholm 1993). Specifically, under the assumption that the input is inside the dynamic range of the quantizer, the output can be written as the sum of the input and an additive white quantization noise signal. Therefore, the structure of the hardware-limited task-based quantization system depicted in

5 Hardware-Limited Task-Based Quantization in Systems

127

Fig. 5.7 is now a linear one. This facilitates an explicit derivation of the achievable distortion and to characterize the system that achieves minimal distortion. For any quantized representation .sˆ, it follows from the orthogonality principle that the meansquare error (MSE), .E[||s − sˆ||2 ], equals the sum of the estimation error of the MMSE estimate, .E[||s−˜s||2 ], and the distortion with respect to the MMSE estimate, .E[||˜ s − sˆ||2 ]. In the following, for simplicity, the MSE distortion refers to the second term, i.e., .E[||˜s − sˆ||2 ], since the first term does not depend on .sˆ (Shlezinger et al. 2019c). Let . x be the covariance matrix of observations .x, .γ the dynamic range of the scalar quantizers, .zl , l = 1, . . . , p the dither signals, and .A◦ and .B◦ , respectively, the optimal analog and digital processing matrices that achieve the minimal MSE distortion. We then have the following results (Shlezinger et al. 2019c): Theorem 1.3.1 For any analog combining matrix .A and dynamic range .γ such that .Pr(|(Ax)l + zl | > γ ) = 0, namely, the quantizers operate within their dynamic range with probability one, the digital processing matrix that minimizes the MSE is given by  ◦

B (A) =  x A

.

T

2γ 2 Ip A x A + 3M˜ p2

−1

T

.

(5.11)

Theorem 1.3.2 For the hardware-limited quantization system based on the model depicted in Fig. 5.7, the analog combining matrix .A◦ is given by .A◦ = −1/2 UA A VTA  x , where: 1/2 • .VA ∈ Rn×n is the right singular vectors matrix of .˜   x . • .A ∈ Rp×n is a diagonal matrix with diagonal entries

(A )2i,i =

.

 where .κp = η2 1 −

η2 3M˜ p2

+ 2κp ζ · λ,i −1 , ˜ 3M˜ p2 · p

−1 with .η denoting a constant that is set to guarantee

that the quantizer operates within the dynamic range (Shlezinger et al. 2019c), ˜ {λ,i ˜ } are singular values of . arranged in a descending order, and .ζ is chosen such that

.

.

p

+ 2κp  ζ · λ,i − 1 = 1. ˜ 3M˜ p2 · p i=1

• .UA ∈ Rp×p is a unitary matrix that guarantees that .UA A TA UTA has identical diagonal entries. The dynamic range of the quantizer is given by

128

D. Malak et al.

η2 .γ = p 2



η2 1− 3M˜ p2

−1 (5.12)

,

and the resulting minimal achievable distortion is

E[||˜s − sˆ||2 ] =

.

⎧ k ⎪ ⎪ ⎪ ⎨ i=1 p ⎪ ⎪ ⎪ ⎩ i=1

λ2˜ ,i

+ , +1 ζ ·λ,i ˜ −1 λ2˜ ,i

+ +1 ζ ·λ,i ˜ −1

+

p≥k k

2 , i=p+1 λ,i ˜

(5.13) p < k.

The characterization of the task-based quantization system in Theorem 1.3.2 gives rise to some non-trivial insights (Shlezinger et al. 2019c): (1) in order to minimize the MSE, p must not be larger than the rank of the covariance matrix of the MMSE estimate .s˜. This implies that reducing the dimensionality of the input prior to quantization contributes to recovering the task vector as higher resolution quantizers can be used without violating the overall bit constraint; and (2) when the covariance matrix of .s˜ is non-singular, quantizing the MMSE estimate minimizes the MSE if and only if the covariance matrix of .s˜ equals .Ik up to a constant factor. This indicates that, except for very specific models, quantizing the entries of the MMSE estimate vector, which is the optimal strategy when using vector quantizers (Wolf and Ziv 1970), does not minimize the MSE when using uniform scalar ADCs.

5.3.3 Quadratic Estimation Tasks In some scenarios such as covariance estimation (Rodrigues et al. 2017) and direction of arrival (DOA) recovery (Yu et al. 2016a), the desired parameter can be extracted from a quadratic function of the observed signal. In this subsection, we show how the hardware-limited task-based quantization scheme provided in the previous subsection can be applied for tasks that are a quadratic function of .x. Assume the observations .x ∈ Rn obey a Gaussian distribution, and the task is to recover a set of quadratic functions .{xT Ci x}ki=1 , where each .Ci ∈ Rn×n satisfies T .E[x Ci x] < ∞. For example, in the problem of covariance estimation, each element of the desired covariance matrix .E[xxT ] can be represented as a quadratic function of .x with appropriate .Ci . Correspondingly, in the hardware-limited taskbased quantization system, these desired quadratic quantities can be represented by the vector .s ∈ Rk with entries .(s)i  xT Ci x. By defining the random vector 2 ¯  vec(xxT ) and the matrix .G ∈ Rk×n whose ith row is given by .vecT (Ci ), .x and the task vector .s can be written as .s = G¯x. We then use the linear task-based quantization framework described in the previous subsection for the recovery of ¯ . Under the constraint that the overall number of .s from its linear measurements .x

5 Hardware-Limited Task-Based Quantization in Systems

129

quantization levels, that is, M, is fixed, the achievable MSE is given in the following theorem (Shlezinger and Eldar 2020; Salamatian et al. 2019a): Theorem 1.3.3 For any analog combining matrix .A and dynamic range .γ such that .Pr(|(A¯x)l + zl | > γ ) = 0, namely, the quantizers operate within their dynamic range with probability one, the following MSE is achievable: ⎛



MSE(A) = Tr ⎝G x¯ GT − G x¯ AT

.

2γ 2 Ip A x¯ AT + 3M˜ p2

⎞

−1

A x¯ GT ⎠ , (5.14)

where . x¯ denotes the covariance matrix of .x¯ . The minimum MSE is achievable by setting the digital matrix .B as  B = G x¯ AT

.

2γ 2 Ip A x¯ AT + 3M˜ p2 −1/2

and the analog matrix .A as .A = UA A VTA  x¯ • .VA ∈ R

n2 ×n2

• .A ∈ R

p×n2

−1 ,

(5.15)

, where:

˜  G 1/2 . is the right singular vectors matrix of .G x¯ is a diagonal matrix with diagonal entries (A )2i,i =

.

 where .κp = η2 1 −

η2 3M˜ p2

+ 2κp ζ · λG,i −1 , ˜ 3M˜ p2 · p

−1 with .η denoting a constant that is set to guarantee

that the quantizer operates within the dynamic range (Salamatian et al. 2019a), ˜ {λG,i ˜ } are singular values of .G arranged in a descending order, and .ζ is set such that

.

.

p

+ 2κp  ζ · λG,i − 1 = 1. ˜ 3M˜ p2 · p i=1

• .UA ∈ Rp×p is a unitary matrix that guarantees that .UA A TA UTA has identical diagonal entries. By properly pre-processing the observations, the linear task-based quantization framework can be applied to the quadratic setting, and significant reduction on the data dimensionality can be achieved by using an analog matrix, i.e., from .n2 to p. This allows more bits and thus higher resolution for each channel compared with the task-ignorant scheme, under the assumption that the total number of bits is fixed. For more complex scenarios where accurate knowledge of the statistical relationship between the observations and the task, that is, .fx|s , is not available or .fx|s is too

130

D. Malak et al.

complex to be expressed in linear or quadratic forms, Shlezinger and Eldar (2021) propose to learn it in a data-driven manner, in particular, by utilizing machine learning (ML) methods. The proposed deep task-based quantization scheme learns the analog and digital processing parts, parameterized as layers of deep neural networks (DNNs), and the scalar ADC, which is modeled as an activation function between two intermediate layers, in an end-to-end manner from a set of training data. With such a system architecture, tasks including estimation and classification can be performed by respectively setting the loss functions as empirical MSE and cross-entropy; further details can be found in Shlezinger and Eldar (2021).

5.3.4 Applications and Numerical Study To evaluate the performance of the hardware-limited task-based quantization system shown in Fig. 5.7, two scenarios involving linear and quadratic parameter acquisition from quantized measurements are considered: finite intersymbol interference (ISI) channel estimation in Sect. 5.3.4.1 and covariance recovery in Sect. 5.3.4.2 (Shlezinger et al. 2019c; Salamatian et al. 2019a).

5.3.4.1

ISI Channel Estimation

Consider the estimation of an ISI channel from quantized observations, as in Zeitler et al. (2012). Here, the task vector .s represents the coefficients of a multipath channel with k taps. The channel is estimated from a set of .n = 120 noisy observations .x, given by (x)i =

k 

.

(s)l ai−l+1 + wi ,

i ∈ {1, 2, . . . , n},

(5.16)

l=1

where .{ai } is the known training sequence, and .{wi } are samples from an i.i.d. zero-mean unit variance Gaussian noise process independent of .s. In particular, the channel .s is modeled as a zero-mean Gaussian vector with covariance matrix . s , with .( s )i,j = e−|i−j | , .i, j ∈ {1, 2, . . . , k}, and the training sequence is given by .ai = cos(2π i/n) for .i > 0 and .ai = 0 otherwise. Note that .s and .x are jointly Gaussian, and thus the MMSE estimate .s˜ is a linear function of .x so that the linear task-based quantization framework is employed here. Figure 5.8 shows the achievable MSE distortion of the resulting hardwarelimited task-based quantization system with .p = k = 2, and the total number of quantization bits .logM ∈ [2k, 8k]. Since dithering increases the energy of the quantization noise, we also compute the achievable MSE of the proposed systems when the ADCs implement uniform quantization without dithering. Furthermore, comparisons with the optimal task-based vector quantizer and the task-ignorant

5 Hardware-Limited Task-Based Quantization in Systems

131

0.85 No quantization Optimal, lower bound Optimal, upper bound Task-ignorant Scalar ADC, w dither Scalar ADC, w/o dither

0.8 0.75

Distortion

0.7 0.65 0.6 Achievable optimal distortion region

0.55 0.5 0.45 4

6

8

10

12

14

16

Number of bits log2 M

Fig. 5.8 Distortions of ISI channel estimation versus the number of bits (Shlezinger et al. 2019c)

quantizer discussed in Shlezinger et al. (2019c) are also provided. It can be seen that hardware-limited task-based quantizers substantially outperform task-ignorant vector quantization and approach the optimal performance as M increases. In particular, when each scalar quantizer uses at least five bits, i.e., .logM ≥ 5k, the quantization error becomes negligible, and the overall distortion is effectively the MMSE. This implies that compared with the conventional high-resolution quantization scheme, the task-based quantization can achieve comparable performance with a small number of bits, addressing the power consumption and storage issues of conventional ADCs.

5.3.4.2

Empirical Covariance Estimation

We next consider applying the task-based quantization system in recovering quadratic functions (Salamatian et al. 2019a). In particular, consider an empirical covariance estimation problem where the input is given by .x = [vT1 , . . . , vT4 ]T with 4 .{vi } i=1 denoting i.i.d. .3 × 1 zero-mean Gaussian random vectors. Therefore, the length of .x is .n = 12. The entries of the covariance matrix of .vi , denoted as . v , are .( v )i,j = e−|i−j | . The parameter of interest is the .3 × 3 empirical covariance  matrix . 14 4i=1 vi vTi , which is completely determined by its upper triangular matrix, stacked as the desired task vector .s, thus .k = 6. For the considered scenario, the following quantization systems are evaluated in Fig. 5.9 in terms of the achievable MSE versus the number of bits (Salamatian et al. 2019a): • The quadratic task-based quantization system presented in Sect. 5.3.3 with .p = 6 identical quantizers

132

D. Malak et al.

10

1

100

10-1

10-2

10-3

10

20

30

40

50

60

Fig. 5.9 Distortions of empirical covariance recovery versus the number of bits (Salamatian et al. 2019a)

• The linear quantization system described in Sect. 5.3.2, which linearly combines 1 T n T T T T .x, quantizing the . × 1 vector . [v + v , v + v ] , and computes the empirical 2 3 4 2 1 2 covariance at the output of the ADC • A task-ignorant system that quantizes .x and computes the empirical covariance at the output of the ADC • A system that recovers .s in analog and sets .sˆ to be the output of the ADC. Observing Fig. 5.9, we note that the quadratic task-based quantizer achieves the best MSE performance. While quantizing .s directly results in notable quantization errors when operating with a small number of bits, due to the need to set the dynamic range to a relatively large value resulting in coarse quantization, the distortions of the task-ignorant quantization and the linear combining one are comparatively high even with large quantization bits.

5.3.5 Conclusions In this section, we introduced hardware-limited task-based quantization, including the general system model, the specific system design for linear and quadratic tasks, followed by numerical examples. From the numerical studies, we see that the taskbased quantization system significantly outperforms its task-ignorant counterpart when the total number of quantization bits is fixed. This is achieved by introducing an analog combiner that greatly reduces the dimensionality of the input to the quantizers so that more bits can be assigned to each channel to obtain a more precise resolution. By exploiting the prior knowledge of the task, the hardwarelimited task-based quantization system approaches satisfying performance with fewer bits compared with conventional schemes that use high-resolution quantizers.

5 Hardware-Limited Task-Based Quantization in Systems

133

This implies that the task-based quantization scheme can be a promising solution to the high power consumption and storage problems encountered by conventional quantization systems. Therefore, applying the task-based quantization scheme in practical scenarios is of great interest, especially considering the fact that in many applications the analog signals are acquired not with the goal of being reconstructed, but for a specific task. For instance, in Shlezinger et al. (2019b), task-based quantization is applied to massive MIMO communications for estimation of the underlying channel from the high-dimensional received signals. The results demonstrate that the task-based system that uses low-resolution scalar ADC approaches the optimal channel estimation performance. In Xi et al. (2021), the authors consider target identification in radar; here the task-based quantization system operating with a bit budget equivalent to one bit per sample achieves target estimation accuracy comparable to that of costly MIMO radars operating with unlimited resolution ADCs. The task-based quantization has also been applied to graph signal processing (Li et al. 2022) and joint radar communications (Ma et al. 2021).

5.4 Architectures and Hardware Considerations Confidential Until March 1st, 2022—Please Do Not Share—Currently Under Process The wide use of multiple-input, multiple-output (MIMO) communication systems in 5G and beyond is due to the improved data capacity and coverage offered by the MIMO systems (Larsson et al. 2014; Agiwal et al. 2016; Shlezinger and Eldar 2018). However, traditionally, MIMO hardware implementations encounter a performance versus power consumption tradeoff (Ngo et al. 2013; Gesbert et al. 2003; Rusek et al. 2013; Shlezinger et al. 2019a). MIMO receivers typically utilize multiple signal acquisition chains consisting of a radio-frequency (RF) front end for low-noise amplification and RF-to-baseband downconversion followed by sampling and quantization using a uniform scalar analog-to-digital converter (ADC) for digital spatial-signal processing (Eldar 2015; Walden 1999). Spatialsignal processing in MIMO systems is performed by applying brute-force data acquisition using high-resolution quantization and Nyquist sampling rates. The brute-force data acquisition in MIMO and massive-MIMO systems requires high power consumption and increased hardware design complexity. Reducing the number of RF front-end chains and ADCs compared to the number of antenna elements via the use of analog combiners is commonly utilized for reducing the power consumption and hardware design complexity of hybrid analog/digital beamforming (HBF) receivers (Méndez-Rial et al. 2016; Ioushua and Eldar 2019). With the help of these analog combiners, MIMO receivers can achieve dimensionality reduction without degrading the directed beamforming capabilities via, e.g., holographic schemes (Huang et al. 2020). Recent MIMO architectures (Soer et al. 2011, 2017; Krishnaswamy and Zhang 2016; Golabighezelahmad et al.

134

D. Malak et al.

2020) also use these HBF techniques for improving the spatial interferer rejection performance in the analog domain without RF chain reduction. However, analog combiners typically consist of power-hungry active hardware components. Using low-resolution ADCs (Mo et al. 2017; Roth et al. 2018; Choi et al. 2018; Li et al. 2017b) reduces the power consumption and simplifies the hardware design of MIMO receivers at the expense of distortion due to coarse quantization, hence degradation in signal recovery performance. A task-specific (task-based) quantization for accurate signal recovery using lowquantization-rate ADCs has recently been proposed in theory (Shlezinger et al. 2019c; Neuhaus et al. 2021; Shlezinger and Eldar 2021; Salamatian et al. 2019b). A MIMO receiver front end combines the RF signals in the analog domain by taking the desired task and low-quantization-rate ADCs into consideration for taskspecific quantization. The underlying idea demonstrated in task-specific MIMO communications (Shlezinger et al. 2019b; Wang et al. 2021; Shlezinger et al. 2020a) and radar systems (Xi et al. 2021) is to ensure that the quantization distortion does not have a significant impact on the low-dimensional task information recovery in the digital domain. This section presents a power-efficient hardware algorithm co-design methodology for task-specific MIMO hardware systems (Zirtiloglu et al. 2022). The task is to achieve optimal recovery of desired signals while suppressing spatial interferers. Despite the numerically and theoretically evaluated performance gains of taskspecific quantization, designing these task-specific MIMO receivers requires an analog pre-processing front end, i.e., analog combiner, which is typically costly and power-hungry. In this section, we only consider analog combiners implemented as vector modulators (VMs) (Ellinger et al. 2010) for consistency in our analysis. To reduce the power consumption and hardware complexity of task-specific MIMO receivers, we exploit sparsity and discrete quantization of the VMs forming the analog combiner (Zirtiloglu et al. 2022). We present a task-specific algorithm codesigned with the receiver architecture under these hardware power constraints for accurate task recovery. By introducing a model for an end-to-end system evaluation, we compare the task-specific system performance with the task-agnostic MIMO systems for signal recovery accuracy and directed beam patterns through numerical simulations (Zirtiloglu et al. 2022). Finally, we provide power consumption estimates for the task-specific and task-agnostic systems derived from the measured power consumption of the state-of-the-art (SOA) integrated hardware implementations. We show that a task-specific design using a significantly reduced quantization rate achieves accurate signal recovery comparable to the performance of the fully digital MIMO receivers using high-resolution analog-to-digital converters (ADCs) (Zirtiloglu et al. 2022). The task-specific MIMO receiver reduces the power consumption by at least 58% compared to the task-agnostic fully digital MIMO and hybrid analog/digital beamforming (HBF) receivers while outperforming these taskagnostic architectures operating under similar bit constraints (Zirtiloglu et al. 2022). The rest of the chapter is organized as follows: Sect. 5.4.1 reviews the system model. The task-specific HBF receiver design algorithm with algorithmic design decisions is discussed in Sect. 5.4.2. Section 5.4.3 presents the receiver hardware

5 Hardware-Limited Task-Based Quantization in Systems

135

architecture with a model-based performance evaluation supported by numerical simulations and power consumption estimates. This section is followed by conclusions and future directions in Sect. 5.4.4.

5.4.1 System Model Here, we introduce a system model in detail for a task-specific hybrid MIMO embedding beamforming.

5.4.1.1

Task-Specific Hybrid MIMO with Embedded Beamforming

We consider a hybrid MIMO receiver with N antennas and P RF hardware chains and ADCs (Zirtiloglu et al. 2022). Let .x = [x1 , . . . , xN ]T be the input signal observed at the N antenna elements. This received input signal is initially processed in analog, yielding the vector .z = [z1 , . . . , zP ]T , which is provided to the ADCs for sampling and quantization. Given .A is the .P × N analog combiner matrix, the output vector .z is z = Ax.

(5.17)

.

The vector .z is converted into a digital signal using P identical uniform ADCs, each with b levels, i.e., the total number of bits used is .P log2 b. The vector processed in digital is .Qb (z), where .Qb (·) is the element-wise uniform quantization operator with b levels. The received vector .x is composed of a set of K desired signals, denoted by .s1 , . . . , sK , received from sources at relative angles .θ1 , . . . , θK , respectively. In addition, the received vector also consists of M interferer components .v1 , . . . , vM received from sources at relative angles .φ1 , . . . , φM , respectively. We assume all sources are narrowband and placed at the far field, and hence, the received signal with noise is given by x=

K 

.

k=1

sk a(θk ) +

M 

vm a(φm ) + w.

(5.18)

m=1

In (5.18), .w is additive white Gaussian noise with variance .σw2 , and .a(θ ) is the .N ×1 steering weights vector. Steering weights vector entries are given by d

[a(θ )]n = e−j 2π n λ sin(θ) ,

.

(5.19)

where d is the element spacing and .λ is the wavelength. We illustrate this system in Fig. 5.10 (Zirtiloglu et al. 2022).

136

D. Malak et al.

Inputs

V1

Bit-Constrained MIMO Receiver

X1

Analog Combiner

X2 ϕ1

S1

ANxP

Qb(z)= Qb(Ax)

ADC I1 ADC Q1

A1,1 ... A1,P A2,1 ... A2,P

θ1 VM

z = Ax

ϕM θK

XN-1

...

...

...

ADC IP

...

...

...

ADC QP

...

...

...

AN,1 ... AN,P

XN

Ŝ= Digital BQb(Ax) Filter

B

TaskSpecific Recovery

MSE

Control Angles of Arrival & Power Levels

SK

Fig. 5.10 Task-specific hybrid MIMO receiver system with embedded beamforming and lowquantization-rate ADCs (Zirtiloglu et al. 2022)

We can represent (5.18) as x = Aθ s + Aφ v + w,

.

(5.20)

N ×K and .A N ×M such that by defining the steering matrices .Aθ ∈ C φ ∈ C   .[Aθ ]n,k = [a(θk )]n and . Aφ = [a(φm )]n , as well as .s = [s1 , . . . , sK ]T n,m T and .v = [v1 , . . . , vM ] . Using (5.20), the second-order statistical moment of the H 2 observed .x is .C x = Aθ C s AH θ + Aφ C v Aφ + σw I N , while its correlation with the task of interest .s is given by .C sx = C s AH θ . Here, .C s and .C v indicate the covariance matrices of .s and .v, respectively. We utilize these statistical measures for the taskspecific recovery.

5.4.1.2

System Problem Formulation

The underlying idea is to tune the reconfigurable analog combiner .A of the HBF MIMO receiver using the signal model in (5.20). We jointly optimize the design for multiple tasks simultaneously to achieve accurate signal recovery in a powerefficient manner while suppressing spatial interferers, as discussed in detail below. Accurate Signal Recovery The main system task is to recover the desired signal s from the digital representation .Qb (z) obtained using low-quantization-rate ADCs. We use the mean-squared error (MSE) as our design metric for the task-specific signal recovery. The MSE is given by

.

  MSE(A) := E s − E{s|Qb (Ax)}2 .

.

(5.21)

5 Hardware-Limited Task-Based Quantization in Systems

137

Spatial Interferer Suppression The first task performance metric, MSE, focuses only on the ability to recover .s accurately. In practical scenarios, we prefer rejecting interferers in the analog domain. Otherwise, these interferers lead to receiver desensitization and increased ADC dynamic range requirement (Krishnaswamy and Zhang 2016). Analog combiners should suppress the spatial interferers while accurately recovering the desired signals. Given the contribution of .v on the analog combiner output .z is defined as .AAφ , we penalize interferer rejection using the max norm of .AAφ , i.e.,   IntRej(A) := AAφ max = max |[AAφ ]i,j | .

.

i,j

(5.22)

Power Consumption The power consumption of MIMO receiver depends on the individual power consumption of hardware components, including RF amplifiers, RF mixers, low-pass filters, local oscillator generator (LO Gen), and ADCs (Zirtiloglu et al. 2022). Since we use low-resolution quantizers in the task-specific quantization schemes, the ADC power consumption will be reduced significantly, which is approximately proportional to the number of levels b (Walden 1999). In Zirtiloglu et al. (2022), we consider the design of the analog combiner .A implemented using vector modulators (VMs) in the RF front end. The matrix entries of .A cannot take any value in .C for VMs implementation in practice and are constrained to a discrete set .A ⊂ C, including .0 ∈ A, i.e., when VM is deactivated for sparsity exploitation. Therefore, the analog combiner power consumption has two main contributors: how many different values can the matrix entries take, i.e., .|A|, and which of those matrix entries are active, namely, the sparsity level of .A (Zirtiloglu et al. 2022). As demonstrated in Zirtiloglu et al. (2022), to reduce the analog combiner power consumption, a coarse and sparse .A should be preferred.

5.4.2 Hardware-Constrained Task-Specific Acquisition In Zirtiloglu et al. (2022), we presented a design algorithm for the HBF MIMO receiver. We consider a scenario where the number of ADCs, the ADC resolution (b), and mapping of the VMs to a discrete set (.A) are constrained by the hardware design complexity and power consumption. Therefore, we optimize .A according to these hardware constraints. For this scenario, we discuss the optimization of the first task: signal recovery in Sect. 5.4.2.1. Furthermore, we present the joint optimization of the second task: interferer suppression with additional hardware power reduction techniques in Sect. 5.4.2.2. We summarize the system design choices in Sect. 5.4.2.3.

138

5.4.2.1

D. Malak et al.

First Task: Signal Recovery

Obtaining .A that minimizes (5.21) is a special case of the task-specific (task-based) quantization setup studied in Shlezinger et al. (2019c). As shown in Shlezinger et al. (2019c), (5.21) minimization is likely analytically intractable. However, it is possible to achieve accurate signal recovery by using the non-subtractive dithered quantization model. In the framework (Shlezinger et al. 2019c), recovering the linear minimal MSE estimate of .s is the objective. The digital processing output is an estimate following the form .sˆ = BQb (Ax) for some .B ∈ CK×P (Zirtiloglu et al. 2022). −1/2 We define .  C sx C x , and let .{λ,i } be its singular values arranged in descending order to formulate the MSE objective for complying with the above  η2 −1 , where .η is a coefficient considerations. Furthermore, we set .κ  η2 1 − 3b 2 tuned for ensuring negligible overloading probability of the ADCs (set here to .η = 3) (Zirtiloglu et al. 2022). For complex signal processing, we re-formulate the MSE in (5.21) as indicated in the below lemma (adapted from (Shlezinger et al. 2019c, Lem. 1)): Lemma 1 Under the assumption that the ADCs is modeled as non-subtractive dithered quantizers with vanishing overloading probability, the MSE objective (5.21) becomes   H H AC x AH .MSE(A) = Tr C x  −C x A 2κ · Tr(AC x AH ) IP + 3b2 · P

−1  H AC x  .

This MSE is obtained by setting the digital filter to be −1  2κ · Tr(AC x AH ) H H AC x A + IP . .B (A) = C x A 3b2 · P

(5.23)

(5.24)

Lemma 1 holds under the assumption of using non-subtractive dithered quantizers. It is also approximately valid when these assumptions are not satisfied for various input signals (Shlezinger et al. 2019c). As presented in Shlezinger et al. (2019c), the MSE objective in (5.23) is convex. In Zirtiloglu et al. (2022), we exploited this convexity for introducing additional design considerations for .A as detailed in Sect. 5.4.1.2.

5.4.2.2

Second Task: Interferer Mitigation

The objective (5.23) acknowledges a closed-form minimizer, as shown in (Shlezinger et al. 2019c, Thm. 1). This design only achieves the main task

5 Hardware-Limited Task-Based Quantization in Systems

139

of signal recovery but does not impose any structure on .A. To introduce the interference rejection requirement while reducing the power consumption of the analog combiner hardware, we formulated our design objective in Zirtiloglu et al. (2022) as L(A) = MSE(A) + γI IntRej(A) + γS A1,1 .

.

(5.25)

In (5.25), .·1,1 is the entry-wise .1 norm operator, while .γI , γS ≥ 0 are regularization coefficients, simultaneously achieving accurate signal recovery by minimizing MSE while rejecting spatial interferers, and imposing an analog combiner sparsity level in the overall loss measure .L(A) (Zirtiloglu et al. 2022). The regularization term for suppressing interferers defined in (5.22) is convex (Boyd and Vandenberghe 2004, Ch. 3.2). In addition, the sparsity of .A is exploited while preserving the convexity via the entry-wise .1 norm (Zirtiloglu et al. 2022). The final optimization problem is formulated as the following by taking into account the power-efficient hardware implementation of .A using RF VMs. Ao = arg min L(A).

.

(5.26)

A∈AP ×N

Since the optimization problem (5.21) is formulated over a discrete (i.e., nonconvex) search space, it is very difficult to obtain the analog combiner that optimizes the objective function. Given that the objective .L(A) is convex, we utilize discrete optimization tools to recover an optimal design for .A as demonstrated in Zirtiloglu et al. (2022).

5.4.2.3

Analog Combiner Algorithm for Joint Task-Specific Optimization

The optimization problem given in (5.26) aims to minimize a convex objective over a discrete set. In Zirtiloglu et al. (2022), we achieved this goal by using a projected convex optimization. Our design strategy in Zirtiloglu et al. (2022) is to utilize .kmax rounds of a convex optimizer for minimizing .L(A) over .CP ×N , while periodically projecting onto the discrete .A. In Zirtiloglu et al. (2022), we define the .OL (·) as the iterative optimizer with a loss measure .L and use proximal gradient descent ˜ := MSE(A) + γI IntRej(A) algorithm with a step-size .μ > 0 while treating .L(A) as the task term and .A1,1 as the prior, while maintaining the analog combiner sparsity. This iterative optimizer is defined as the following (Foucart and Rauhut 2013, Ch. 3): 2   ˜ − A + μ∇A L(A) ˜ ˜ 1,1 + 1  OL (A) = arg min γS A  A P ×N 2,2 2 ˜ A∈C

.

= TγS {A − μ∇A (MSE(A) + γI IntRej(A))} ,

(5.27)

140

D. Malak et al.

where .T is the element-wise complex soft-thresholding operator, given by .Tλ (z) := ej arg(z) max(|z| − λ, 0). The intermediate .A(k) is projected to consider the analog combiner matrix constrained to a discrete set due to the VMs-based hardware implementation via the element-wise projection operator .PA (z) := arg mina∈A a − z2 for every .kproj iterations. The proposed algorithm in Zirtiloglu et al. (2022) is given in Algorithm 1. Data: Fix A(0) for k = 1, 2, . . . , kmax do   Update A(k) ← OL A(k−1) if mod (k, kproj ) = 0 then Project via A(k) ← PA (A(k) ) end end Output: Analog combiner A(kmax ) .

Algorithm 1: Analog combiner setting (Zirtiloglu et al. 2022)

The essential hyperparameters of Algorithm 1 are the regularization coefficients γI , γS , the iteration limits .kmax , kproj , and the initial setting of .A(0) (Zirtiloglu et al. 2022). These hyperparameters are combined with the individual hyperparameters of the convex optimizer .OL (·). For the study in Zirtiloglu et al. (2022), we set the digital filter via (5.24) and use .A(0) =  as the initial estimate for the case when the number of complex ADCs is equal to the number of desired signals, i.e., .P = K.

.

5.4.3 Task-Specific Hybrid MIMO System Evaluation This section discusses the hardware architecture, model-based simulation setup, and recovery performance of the task-specific hybrid MIMO system. Furthermore, we show the angular-dependent beam patterns of the analog combiner front end and provide power consumption estimates for task-specific and task-agnostic receivers.

5.4.3.1

Hardware Model

To assess the performance of the task-specific MIMO system, we created a hardware model in Zirtiloglu et al. (2022). The hardware system in Zirtiloglu et al. (2022) includes an RF analog-combiner front end and digital signal processing back end for task-specific recovery similar to the system representation shown in Fig. 5.10. The RF analog-combiner front end is modeled as reconfigurable low-noise VMs. Each of these vector modulators represents a specific matrix .A entry. For an .N × P front end, a total of .N × P VMs are deployed. The system operates under prior knowledge of the angle of arrivals and signals power. The coefficients of

5 Hardware-Limited Task-Based Quantization in Systems

141

the analog-combiner matrix .A are calculated using the Algorithm 1 in the digital signal processing back end and used for VMs’ configuration. The N-element input observations .x = [x1 , . . . , xN ]T are provided to the analog combiner. These input signals are downconverted from RF to a low baseband frequency and combined in the analog domain. The P -combined signals are then sampled and quantized by the low-quantization-rate ADCs in both in- and quadrature-phase hardware branches (I .& Q). The digital filter .B, calculated using (5.24), is applied in the digital back end. The task-specific signal recovery is performed and characterized by the MSE performance metric for recovery accuracy. In Zirtiloglu et al. (2022), we considered an .8 × 2 hybrid MIMO system for the performance comparison and simulations, with .K = 2 desired signals at angles π π .θ1 = 8 , θ2 = − 4 with variances .1.5 and .0.5, respectively. Furthermore, we π assumed .M = 2 unwanted interferers at angles .φ1 = − 18 , φ2 = π3 with variances 5 for both sources in the model. The spatial interferers were set to a significantly stronger power level than the power level of the desired signals. The noise level was set to .σw2 = 1. In Zirtiloglu et al. (2022), we also provided the system performance evaluation without any quantization. We created a model of a taskagnostic .8 × 2 conventional hybrid MIMO receiver with embedded beamforming in analog and recovery in digital and a task-agnostic .8 × 8 fully digital MIMO receiver recovering the data solely in the digital domain, as performance benchmark systems in Zirtiloglu et al. (2022).

5.4.3.2

Mean-Squared Error Performance

In Zirtiloglu et al. (2022), to evaluate the signal recovery MSE performance, achieved using Algorithm 1, we varied the sparsity level of analog combiner matrix .A and VM resolution. The mean-squared error (MSE) performance for .s recovery is shown in Fig. 5.11 against the overall number of bits, i.e., .P log2 b on the x axis (Zirtiloglu et al. 2022). The numerical simulation results are shown for sparsity levels of 0 and 25.% with non-quantized, continuous (Cont.) matrix .A, and for 25.% sparse .A with a low-resolution VMs, e.g., 4-bit resolution for the VMs (Zirtiloglu et al. 2022). As illustrated in Fig. 5.11, a task-specific hybrid MIMO receiver, utilizing low-quantization-rate ADCs and using Algorithm 1, provides a similar MSE performance to the performance achieved without any quantization. This MSE performance is obtained for the task-specific hybrid MIMO receiver using low-resolution quantized VMs, e.g., .24 = 16 different settings for 4 bits, and deactivating .25% of VMs for sparsity to reduce hardware power consumption. On the other hand, the task-agnostic fully digital MIMO receiver model performs substantially worse in terms of recovery accuracy for a comparable overall number of ADC bits. It is possible to reduce the quantization rate by more than a factor of four as demonstrated in Zirtiloglu et al. (2022) to maintain the same target MSE performance floor shown in Fig. 5.11. The conventional hybrid MIMO receiver only beamforms toward the desired angles and does not account for the interferer suppression. The task-specific hybrid MIMO system demonstrates .1.5× lower MSE

142

D. Malak et al.

Fig. 5.11 MSE vs. the total number of quantization bits for recovering .K = 2 desired signals in the presence of .M = 2 spatial interferers (Zirtiloglu et al. 2022)

by using a total of 16 bits for 2 outputs consisting of I .& Q hardware branches, i.e., 4 bits ADCs are utilized for each hardware branch, compared to a conventional hybrid MIMO receiver with the same quantization rate as shown in Fig. 5.11 (Zirtiloglu et al. 2022).

5.4.3.3

Task-Specific Beamforming

Algorithm 1 also optimizes the analog combiner for the spatial interferer rejection in the analog domain, jointly with the main system task of accurate desired signal recovery, .s, in the digital domain. An array factor (AF) is a measure of MIMO receiver gain as a function of signal angular direction and defined by .AF (θ ) = N A ej π i sin(θ) , where .Ai is a specific complex gain coefficient of the input i i=1 signal, and N is the number of antennas (Soer et al. 2011). There are P independent beams directed toward a specific angle for an .N × P MIMO receiver. In Zirtiloglu et al. (2022), we modeled and computed the array factor at the .P = 2 analog outputs of the proposed system. These task-specific array factors are illustrated in Fig. 5.12 with the array factors for a task-agnostic conventional hybrid MIMO receiver with an analog combiner beamforming toward .θ1 for the first output (Fig. 5.12a) and .θ2 for the second output (Fig. 5.12b). The beam patterns for a task-specific hybrid MIMO receiver using Algorithm 1 are directed toward both of the desired angles .θ1 and .θ2 at both of the outputs, hence forming a linear combination of the desired signals at the first and second outputs, while rejecting the interferers at angles .φ1 , .φ2 by .≥ 36dB. The task-specific analog combiner is not solely optimized for beamforming but designed for ensuring accurate recovery from low-quantization-rate observations. Therefore, the taskspecific analog combiner provides a lower AF gain for the desired signals compared

5 Hardware-Limited Task-Based Quantization in Systems

143

Fig. 5.12 Array factor plots versus angle of arrival for .P = 2 outputs for a task-specific hybrid MIMO receiver and a task-agnostic conventional hybrid MIMO receiver (dB scale, 20.log) (Zirtiloglu et al. 2022). (a) Array factor plot for the first output for a task-specific hybrid MIMO receiver (red curve) and a task-agnostic conventional hybrid MIMO receiver (green curve) (Zirtiloglu et al. 2022). (b) Array factor plot for the second output for a task-specific hybrid MIMO receiver (blue curve) and a task-agnostic conventional hybrid MIMO receiver (green curve) (Zirtiloglu et al. 2022)

to conventional beamforming as shown in Fig. 5.12 (Zirtiloglu et al. 2022). This lower AF gain does not degrade the task-specific recovery accuracy as illustrated in Fig. 5.11 (Zirtiloglu et al. 2022).

5.4.3.4

Power Consumption Model

Power consumption of the task-specific HBF MIMO receiver and the task-agnostic benchmark systems are estimated in Zirtiloglu et al. (2022) based on the measured power consumption of each hardware component reported in the state-of-the-art integrated hardware designs (Soer et al. 2017; Joram et al. 2009; Amer et al. 2007; Kibaroglu and Rebeiz 2017; Méndez-Rial et al. 2016; Ho and Lee 2012; Lee and Sodini 2008). Power consumption of an .N × N fully digital MIMO receiver is given by PFD = N · PLNA + N · PMIX + 2 · N · PBB + 2 · N · PADC .

.

(5.28)

Here, .PLNA represents the power consumption of a low-noise amplifier, .PMIX denotes the power consumption of a mixer, and .PBB and .PADC are baseband amplifier and ADC power consumption, respectively. Baseband amplifier and ADC power consumption is reported for both in-phase and quadrature-phase paths (I .& Q paths). The power consumed by an .N × P hybrid MIMO receiver is estimated by PHYB = γSP · N · P · PVM + P · PMIX + 2 · P · PBB + 2 · P · PADC .

.

(5.29)

144

D. Malak et al.

Table 5.1 Estimated power consumption for a task-specific hybrid MIMO receiver compared to the estimated power consumption of task-agnostic receivers, including a fully digital MIMO receiver and a conventional hybrid MIMO receiver (Zirtiloglu et al. 2022). Bold signifies the reduced power consumption Hardware component/system LNA/VM .PLNA/VM (1–5 GHz 8 bit/4 bit) (Soer et al. 2017; Joram et al. 2009) Mixer with LO Gen (1–5 GHz) .PMIX (Amer et al. 2007; Kibaroglu and Rebeiz 2017) Baseband amplifier .PBB (Méndez-Rial et al. 2016) ADC .PADC (100 MS/s 10 bit/4 bit) (Ho and Lee 2012; Ginsburg and Chandrakasan 2006) Fully digital MIMO receiver (.8 × 8) Conventional hybrid MIMO receiver (.8 × 2) Task-specific hybrid MIMO receiver (.8 × 2)

Power (mW) 20/10 15 5 10/0.5 520 410 172

PVM is the power consumed by a low-noise VM amplifier, and .γSP is the analogcombiner sparsity coefficient. .γSP = 1 represents a non-sparse .A, while .γSP = 0.75 corresponds to 25.% sparsity. We summarize the estimated power consumption of each hardware component and the total power consumption of task-specific and task-agnostic MIMO receivers in Table 5.1 (Zirtiloglu et al. 2022). The power consumption scaling for different quantization levels of the VMs is derived from Qian et al. (2019) when 8 bits represent high-resolution VMs and 4 bits represent low-resolution VMs. We utilize Walden FoM (Skrimponis et al. 2020; Lee and Sodini 2008) for the ADC power estimation. Our results in Zirtiloglu et al. (2022) show that the proposed power-saving techniques, exploiting 25.% sparsity, 4-bit VMs, and 4-bit ADCs, provide more than 58.% reduction in power compared to the task-agnostic MIMO architectures using high-resolution ADCs, high-power LNAs, and high-resolution VMs, while providing an MSE performance improvement observed from simulation results shown in Fig. 5.11 and the calculated spatial interferer rejection shown in Fig. 5.12.

.

5.4.4 Discussion and Conclusions This section discussed a power-efficient hybrid MIMO receiver design embedding beamforming and low-resolution ADCs using task-specific quantization. We presented a joint optimization framework for analog pre-processing and digital signal processing to achieve low-power operation. This joint optimization enabled optimal MSE performance at significantly lower power consumption through sparse and low-resolution analog-combiner design (Zirtiloglu et al. 2022). Furthermore, we demonstrated mitigation of undesired spatial interferers. The task-specific hybrid MIMO receiver outperforms the task-agnostic fully digital MIMO receivers as supported by our numerical simulation results (Zirtiloglu et al. 2022).

5 Hardware-Limited Task-Based Quantization in Systems

145

The analog-combiner tuning discussed in this section requires prior knowledge of the angle of arrivals and power levels. We utilized this knowledge for creating the correlation matrices .C x and .C sx for the algorithm presented in Zirtiloglu et al. (2022). This information can be estimated or provided by a spatial sensor in practice (Zirtiloglu et al. 2022). An open research question is to assess the impact of hardware nonidealities on the task-specific MIMO system performance, including frequency-independent and -dependent gain and phase imbalance of vector modulators. We can design co-optimized algorithms and hardware to cope with these hardware nonidealities for the task-specific systems. Lastly, in our simulations, we manually set the hyperparameters for Algorithm 1. Given the recent success of deep learning tools in enabling rapid optimization (Shlezinger et al. 2020b), we can design data-driven hyperparameter setting, e.g., via the learn-tooptimize framework (Chen et al. 2021) or via deep unfolding (Monga et al. 2021). As recently shown in Khobahi et al. (2021), deep unfolding is particularly suitable for achieving accurate and fast optimization of convex objectives over discrete sets similar to (5.26).

5.5 Networks with Task-Based Quantization The focus of the previous section is on the design MIMO architectures for recovering tasks via taking into account hardware constraints and suppressing the interference. In this section, we consider an abstraction for next-generation communication networks consisting of a number of nodes where tasks travel among the nodes following a fixed routing matrix, and nodes have variable computation and service rates. This scheme can also encompass MIMO for sending and receiving multiple signals. To that end, we combine queueing theory, optimization, and entropy coding to capture the critical aspects of a network, such as congestion and latency, for task distribution. For instance, given a set of functions to be computed, our goal is to find the data and task distribution across the nodes, as well as probabilistic bandwidth reservation between each node in order to minimize a measure of cost, such as latency or job completion time. Taking a queueing theory approach adds interesting angels to the problem. This is because the performance of an interacting queue model is highly non-linear in terms of the load distribution. On one hand, the larger the number of parallel nodes, it is possible to significantly reduce the waiting time and the queue length. On the other hand, adding nodes incurs additional cost to serve the requests.

5.5.1 Leveraging Little’s Law for Distributed Computing We provide a utility-based approach for general functions where we connect function computation problem with queueing theory’s Little’s law. Consider a

146

D. Malak et al.

Fig. 5.13 Illustration of computational flow j (or task j j ) at node .v ∈ V where .λv is the total arrival rate of computational flow j at node v that incorporates the j original arrivals .βv and the arrivals routed from .u ∈ V . The total generated rate of j j at v is .γf (λv ), and .u, v ∈ V represent the nodes where arrivals routed from/to

general network topology where nodes represent compute resources. We abstract away the switches in the topology and connect two nodes with an edge if there is a path between the two nodes. The weight on each edge will capture network properties such as path length, congestion on the path, capacity, latency, etc. Figure 5.13 illustrates the computational flow perspective. Little’s law states that the long-term average number L of packets in a stationary system is equal to the long-term average effective arrival rate .λ multiplied by the average time W that a packet spends in the system (Leon-Garcia 2017). More formally, it can be expressed as .L = λW . The result applies to any system that is stable and non-preemptive, and the relationship does not depend on the distribution of the arrival process, the service distribution, and the service order (Kleinrock 1975). In our setting, the average time a packet spends in the system is given by the addition of the total time required by computation followed by the total time required by communications. The goal of the Function Manager module is to come up with the best allocation. As a result, it should solve an optimization problem of the form  MinCost :min C = T askCompletionT ime(v, j ) j {ρv } v∈V j ∈J . (5.30) s.t. ρvj < 1,

∀j ∈ J, v ∈ V ,

where .T askCompletionT ime(v, j )

= ComputeT ime(v, j ) + CommunicationT ime(v, j )

captures the total completion time of computational flow (or task) j on node v, ComputeT ime(v, j ) captures the computation time, and .CommunicationT ime (v, j ) captures the communication time of task j on node v. Both .ComputeT ime (v, j ) and .CommunicationT ime(v, j ) are positive delay cost functions that are j non-decreasing in flow or load .ρv , .j ∈ J, .v ∈ V .

.

5 Hardware-Limited Task-Based Quantization in Systems

147

Note that the proposed scheme assumes there are two queues at each node (or server), a computation queue followed by a communication queue that forwards the generated data to the other nodes. The computation delay at node .v ∈ V for function of type .j ∈ J is ComputeT ime(v, j ) =

.

1 j

λv

T askT imeComplexity(Mvj ),

where T askT imeComplexity models the time complexity of computation, i.e., the total time needed to process all the incoming packets and generate the desired j function outcomes, and .Mv is the long-term average number of packets for function of type .j ∈ J in v waiting for communications service. Arrival rate of type .j ∈ J j j flow at node v is .λv . Service rate of type j flow at node v is given by .μv . As an example, we consider three different function categories and the time complexity of these: Search, MapReduce, and Classification. The Search function tries to locate an element in a sorted array; hence, a search algorithm can run in logarithmic time. For the MapReduce function (or linear reduce function), since the reduce functions of interest are linear, the algorithm runs in linear time. And for the Classification function, we can consider the set of all decision problems that have exponential runtime, which is of high complexity. The time complexity, i.e., the order of the count of operations, of these functions satisfies

j .T askT imeComplexity(Mv )

=

⎧ j ⎪ ⎪ ⎨O(log(Mv )),

Search,

j O(Mv ),

MapReduce, ⎪ ⎪ ⎩O(exp(M j )), Classification. v

(5.31)

To model the communications time of each server, we use an M/M/1 queue. The communications delay at node .v ∈ V for function of type .j ∈ J is CommunicationT ime(v, j ) =

.

1 j μv

j

− γf (λv )

,

(5.32)

j

where .γf (λv ) characterizes the amount of computation flow rate generated by node v as a result of computing .fj , i.e., the processing (or surjection) factor of a node v. Hence, the second term on the right-hand side captures the waiting time, i.e., the queueing plus service time of a packet. By applying Little’s law, we can solve the MinCost formulation in (5.30). This law is the key to capturing the relationship between ComputeT ime and j CommunicationT ime. Hence, we expect the long-term average number .Lv of packets in node v for function of type .j ∈ J satisfies the following relation: NumberOf SubT asks(v, j ) = OutputRate(v, j )×T askCompletionT ime(v, j )

.

148

D. Malak et al.

Ljv = γf (λjv ) × Wvj ,

(5.33)

j

where we aim to infer the value of .γf (λv ) with Little’s law. This approach captures the joint behavior of the communications queue preceded by the computation queue. Note that in Eq. (5.30), we assume the time complexity of different classes, j i.e., .T askT imeComplexity(Mv ), is known. Observe that .ComputeT ime(v, j ) j j decreases in .γf (λv ), and .CommunicationT ime(v, j ) increases in .γf (λv ). Note j also that the values of .λv should be jointly optimized to minimize C.

5.5.2 Probabilistic Bandwidth Reservation The previous subsection described a utility optimization approach to divide computation into tasks and assign them to servers. The function manager also determines the probabilistic bandwidth reservations, .Pu,v , of the links between servers. To do so, first we consider the behavior of each node in isolation, which is allowed when the network is quasi-reversible or in product form (Walrand 1983). For example, a Jackson network exhibits this behavior, and the total arrival rate of class j packets to server v can be related to probabilistic reservations via a Markov routing policy (Nelson 2013, Ch. 10.6.2): λjv = βvj +



.

j

γf (λu )Pu,v (j ),

(5.34)

u∈V

where .Pu,v (j ) is the probability that a class j packet that finishes service at node u j is routed to node v, and.βv is the arrival rate of class j packets to server v. The first term on the RHS denotes the original arrival rate of class j packets that are assigned to node v, and the second term on the RHS denotes the arrival rate of class j packets that are routed to node v after finishing service at other nodes .u ∈ V . Note that the j term .γf (λu ) denotes the total departure rate of class j packets from node u as a result of computation. Figure 5.13 illustrates these various aspects of computation and communication for a given worker node .v ∈ V . The min-cut denotes the total arrival rate of j computational flow (a.k.a. task) j and is given by .λv . This captures the rate of j original arrivals that is .βv , and the arrivals routed from any other node .u ∈ V in j j the network. If there is no .u ∈ V such that .Pu,v (j ) > 0, then .λv = βv . The cut j .γf (λv ) denotes the total generated rate (or processing factor) of computational flow j at node v. The processed flow can be routed to any .v ∈ V in the network if j .Pv,v (j ) > 0. If there is no such node, then .γf (λv ) departs the system. The advantage of having a Jackson type network is that the nodes can be considered in isolation (Nelson 2013). Each node only needs to know how much it needs to manage, which is less complicated than when nodes need the topological

5 Hardware-Limited Task-Based Quantization in Systems

149

information to manage individual computational flows. The routing policy in (5.34) assumes that packets do not change their class when routed from one node to another. We further plan to study the multi-class generalization because in general packets can change their class when routed (Nelson 2013). Given the cost functions .ComputeT ime(v, j ) and .CommunicationT ime(v, j ) and the probabilistic reservations .Pu,v , we can solve for the optimal values of j .γf (λv ), .v ∈ V , .j ∈ J that minimize the MinCost problem in Equation (5.30). However, this approach by itself does not say much about the relation between j j the load .ρv before computation and the load .γf (λv ) generated at server v as a result of computing the outcome of function outcome of type j , denoted by .fj (X), where X can be multi-variate. To do so, we develop an innovative perspective in Sect. 5.5.4, which maps the surjectivity of the different classes of functions to j the corresponding .γf (λv ) values. The techniques in this section along with the perspective in Sect. 5.5.4 will enable a system’s perspective that combines the scheduler with the abstraction for distributed computation.

5.5.3 Using Conflict Graphs in Distributed Computation As another example, we provide the connection between a recent paper for distributed machine learning called Matcha (Wang et al. 2019) and conflict graphs. The assignment of a schedule though matchings, as described in the Matcha paper, is an instance of using a conflict graph formulation. A common application of the principle in current networks is the use of bipartite matchings between input and output ports in switches (McKeown et al. 1999). Through a schedule that samples, whether in a deterministic or probabilistic fashion, in a weighted fashion across such matchings, full utilization can be achieved. Optimal flow in switches through the use of weighted scheduling over input–output matchings, which was examined in the paper “Achieving 100% throughput in an input-queued switch,” is in effect an instance of the classic Birkhoff–von Neumann theorem (Von Neumann 1953; Birkhoff 1946). The latter is itself an instance of the problem of finding a stable set polytope of a conflict graph. A conflict graph is a graphical representation of taboo concurrencies or conflicts. For example, in the case of a switch, only one output node may be active at a time, so a conflict graph representation captures the fact that an allowable state of the switch cannot have two input nodes linking to a same output node. Each allowable instantaneous state of a system is represented thus by a stable set, or a set of nodes where no two nodes share an edge on the conflict graph. Each allowable state of the system represents one state in a schedule. Consider the vertices of a polytope formed by the convex combination of these allowable states, or stable sets of the conflict graph. That polytope encompasses all the possible schedules. Finding that stable set is in general NP-hard, but, in certain cases, the problem is polynomial time. For example, the Birkhoff–von Neumann theorem, mentioned above, can be

150

D. Malak et al.

proved immediately by noting that the matching problem leads to a conflict graph that is perfect, in which case determining the stable set polytope is polynomial time. Other examples, such as conflict graphs that are claw-free (Köse and Médard 2017), also lend themselves to a polynomial time determination of the stable set polytope. Note that our discussion suggests intriguing possibilities for simple scheduling algorithms. Returning to the switch example, a simple greedy scheduling algorithm such as iSLIP (McKeown 1999) provides full utilization with a speed-up of at most two. This property is a direct consequence of the structure of the underlying conflict graph. Identifying scheduling schemes such as Matcha as special cases of operating on stable set polytopes strongly suggests that similar results are possible and, indeed, likely. The stable set polytope circumscribes the region of communication resources that are available for allocation over the network. The behavior of cost over that allowable region may depend heavily on load. Consider, for example, the usual model for congestion, or, equivalently, delay, over a link. Using the usual nomenclature of queuing theory, let .λ(i,j ) be the traffic on link .(i, j ), and let the bandwidth, or service rate, be .μ(i,j ) . The load is then defined in the usual fashion λ ) by .ρ(i,j ) = μ(i,j and must also be below 1 to ensure system stability. Through (i,j ) the use of such canonical models such as M/M/1 queues, cost depends on .ρ(i,j ) as . 1−ρ1(i,j ) . Considering the Matcha assignment in the above context, assigning matchings in a manner that takes degree into account is a heuristic form of load balancing. A more nuanced and principled approach would consider a cost metric, such as .(i,j ) 1−ρ1(i,j ) , over the possible set of .μ(i,j ) determined by the stable set polytope. Further refinements include different weights to reflect such aspects as differing costs of links.

5.5.4 An Information Theory Perspective to Distributed Computation This section aims to provide an understanding of the information theoretical concepts and how we can use them in distributed function computation. To understand the fundamental limits of distributed function computation, we start with a canonical example that includes a single node with computation capability. In this example, the node wants to compute the value of a function f applied to data X, denoted as .f (X). The total incoming flow rate of X to the node should be at least .Entropy(X) = H (X), which is the Shannon entropy—a measure of the minimal representation—of X measured in terms of bits required to represent data. The node wants to perform a computation on X and generate .f (X). The node then wants to communicate the function value .f (X). In this case, the total amount of generated flow rate should be at least .Entropy(f (X)) = H (f (X)) that is typically smaller than .Entropy(X) due to function’s surjectivity.

5 Hardware-Limited Task-Based Quantization in Systems

151

5.5.5 Computing General Functions in Large-Scale Topologies While many reduce functions of interest in the literature are linear, e.g., MapReduce (Li et al. 2018), some functions may exhibit concave or convex cost characteristics. How to distribute the tasks among multiple servers depends on the computation cost characteristics (can effectively be done in parallel or not). This is further related to how surjective the function is on its domain. The higher the surjectivity of the function is, the harder it becomes to compress it. The higher the compute rate, the less the outgoing flow .Entropy(f (X)) is, and this can provide significant gains in terms of communication cost. Therefore, as also illustrated by the above examples, it is clear that surjectivity of the function determines how much we can move along the communication cost curve. Reliable evaluation of a function means asymptotically lossless computation of a function as the blocklength goes to the infinity. In particular, this problem is referred to as Distributed Functional Compression and has been studied under various forms since the pioneering work of Slepian and Wolf in 1973 (Slepian and Wolf 1973). We now discuss the general distributed source coding problem in which the function is the same as the source data, and we ask the following question: do we have to transmit all data? For sake of presentation, we focus on the case of two random variables .X1 and .X2 , which are jointly distributed according to .PX1 ,X2 . We first consider the natural scenario where the function .f (X1 , X2 ) is the identity function, i.e., the case of distributed lossless compression. Source random variable .X1 can be asymptotically compressed up to the rate .Entropy(X1 |X2 ) that is smaller than .Entropy(X1 ) when .X2 is available at the receiver (Slepian and Wolf 1973). It intuitively makes sense that the side information at the receiver should help the compression of the sources. Given two statistically dependent i.i.d. finite alphabet sequences .X1 and .X2 , the Slepian–Wolf theorem gives a theoretical bound for the lossless coding rate for distributed coding of the two sources (Slepian and Wolf 1973), which is given by (5.8). We denote this rate region by .R and depict it in Fig. 5.14-(Left) by the inner bound .I. This result states that in order to recover a joint source .(X1 , X2 ) at a receiver, it is both necessary and sufficient to encode separately sources .X1 and .X2 at rates .(RX1 , RX2 ) ∈ R (Doshi et al. 2010). The encoding is done in a truly distributed way, i.e., no coordination is necessary between the encoders. Distributed coding can achieve arbitrarily small error probability for long sequences. Furthermore, the decomposition of .J ointEntropy(X1 , X2 ) = Entropy(X1 |X2 ) + Entropy(X2 ) implies the feasibility of time-sharing between sources. We next seek whether it is possible to generalize the Slepian–Wolf scheme to determine a bound for the coding rate when computing general functions .f (X1 , X2 ) different from the identity function. Challenges in function computation involve the function f on the source variables (or data) X itself as well as the correlations among X, due to the mapping from the sources to the destinations (determined by connectivity, hardware constraints, stragglers, and finite link capacities), the codebook design becomes challenging. Since the rate region of the distributed

152

D. Malak et al.

function computation problem depends on the function, designing achievable schemes for the optimal rate region for function computation (or compression) (for general functions, with/without correlations) remains an open problem (Gamal and Kim 2011, Ch. 17). We aim to develop a tractable approach for computing general functions in general topologies using tools from information theory and graph theory. We illustrate the characteristic graph and its relevance in compression through the following examples. Example 4 (Compression Below Entropy Rate) Consider that X is a random variable with uniform distribution over .{0, 1, 2, 3, 4}, and Z is a random variable and .f (X, Z) such that we have the graph .GX with 5 vertices, with each vertex has only two edges to two other vertices. This graph is a valid coloring using 3 colors. The vertices that have the same color yield the same function output. Let this coloring random variable be .cGX . In this example, .cGX = {c1 , c2 , c3 } with a distribution .P (c1 ) = P (c2 ) = 2/5 and .P (c3 ) = 1/5. This yields an .Entropy(cGX ) ≈ 1.52. Now, we encode a random variable .X2 , which can take 25 values 2 .{00, 01, . . . , 44}. To construct the characteristic graph for .X , i.e., the second 2 power graph of .GX (referred by .GX ), we connect two vertices if at least one of coordinates is connected in .GX . It can be shown that one can color .G2X by using 8 colors. Its entropy satisfies . 12 Entropy(cG2 ) ≈ 1.48 < Entropy(cGX ) ≈ 1.52 < X Entropy(X) ≈ 2.32. Example 4 demonstrates that if we assign colors to a sufficiently large power graph of .GX , we can compress source random variables more. In general, finding minimum entropy colorings of characteristic graphs is NP-hard. However, in some instances, it is possible to efficiently compute these colorings. In Feizi and Médard (2014), authors showed that sending colorings of sufficiently large power graphs of characteristic graphs followed by Slepian–Wolf compression leads to an achievable scheme under some conditions. An object of interest in the study of these fundamental limits is the characteristic graph of a function f , and in particular its coloring. In the characteristic graph, each vertex represents a possible different sample value, and two vertices are connected if they should be distinguished. Hence, this structure tells us about the fundamental limits of compression to fully represent a function.

5.5.6 Characterizing Communication Rates with Graph Entropy The graph entropy is the minimum rate at which a single source can be encoded so that a function of that source can be computed with zero distortion (Körner 1973; Alon and Orlitsky 1996; Witsenhausen 1976). The notion of graph entropy is useful in determining the rate region of the distributed functional compression problem. In Feizi and Médard (2014), authors have determined the rate region for a distributed

5 Hardware-Limited Task-Based Quantization in Systems

153

Fig. 5.14 (Left) Example rate region for the zero distortion distributed functional compression problem (Doshi et al. 2010). .S denotes the shaded region between .J ointEntropy(X1 , X2 ) curve (inner bound .I) and .J ointGraphEntropy(X1 , X2 ) (outer bound .O). Note that any point above .I is the Slepian–Wolf achievable rate region, and .O is characterized by how surjective the graph entropy is. The axes are color-labeled, and only the labels corresponding to .RX2 are provided. (Right) Example scenarios with achievable rates: rate region for (i) source compression, (ii) functional compression, (iii) distributed source compression with two transmitters and a receiver, and (iv) distributed functional compression with two transmitters and a receiver. Note that in (iv) the main benefit of y .J ointGraphEntropy(X1 , X2 ) is that it is less than the sum of the marginal graph entropy of source .X1 , i.e., .GraphEntropy(X1 ), and the conditional graph entropy of source .X2 given .X1 , i.e., .GraphEntropy(X2 |X1 )

functional compression problem with two sources and a receiver: R11 ≥ GraphEntropy(X1 |X2 ), R12 ≥ GraphEntropy(X2 |X1 ),

.

R11 + R12 ≥J ointGraphEntropy(X1 , X2 ),

(5.35)

where .J ointGraphEntropy(X1 , X2 ) is the joint graph entropy of the sources. In Fig. 5.14, we illustrate the inner bound .I determined by the Slepian–Wolf compression rate region versus the outer bound .O determined by the joint graph entropy of variables .X1 and .X2 in (5.35). In the graph, the region between two bounds, denoted by .S, determines the limits of the functional compression. The depth of this region indicates that there could be potentially a lot of benefit in exploiting the compressibility of the function to reduce communication. .SideI nf ormationGain(f ) captures the gain of side information provided by the other source under graph entropy. .J ointCompressionGain(f ) takes into account the function’s surjectivity and joint compressibility of both sources under graph entropy—unlike J ointEntropy that allows a linear splitting of the source entropies via the chain rule and does not take the function structure into account, J ointGraphEntropy can provide a better compression rate than J ointEntropy because function is compressed along with the sources and it does not satisfy the

154

D. Malak et al.

Fig. 5.15 A two-stage tree network (Feizi and Médard 2014)

chain rule (convex). Hence, we expect to have .J ointCompressionGain(f ) ≥ SideI nf ormationGain(f ) in the left figure. Example 5 (In-Network Computation (Sapio et al. 2019a; Ports and Nelson 2019)) To motivate the rate gains that can be obtained via in-network function computation, we next consider a simple tree network to compute a linear function. This aggregation method is used to scale distributed machine learning to accelerate parallel training (Sapio et al. 2019b). The example network that is illustrated in Fig. 5.15 contains 2 stages and has 4 source nodes .Xi , .i = 1, 2, 3, 4, each taking binary values. The receiver computes a parity check function .f (X1 , X2 , X3 , X4 ) = (X1 + X2 + X3 + X4 )mod 2. If intermediate nodes act like relays, i.e., they do not perform any computations, then the following set of rates is an achievable scheme: R2j ≥ 1, 1 ≤ j ≤ 4,

.

R1j ≥ 2, 1 ≤ j ≤ 2.

If intermediate nodes perform computation, then the set of rates .Rij ≥ 1 is achievable. In this scheme, intermediate nodes need to transmit one bit, which reduces the total communication rate by 2. This is achieved via the following encoding/decoding scheme (Feizi and Médard 2014). Source nodes send their coloring random variables that are equal to source random variables because their characteristic graphs are complete. Then, each intermediate node makes its own characteristic graph, and by using the received colors, picks a corresponding color for its own characteristic graph and sends that color. The receiver, by using the received colors of intermediate nodes’ characteristic graphs and a look-up table, can compute its desired function.

5.5.7 Conclusions We provided a novel perspective to function computation in networks. We introduced the notion of entropic surjectivity to assess the functional complexity. Extending Little’s law to computing, we derived regimes for which intermediate computations can provide significant savings. This approach provides a baseline toward understanding how to distribute computation and balance functional load

5 Hardware-Limited Task-Based Quantization in Systems

155

in networks. Future directions include devising coding techniques for in-network functional compression, by using compressed sensing and the compression theorem of Slepian and Wolf, employing the concepts of graph entropy, and exploiting function surjectivity. They also include more general network models beyond stationary and product form.

5.6 Summary In this chapter, we detailed several key frameworks that enable task-based representations in systems. In Sect. 5.2, we discussed an information-theoretically motivated approach to efficient distributed compression for deriving vector quantized functional representations for task-based systems. In Sect. 5.3, we detailed the theoretical principles for task-based quantization under hardware and bit budget constraints, and in Sect. 5.4, we focused on a MIMO-based communication architecture for realizing hardware-limited task-based quantization. Finally, in Sect. 5.5, we described how task-based quantization can be implemented in networks, via leveraging tools from graph coloring, queueing theory, and optimization, to capture the tradeoff space between communications and computation for various tasks. In the next generation of communication and computing systems and applications, with the growing complexity of tasks and demand and scarcity of resources, the system designer should pay utmost attention to efficient compression of signals and data. We believe that task-based quantization, which is a radically different approach to conventional quantization techniques, can be a promising solution to the exploding data, power, and storage demand in current and future systems and networking applications.

References Agiwal M, Roy A, Saxena N (2016) Next generation 5G wireless networks: A comprehensive survey 18(3):1617–1655 Alon N, Orlitsky A (1996) Source coding and graph entropies. IEEE Trans Inf Theory 42(5):1329– 39 Amer A, Hegazi E, Ragaie HF (2007) A 90-nm wideband merged CMOS LNA and mixer exploiting noise cancellation. IEEE J Solid State Circuits 42(2):323–328 Appuswamy R, Franceschetti M (2014) Computing linear functions by linear coding over networks. IEEE Trans Inf Theory 60(1):422–431 Bajcsy J, Mitran P (2001) Coding for the Slepian-Wolf problem with turbo codes. In: Proc., IEEE Globecom, San Antonio, TX, vol 2, pp 1400–1404 Basu S, Seo D, Varshney LR (2020) Hypergraph-based coding schemes for two source coding problems under maximal distortion. In: Proc., IEEE ISIT, Los Angeles, CA, pp 2426–2431 Berger T (1978) Multiterminal source coding. The Inf Theory Approach to Commun Birkhoff G (1946) Tres observaciones sobre el algebra lineal. Univ Nac Tucuman A 5:147–154 Boyd S, Vandenberghe L (2004) Convex optimization. Cambridge University Press

156

D. Malak et al.

Cardinal J, Fiorini S, Van Assche G (2004) On minimum entropy graph colorings. In: Proc., IEEE ISIT, Chicago, Illinois, p 43 Chen T, Chen X, Chen W, Heaton H, Liu J, Wang Z, Yin W (2021) Learning to optimize: A primer and a benchmark. Preprint. arXiv:210312828 Choi J, Mo J, Heath RW (2016) Near maximum-likelihood detector and channel estimator for uplink multiuser massive MIMO systems with one-bit ADCs. IEEE Trans Commun 64(5):2005–2018 Choi J, Sung J, Evans BL, Gatherer A (2018) Antenna selection for large-scale MIMO systems with low-resolution ADCs. In: 2018 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP), pp 3594–3598 Courtade TA, Wesel RD (2011) Multiterminal source coding with an entropy-based distortion measure. In: Proc., IEEE Int. Symp. Inf. Theory, Saint-Petersburg, Russia, pp 2040–2044 Cover T (1975) A proof of the data compression theorem of Slepian and Wolf for ergodic sources. IEEE Trans Inf Theory 21(2):226–228 Cover TM, Thomas JA (2012) Elements of information theory. Wiley Dattorro J (2005) Convex optimization & Euclidean distance geometry. Meboo Publishing Delgosha P, Anantharam V (2019) A notion of entropy for stochastic processes on marked rooted graphs. Preprint. arXiv:190800964 Doshi V, Shah D, Médard M, Effros M (2010) Functional compression through graph coloring. IEEE Trans Inf Theory 56 Eldar YC (2015) Sampling theory: beyond bandlimited systems. Cambridge Univ. Press, Cambridge, UK Ellinger F, Mayer U, Wickert M, Joram N, Wagner J, Eickhoff R, Santamaria I, Scheytt C, Kraemer R (2010) Integrated adjustable phase shifters. IEEE Microwave Mag 11(6):97–108 Fang J, Li H (2009) Hyperplane-based vector quantization for distributed estimation in wireless sensor networks. IEEE Trans Inf Theory 55(12):5682–5699 Feizi S, Médard M (2014) On network functional compression. IEEE Trans Inf Theory 60(9):5387–5401 Feng H, Effros M, Savari S (2004) Functional source coding for networks with receiver side information. In: Proc., IEEE Allerton Conf. Comm., Control and Comput., Monticello, IL, pp 1419–27 Foucart S, Rauhut H (2013) A mathematical introduction to compressive sensing. Springer Gács P, Körner J (1973) Common information is far less than mutual information. Probl Control Inf Theory 2(2):149–162 Gallager R (1988) Finding parity in a simple broadcast network. IEEE Trans Inf Theory 34(2):176– 180 Gamal AE, Kim YH (2011) Network information theory. Cambridge University Press Gesbert D, Shafi M, shan Shiu D, Smith P, Naguib A (2003) From theory to practice: an overview of MIMO space-time coded wireless systems 21(3):281–302 Ginsburg B, Chandrakasan A (2006) A 500MS/s 5b ADC in 65nm CMOS. In: 2006 Symposium on VLSI Circuits, 2006. Digest of Technical Papers, pp 140–141 Giridhar A, Kumar P (2005) Computing and communicating functions over sensor networks. IEEE J Sel Areas Commun 23 Golabighezelahmad S, Klumperink EA, Nauta B (2020) A 0.7-5.7 GHz reconfigurable MIMO receiver architecture for analog spatial notch filtering using orthogonal beamforming Gorodilova A (2019) On the differential equivalence of APN functions. Crypto Commun 11(4):793–813 Gray RM, Stockholm TG (1993) Dithered quantizers. IEEE Trans Inf Theory 39(3):805–812 Heath RW, González-Prelcic N, Rangan S, Roh W, Sayeed AM (2016) An overview of signal processing techniques for millimeter wave MIMO systems. IEEE J Sel Top Signal Process 10(3):436–453 Ho CC, Lee TC (2012) A 10-bit 200-ms/s reconfigurable pipelined A/D converter. In: Proceedings of Technical Program of 2012 VLSI Design, Automation and Test, pp 1–4

5 Hardware-Limited Task-Based Quantization in Systems

157

Ho T, Médard M, Koetter R, Karger D, Effros M, Shi J, Leong B (2006) A random linear network coding approach to multicast. IEEE Trans Inf Theory 52:4413–30 Huang C, Tan Z, Yang S, Guang X (2018) Comments on cut-set bounds on network function computation. IEEE Trans Inf Theory 64(9):6454–6459 Huang C, Hu S, Alexandropoulos GC, Zappone A, Yuen C, Zhang R, Di Renzo M, Debbah M (2020) Holographic MIMO surfaces for 6G wireless networks: Opportunities, challenges, and trends 27(5):118–125 Iosif A, Ding X, Yu Y (2012) Lecture notes in optimization Ioushua SS, Eldar YC (2019) A family of hybrid analog-digital beamforming methods for massive MIMO systems 67(12):3243–3257 Joram N, Mayer U, Eickhoff R, Ellinger F (2009) Fully integrated active CMOS vector modulator for 802.11a compliant diversity transceivers. In: 2009 IEEE International Conference on Microwaves, Communications, Antennas and Electronics Systems, pp 1–4 Kamran K, Yeh E, Ma Q (2019) DECO: Joint computation, caching and forwarding in data-centric computing networks. In: Proc., ACM MobiHoc, Catania, Italy, pp 111–120 Khobahi S, Shlezinger N, Soltanalian M, Eldar YC (2021) LoRD-Net: unfolded deep detection network with low-resolution receivers. IEEE Trans Signal Process 69(11):5651–5664 Kibaroglu K, Rebeiz GM (2017) A 0.05–6 GHz voltage-mode harmonic rejection mixer with up to 30 dBm in-band IIP3 and 35 dBc HRR in 32 nm SOI CMOS. In: 2017 IEEE Radio Frequency Integrated Circuits Symposium (RFIC), pp 304–307 Kleinrock L (1975) Queuing systems vol. I: theory. Wiley, New York Koetter R, Médard M (2003) An algebraic approach to network coding. IEEE/ACM Trans Netw 11(5):782–795 Körner J (1973) Coding of an information source having ambiguous alphabet and the entropy of graphs. In: Proc., 6th Prague Conf. Inf. Theory, Prague, Czech Republic, pp 411–425 Köse A, Médard M (2017) Scheduling wireless ad hoc networks in polynomial time using clawfree conflict graphs. In: 2017 IEEE 28th Annual International Symposium on Personal, Indoor, and Mobile Radio Communications (PIMRC), IEEE, pp 1–7 Kowshik H, Kumar P (2012) Optimal function computation in directed and undirected graphs. IEEE Trans Inf Theory 58(6):3407–3418 Krishnaswamy H, Zhang L (2016) Analog and RF interference mitigation for integrated MIMO receiver arrays 104(3):561–575 Larsson EG, Edfors O, Tufvesson F, Marzetta TL (2014) Massive MIMO for next generation wireless systems 52(2):186–195 Lee HS, Sodini CG (2008) Analog-to-digital converters: Digitizing the analog world. Proc IEEE 96(2):323–334 Leon-Garcia A (2017) Probability, statistics, and random processes for electrical engineering Li P, Shlezinger N, Zhang H, Wang B, Eldar YC (2022) Graph signal compression by joint quantization and sampling. IEEE Trans Inf Theory 70:4512–4527 Li S, Maddah-Ali MA, Yu Q, Avestimehr AS (2018) A fundamental tradeoff between computation and communication in distributed computing. IEEE Trans Inf Theory 64:109–128 Li SYR, Yeung RW, Cai N (2003) Linear network coding. IEEE Trans Inf Theory 49(2):371–381 Li Y, Tao C, Seco-Granados G, Mezghani A, Swindlehurst AL, Liu L (2017a) Channel estimation and performance analysis of one-bit massive MIMO systems. IEEE Trans Signal Process 65(15):4075–4089 Li Y, Tao C, Seco-Granados G, Mezghani A, Swindlehurst AL, Liu L (2017b) Channel estimation and performance analysis of one-bit massive MIMO systems 65(15):4075–4089 Ma D, Shlezinger N, Huang T, Liu Y, Eldar YC (2021) Bit constrained communication receivers in joint radar communications systems. In: Proc., IEEE ICASSP, Toronto, Canada, pp 8243–8247 Malak D, Médard M (2020) Hyper binning for distributed function coding. In: Proc., IEEE Int. Workshop on Signal Process. Advances in Wireless Commun. (SPAWC) Malak D, Médard M (2023) A distributed computationally aware quantizer design via hyper binning. IEEE Trans Signal Process 71:76–91 Max J (1960) Quantizing for minimum distortion. IRE Trans Inf Theory 6(1):7–12

158

D. Malak et al.

McKeown N (1999) The iSLIP scheduling algorithm for input-queued switches. IEEE/ACM Trans Netw 7(2):188–201 McKeown N, Mekkittikul A, Anantharam V, Walrand J (1999) Achieving 100% throughput in an input-queued switch. IEEE Trans Commun 47(8):1260–1267 Méndez-Rial R, Rusu C, González-Prelcic N, Alkhateeb A, Heath RW (2016) Hybrid MIMO architectures for millimeter wave communications: Phase shifters or switches? IEEE Access 4:247–267 Misra V, Goyal VK, Varshney LR (2011) Distributed scalar quantization for computing: Highresolution analysis and extensions. IEEE Trans Inf Theory 57(8):5298–5325 Mo J, Alkhateeb A, Abu-Surra S, Heath RW (2017) Hybrid architectures with few-bit ADC receivers: Achievable rates and energy-rate tradeoffs 16(4):2274–2287 Monga V, Li Y, Eldar YC (2021) Algorithm unrolling: Interpretable, efficient deep learning for signal and image processing 38(2):18–44 Nelson R (2013) Probability, stochastic processes, and queueing theory: the mathematics of computer performance modeling. Springer Science & Business Media Neuhaus P, Shlezinger N, Dörpinghaus M, Eldar YC, Fettweis G (2021) Task-based analog-todigital converters. IEEE Trans Signal Process 69:5403–5418 Ngo HQ, Larsson EG, Marzetta TL (2013) Energy and spectral efficiency of very large multiuser MIMO systems 61(4):1436–1449 Orlitsky A, Roche JR (2001) Coding for computing. IEEE Trans Inf Theory 47(3):903–17 Padmanabhan M, Bahl LR, Nahamoo D (1999) Partitioning the feature space of a classifier with linear hyperplanes. IEEE Trans Speech Audio Proc 7(3):282–288 Ports DRK, Nelson J (2019) When should the network be the computer? In: Proceedings of the Workshop on Hot Topics in Operating Systems, ACM, New York, NY, USA, HotOS ’19, pp 209–215 Pradhan SS, Ramchandran K (2003) Distributed source coding using syndromes (DISCUS): design and construction. IEEE Trans Inf Theory 49(3):626–643 Qian HJ, Zhang B, Luo X (2019) High-resolution wideband phase shifter with current limited vector-sum. IEEE Trans Circ Syst I Reg Papers 66(2):820–833 Ribeiro A, Giannakis GB (2006a) Bandwidth-constrained distributed estimation for wireless sensor networks-part I: Gaussian case. IEEE Trans Signal Proc 54(3):1131–1143 Ribeiro A, Giannakis GB (2006b) Bandwidth-constrained distributed estimation for wireless sensor networks-part II: unknown probability density function. IEEE Trans Signal Proc 54(7):2784–2796 Rini S, Chataignon J (2019) Comparison-limited vector quantization. Preprint. arXiv:190505401 Rodrigues MRD, Deligiannis N, Lai L, Eldar YC (2017) Rate-distortion trade-offs in acquisition of signal parameters. In: Proc., IEEE ICASSP, New Orleans, LA, USA, pp 6105–6109 Roth K, Pirzadeh H, Swindlehurst AL, Nossek JA (2018) A comparison of hybrid beamforming and digital beamforming with low-resolution ADCs for multiple users and imperfect CSI 12(3):484–498 Rusek F, Persson D, Lau BK, Larsson EG, Marzetta TL, Edfors O, Tufvesson F (2013) Scaling up MIMO: Opportunities and challenges with very large arrays 30(1):40–60 Salamatian S, Cohen A, Médard M (2016) Efficient coding for multi-source networks using GácsKörner common information. In: Proc., IEEE ISITA, Monterey, CA, pp 166–170 Salamatian S, Shlezinger N, Eldar YC, Médard M (2019a) Task-based quantization for recovering quadratic functions using principal inertia components. In: Proc., IEEE ISIT, Paris, France, pp 390–394 Salamatian S, Shlezinger N, Eldar YC, Médard M (2019b) Task-based quantization for recovering quadratic functions using principal inertia components. In: 2019 IEEE International Symposium on Information Theory (ISIT), pp 390–394 Sapio A, Canini M, Ho C, Nelson J, Kalnis P, Kim C, Krishnamurthy A, Moshref M, Ports DRK, Richtárik P (2019a) Scaling distributed machine learning with in-network aggregation. CoRR abs/1903.06701

5 Hardware-Limited Task-Based Quantization in Systems

159

Sapio A, Canini M, Ho CY, Nelson J, Kalnis P, Kim C, Krishnamurthy A, Moshref M, Ports DR, Richtárik P (2019b) Scaling distributed machine learning with in-network aggregation. Preprint. arXiv:190306701 Servetto SD (2005) Achievable rates for multiterminal source coding with scalar quantizers. In: Proc., IEEE Asilomar Conference on Signals, Systems and Computers, Pacific Grove, CA, pp 1762–1766 Shannon CE (1948) A mathematical theory of communication. Bell Syst Tech J 27(3):379–423 Shen L, Suter BW, Tripp EE (2019) Structured sparsity promoting functions. J Optimiz Theory App 183(2):386–421 Shlezinger N, Eldar YC (2018) On the spectral efficiency of noncooperative uplink massive MIMO systems 67(3):1956–1971 Shlezinger N, Eldar YC (2020) Task-based quantization with application to MIMO receivers. Commun Inf Syst 20(2):131–162 Shlezinger N, Eldar YC (2021) Deep task-based quantization. Entropy 23(104):1–18 Shlezinger N, Eldar YC, Rodrigues MR (2019a) Asymptotic task-based quantization with application to massive MIMO 67(15):3995–4012 Shlezinger N, Eldar YC, Rodrigues MRD (2019b) Asymptotic task-based quantization with application to massive MIMO. IEEE Trans Signal Process 67(15):3995–4012 Shlezinger N, Eldar YC, Rodrigues MRD (2019c) Hardware-limited task-based quantization. IEEE Trans Signal Process 67(20):5223–5238 Shlezinger N, van Sloun RJG, Huijben IAM, Tsintsadze G, Eldar YC (2020a) Learning task-based analog-to-digital conversion for MIMO receivers. In: ICASSP 2020 - 2020 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP), pp 9125–9129 Shlezinger N, Whang J, Eldar YC, Dimakis AG (2020b) Model-Based Deep Learning, to appear in Proceedings of the IEEE. Skrimponis P, Dutta S, Mezzavilla M, Rangan S, Mirfarshbafan SH, Studer C, Buckwalter J, Rodwell M (2020) Power consumption analysis for mobile MmWave and sub-THz receivers. In: 2020 2nd 6G Wireless Summit (6G SUMMIT), pp 1–5 Slepian D, Wolf J (1973) Noiseless coding of correlated information sources. IEEE Trans Inf Theory 19(4):471–480 Soer MCM, Klumperink EAM, Nauta B, van Vliet FE (2011) Spatial interferer rejection in a four-element beamforming receiver front-end with a switched-capacitor vector modulator 46(12):2933–2942 Soer MCM, Klumperink EAM, van den Broek DJ, Nauta B, van Vliet FE (2017) Beamformer with constant-GM vector modulators and its spatial intermodulation distortion. IEEE J Solid State Circuits 52(3):735–746 Tung SY (1978) Multiterminal source coding. Cornell University Von Neumann J (1953) A certain zero-sum two-person game equivalent to the optimal assignment problem. Contrib Theory Games 2(0):5–12 Walden R (1999) Analog-to-digital converter survey and analysis 17(4):539–550 Walrand J (1983) A probabilistic look at networks of quasi-reversible queues. IEEE Trans Info Theory 29(6):825–831 Wang H, Shlezinger N, Eldar YC, Jin S, Imani MF, Yoo I, Smith DR (2021) Dynamic metasurface antennas for MIMO-OFDM receivers with bit-limited ADCs 69(4):2643–2659 Wang J, Sahu AK, Yang Z, Joshi G, Kar S (2019) MATCHA: Speeding up decentralized SGD via matching decomposition sampling. Preprint. arXiv:190509435 Wang X, Orchard MT (2001) Design of trellis codes for source coding with side information at the decoder. In: Proc., IEEE Data Compression Conference (DCC), Snowbird, UT, pp 361–370 Witsenhausen H (1976) The zero-error side information problem and chromatic numbers (corresp.). IEEE Trans Inf Theory 22(5):592–593 Wolf J, Ziv J (1970) Transmission of noisy information to a noisy receiver with minimum distortion. IEEE Trans Inf Theory 16(4):406–411 Xi F, Shlezinger N, Eldar YC (2021) BiLiMO: Bit-limited MIMO radar via task-based quantization. Trans Signal Process 69:6267–6282

160

D. Malak et al.

Yu K, Zhang YD, Bao M, Hu YH, Wang Z (2016a) DOA estimation from one-bit compressed array data via joint sparse representation. IEEE Signal Process Lett 23(9):1279–1283 Yu L, Li H, Chen CW (2016b) Generalized common informations: Measuring commonness by the conditional maximal correlation. Preprint. arXiv:161009289 Yu Q, Maddah-Ali MA, Avestimehr AS (2018) The exact rate-memory tradeoff for caching with uncoded prefetching. IEEE Trans Inf Theory 64(2):1281–1296 Zeitler G, Kramer G, Singer AC (2012) Bayesian parameter estimation using single-bit dithered quantization. IEEE Trans Signal Process 60(6):2713–2726 Zirtiloglu T, Shlezinger N, Eldar YC, Yazicigil RT (2022) Power-efficient hybrid MIMO receiver with task-specific beamforming using low-resolution ADCs. In: 2022 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP). https://doi.org/10.1109/ ICASSP43922.2022.9746362

Derya Malak is a tenure track Assistant Professor (a Maitre de Conference) in the Communication Systems Department at Eurecom. Previously, she was an Assistant Professor in the Department of Electrical, Computer, and Systems Engineering at Rensselaer Polytechnic Institute between 2019 and 2021, and a Postdoctoral Associate at MIT between 2017 and 2019. She received a Ph.D. in ECE from the University of Texas at Austin in 2017, where she was affiliated with the Wireless Networking and Communications Group. She received a B.S. in Electrical and Electronics Engineering (EEE) with a minor in Physics from Middle East Technical University, Ankara, Turkey, in 2010, and an M.S. in EEE at Koc University, Istanbul, Turkey, in 2013. Dr. Malak has held visiting positions in INRIA and LINCS, Paris, France, and from Northeastern University, Boston, MA. She has held summer internships at Huawei Technologies, Plano, TX, and Bell Laboratories, Murray Hill, NJ. She was awarded the Graduate School fellowship by UT Austin between 2013 and 2017. She was selected to participate in the Rising Stars Workshop for women in EECS, MIT, Cambridge, MA, in 2018, and the 7th Heidelberg Laureate Forum, Heidelberg, Germany, in 2019. Dr. Malak has expertise in both information theory and networking areas. She has developed novel distributed computation solutions and wireless caching algorithms by capturing the confluence of storage, communication, and computation aspects. Her past work with Huawei and Nokia Bell Labs has led to 3GPP standards contributions in device-to-device communication and radio access networks. Dr. Malak’s research has been funded by the HUAWEI Chair, the National Science Foundation (NSF), the Rensselaer-IBM AI Research Collaboration, and the DARPA Dispersive Computing Programs.

5 Hardware-Limited Task-Based Quantization in Systems

161

Rabia Yazicigil is an Assistant Professor of Electrical and Computer Engineering at Boston University and a Visiting Scholar at MIT. She received her PhD degree in Electrical Engineering from Columbia University in 2016. She received the B.S. degree in Electronics Engineering from Sabanci University, Istanbul, Turkey in 2009, and the M.S. degree in Electrical and Electronics Engineering from Ecole Polytechnique Federale de Lausanne (EPFL), Lausanne, Switzerland in 2011. Her research interests lie at the interface of integrated circuits, signal processing, security, bio-sensing, and wireless communications to innovate system-level solutions for future energyconstrained applications. She has been a recipient of a number of awards, including the “Electrical Engineering Collaborative Research Award” for her PhD research on Compressive Sampling Applications in Rapid RF Spectrum Sensing (2016), the second place at the Bell Labs Future X Days Student Research Competition (2015), Analog Devices Inc. outstanding student designer award (2015) and 2014 Millman Teaching Assistant Award of Columbia University. She was selected among the top 61 female graduate students and postdoctoral scholars invited to participate and present her research work in the 2015 MIT Rising Stars in Electrical Engineering Computer Science. She was selected as a semi-finalist for 2018—35 Innovators Under 35 list sponsored by MIT Technology Review. Muriel Médard is the NEC Professor of Software Science and Engineering in the School of Engineering at MIT and a professor in the Electrical Engineering and Computer Science (EECS) Department at MIT, Cambridge, MA, USA. She leads the Network Coding and Reliable Communications Group in the Research Laboratory for Electronics at MIT and Chief Scientist for Steinwurf, which she has co-founded. She obtained three Bachelor’s degrees, as well as her MS and ScD, all from MIT. Muriel is a member of the US National Academy of Engineering (elected 2020), a member of the German National Academy of Sciences Leopoldina (elected 2022), a fellow of the US National Academy of Inventors (elected 2018), American Academy of Arts and Sciences (elected 2021), and a fellow of the Institute of Electrical and Electronics Engineers (elected 2008). She holds honorary doctorates from the Technical University of Munich (2020) and the University of Aalborg (2022). She was awarded the 2022 IEEE Kobayashi Computers and Communications Award. She received the 2017 IEEE Communications Society Edwin Howard Armstrong Achievement Award and the 2016 IEEE Vehicular Technology James Evans Avant Garde Award. Muriel was co-winner of the MIT 2004 Harold E. Egerton Faculty Achievement Award and was named a Gilbreth Lecturer by the US National Academy of Engineering in 2007. She received the 2019 Best Paper award for IEEE Transactions on Network Science and Engineering, the 2018 ACM SIGCOMM Test of Time Paper Award, the 2009 IEEE Communication Society and Information Theory Society Joint Paper Award, the 2009 William R. Bennett Prize in the Field of Communications Networking, the 2002 IEEE Leon K. Kirchmayer Prize Paper Award, as well as nine conference paper awards. Most of her prize papers are co-authored with students from her group. She currently serves as the Editor-in-Chief of the IEEE Transactions on Information Theory and served previously as Editor in Chief of the IEEE Journal on Selected Areas in Communications. She was elected president of the IEEE Information Theory Society in 2012, and serves on its board of governors, having previously served for 11 years. She has supervised over 40 master students, over 20 doctoral students and over 25 postdoctoral fellows. She received the inaugural MIT Postdoctoral Association Mentoring Award in 2022, the inaugural MIT EECS Graduate Student Association Mentor Award, voted by the students

162

D. Malak et al.

in 2013. She set up the Women in the Information Theory Society (WithITS) and Information Theory Society Mentoring Program, for which she was recognized with the 2017 Aaron Wyner Distinguished Service Award. She has over 60 US and international patents awarded, the vast majority of which have been licensed or acquired. For technology transfer, she has co-founded CodeOn, for intellectual property licensing, and Steinwurf, for reliable and low-latency networking. She serves on the Nokia Bell Labs Technical Advisory Board. Xing Zhang received the B.E. degree in information engineering in 2015 and the Ph.D. degree in information and communication engineering in 2021, both from the School of Information Science and Engineering, Southeast University, China. She is currently a Postdoctoral Researcher with the Signal Acquisition Modeling and Processing Laboratory, Weizmann Institute of Science, Rehovot, Israel. Her research interests include signal processing and its applications on communications.

Yonina C. Eldar received the B.Sc. degree in Physics in 1995 and the B.Sc. degree in Electrical Engineering in 1996 both from Tel-Aviv University (TAU), Tel-Aviv, Israel, and the Ph.D. degree in Electrical Engineering and Computer Science in 2002 from the Massachusetts Institute of Technology (MIT), Cambridge. From January 2002 to July 2002, she was a Postdoctoral Fellow at the Digital Signal Processing Group at MIT. She is currently a Professor in the Department of Mathematics and Computer Science, Weizmann Institute of Science, Rehovot, Israel. She was previously a Professor in the Department of Electrical Engineering at the Technion, where she held the Edwards Chair in Engineering. She is also a Visiting Professor at MIT, a Visiting Scientist at the Broad Institute, and an Adjunct Professor at Duke University and was a Visiting Professor at Stanford. She is a member of the Israel Academy of Sciences and Humanities (elected 2017), an IEEE Fellow and a EURASIP Fellow. Dr. Eldar has received numerous awards for excellence in research and teaching, including the IEEE Signal Processing Society Technical Achievement Award (2013), the IEEE/AESS Fred Nathanson Memorial Radar Award (2014), and the IEEE Kiyo Tomiyasu Award (2016). She was a Horev Fellow of the Leaders in Science and Technology program at the Technion and an Alon Fellow. She received the Michael Bruno Memorial Award from the Rothschild Foundation, the Weizmann Prize for Exact Sciences, the Wolf Foundation Krill Prize for Excellence in Scientific Research, the Henry Taub Prize for Excellence in Research (twice), the Hershel Rich Innovation Award (three times), the Award for Women with Distinguished Contributions, the Andre and Bella Meyer Lectureship, the Career Development Chair at the Technion, the Muriel & David Jacknow Award for Excellence in Teaching, and the Technion’s Award for Excellence in Teaching (twice).

5 Hardware-Limited Task-Based Quantization in Systems

163

She received several best paper awards and best demo awards together with her research students and colleagues including the SIAM outstanding Paper Prize and the IET Circuits, Devices, and Systems Premium Award and was selected as one of the 50 most influential women in Israel. She was a member of the Young Israel Academy of Science and Humanities and the Israel Committee for Higher Education. She is the Editor in Chief of Foundations and Trends in Signal Processing, a member of the IEEE Sensor Array and Multichannel Technical Committee and serves on several other IEEE committees. In the past, she was a Signal Processing Society Distinguished Lecturer, member of the IEEE Signal Processing Theory and Methods and Bio Imaging Signal Processing technical committees, and served as an associate editor for the IEEE Transactions On Signal Processing, the EURASIP Journal of Signal Processing, the SIAM Journal on Matrix Analysis and Applications, and the SIAM Journal on Imaging Sciences. She was Co-Chair and Technical Co-Chair of several international conferences and workshops. She is author of the book “Sampling Theory: Beyond Bandlimited Systems” and co-author of the books “InformationTheoretic Methods in Data Science”, “Compressed Sensing in Radar Signal Processing”, “Compressed Sensing”, and “Convex Optimization Methods in Signal Processing and Communications”, all published by Cambridge University Press.

Chapter 6

Satellite Communications Toward a Sustainable 3D Wireless Network Ana I. Pérez Neira

Acronyms 2D 3D 2G 3G 4G 5G 6G 3GPP ACM AI CD CDMA CTTC dB ESA ETSI FDD Gb/s GEO GNSS HTS HW IoT IP ISL LEO

2 Dimensions 3 Dimensions 2nd generation of wireless communications 3th generation of wireless communications 4th generation of wireless communications 5th generation of wireless communications 6th generation of wireless communications 3rd generation partnership project Adaptive code and modulation Artificial intelligence Compact disk Code division multiple access Centre Tecnològic de Telecomunicacions de Catalunya Decibels European Space Agency European telecommunications standards institute Frequency division duplex Gigabits por segundo Geostationary Global navigation satellite system High throughput satellite Hardware Internet of things Internet protocol Inter satellite link Low earth orbit satellite

A. I. Pérez Neira () Centre Tecnológic de Telecomunicacions de Catalunya - CERCA, Castelldefels (Barcelona), Spain e-mail: [email protected] © The Author(s), under exclusive license to Springer Nature Switzerland AG 2023 M. S. Greco et al. (eds.), Women in Telecommunications, Women in Engineering and Science, https://doi.org/10.1007/978-3-031-21975-7_6

165

166 MEO NFV NGEO RFC SDN SNR SLA SNR Tbp/s TDD UHTS USA

A. I. Pérez Neira Medium earth orbit Network function virtualization Non-geostationary Radio frequency coding Software defined network Signal to noise ratio Service level agreement Signal to noise ratio Terabits por segundo Time division duplex Ultra-high throughput satellite United States of America

6.1 The Sputnik and the Space Race 6.1.1 Decisive Elements in the Development of Digital Communications The era of digital communications began with Claude Shannon’s Information Theory, the foundations of which he published in his 1949 article (Shannon 1949). One of the key concepts in it is that of channel capacity, which, as Moore’s law does, offers a limit of the maximum speed at which it can be transmitted (bits/second or b/s) (6.1) given the bandwidth B, the noise power σ 2 of the transmission channel, and a maximum transmission power P,   P = Blog2 (1 + SNR) .C  Blog2 1 + σ2

[b/s] .

(6.1)

The relationship between the power that accompanies the desired message, P, and the noise power is called the signal-to-noise ratio or SNR (equal to P/σ 2 ). This law was completely disruptive because it showed for the first time that to transmit at high speed the transmission power is not the only tool but that the bandwidth available in the channel is also important. This is the reason why each new generation of satellites and radio communications generally goes up on the carrier frequency, allowing to grow the transmit bandwidth. Another key concept in C. Shannon’s article is that, in a point-to-point communication, to achieve the highest quality, the message must be compressed into bits (source encoding) and then encoded adding the appropriate redundancy (encoding channel), to protect said transmission from errors. His work initially motivated endless theoretical investigations, which had not yet materialized in any technological development. After all, the power required by existing communications to obtain good communication was within the supply range. Also, channel codes were complicated and expensive to implement. It was the space race (accelerated in the USA by the launch of the Russian Sputnik satellite in

6 Satellite Communications Toward a Sustainable 3D Wireless Network

167

Fig. 6.1 Hispasat 36 W-1 satellite at Airbus Defense and Space, compact antenna test in anechoic chamber, Ottobrunn, Germany, September 2016. (© ESA – P. Sebirot)

Fig. 6.2 Hispasat 36 W-1 is the first SmallGEO developed by OHB; it offers 3 kW and covers Europe, the Canary Islands, and America, in Ka and Ku band. (© ESA – P. Carril)

1957) that created the need to implement these codes, as well as the technological development for it. Having power in space is not trivial; it makes the payload of the satellite very expensive (Figs. 6.1 and 6.2 show examples of communications satellites in geostationary orbit), and other ways to combat noise were necessary to achieve error-free long-distance communication (e.g., satellites in geostationary orbit are more than 36,000 km away). In other words, it became necessary to implement channel coding. In 1960, every dB that encoding saved meant more than $ 1,000,000. This is how space communications not only gave the first real application to channel coding but also created the need for rapid technological development. In 1958, Texas Instrument invented the integrated circuit, reducing the cost, weight, and power required for an encoder implementation. At the same time, the decoding speed increased. All of this allowed the development of new coding techniques and new applications outside of space. For example, in 1960 the Reed-Solomon codes were invented, used in reading CDs. In the same way that the lack of HW technology almost calls into question the usefulness of channel encoders, their activation, with the acceleration of

168

A. I. Pérez Neira

Fig. 6.3 Basic architecture of a communication via satellite

the semiconductor revolution in the early 1970s, which gave rise to Intel’s first microprocessor, made viable first commercial applications, that is, outside the space or military environment. The history of these years is fascinating, in 1968 two North American professors, Irwin Jacobs and Andrew Viterbi, created the consultancy Linkabit in the field of encoding and decoding. These years were the era of the Viterbi set-top box and TDMA communications. Linkabit developed the first commercial encrypted system for satellite television, VideoCypher. Both researchers left Linkabit to found Qualcomm, whose objective was exclusively radio communications and created the CDMA system; this is a code access system that spreads the signal spectrum to accommodate multiple users at the same time and on the same frequencies. This system was adopted by both 3G radio systems and satellite communications. All these and subsequent advances in radio communications are based on Information Theory. There are commercial “modems” that practically reach the so-called Shannon capacity limit, that is, they are capable of solving the worst case of transmission through channels with additive Gaussian noise distortion. However, there are still channels to be resolved. These channels are all those in which the communication is not point-to-point, but multipoint, that is, they are communications within a network. In satellite communication, at least three nodes are involved: the transmitter, the satellite, and the receiver (Fig. 6.3). Therefore, strictly speaking, it is not a point-to-point communication and may require new types of encryption. This was the reason why Raymod Yeung devised new codes and coding strategy (Yeung and Zhang 1999): network coding or “network coding,” which was later exported to the rest of radio communications. Again, satellite communications pose new challenges. In a generic communication by satellite, the transmitter and the receiver are at a great distance, and the most basic mission that the satellite performs is to repeat the signal from the transmitter to the receiver. Therefore, although it is a pointto-point communication, geometrically we can speak of a transmission at least

6 Satellite Communications Toward a Sustainable 3D Wireless Network

169

two-dimensional and multidimensional if more than one satellite is used. By taking advantage of the broadcast nature of the wireless medium, R. Yeung observed that the satellite or repeater could implement a coding such that delay time could be saved with respect to a classical wired relay transmission. We note that delay is critical in satellite transmission. R. Yeung conceived this encoding for complex network architectures where intermediate nodes in addition to repeating the signal have to route it. Such was the case of the IridiumTM systems from Motorola or GlobalStarTM from Qualcomm, which in the 1990s offered telephone services (in a coverage of the whole Earth) and data (in a coverage ±70◦ latitude), respectively, to mobile users through multiple satellite links. The satellites were not in geostationary orbits, and therefore a user was covered by more than one satellite and a satellite covered more than one user, thus creating a complex network where intelligent routing had to be carried out. At present, network coding has numerous applications (Fragouli and Soljanin 2008) such as content distribution and sensor networks, among others. What has been exposed so far highlights the particularities of satellite communications with respect to terrestrial radio communications, as well as the relevance of the space sector within the general panorama of digital communications. However, these particularities in turn make it difficult to successfully establish (in terms of complexity versus cost) a satellite-terrestrial hybrid communications system, where the user can have the advantages of both segments in a transparent manner. Next, we present the development of the fifth Generation (5G) standard with the aim that both segments can be integrated.

6.1.2 The Standardization of the Non-terrestrial Segment in 5G and Its Evolution Toward a 3-Dimensional Communication The 5G communications standard is called “New Radio” (NR) (Lin et al. 2019) and had its first delivery in April 2019 (Release 15), and its objective is to cover the following three general use cases: high-speed mobile transmissions, communications between machines and very low latency, and high reliability communications. Simultaneously with its development, we are witnessing a renewed interest in offering connectivity through space. In recent years, different promises of large constellations of low-orbit satellites or LEOs have emerged, such as OneWeb (WorldVu Satellites Limited 2020) and SpaceX [Spa20], which aim to offer broadband access at low cost. For this reason, future NR deliveries intend to integrate satellite communications within 5G, in order to facilitate ubiquitous connectivity. As discussed above, the ambition to offer connectivity from space is not new: IridiumTM and GlobalStarTM emerged in the 1990s. However, their success was limited due to the rapid growth of terrestrial radio communications networks, which

170

A. I. Pérez Neira

were more economical to deploy. In addition, at that time, voice services dominated radio communications, and these are very sensitive to transmission with high delays, a weak point in satellite communications. Currently, different facts have created a good opportunity both for the development of advanced technologies in the satellite segment and for the integration of satellite communications within a generic communications network. Some of these technologies are (Perez-Neira et al. 2019) multi-beam satellites, onboard digital processing, as well as advanced modulation and coding schemes. The main facts are: • The further digitization of all communication networks. • The supremacy of data services over voice services. • The increasing demand for continuity in communication services regardless of the geographical location (mountain, sea, etc.) • A significant reduction in costs in the manufacturing process of a satellite and its launch. • Advances in microelectronics.

6.1.2.1

Use Cases

The satellite has an important complementary role in the communications ecosystem. The reason is that, despite the widespread deployment of land mobile networks, there are still areas with no or very poor service due to cost reasons. For example, offering coverage in rural or remote areas is a challenge for many countries because the investment cost of the terrestrial deployment involved is not justified by the benefits that can be obtained. In contrast, a single communications satellite can cover a large geographic area and can be economically viable and attractive as a solution for extending land coverage. As an example, we will cite that SpaceX has won a tender to provide broadband between 100 and 200 Mb/s in rural USA, the same as Telesat has achieved in Canada. Satellite links also have a constant propagation delay, unlike terrestrial networks with multiple nodes through which a communication can be routed. This controlled delay may be of interest in certain applications such as intercontinental banking transactions. In urban areas, satellites called “high throughput satellites,” or highspeed transmission satellites, can also help decongest the traffic that terrestrial networks have to carry. Another possible growing business area is providing coverage on airplanes or in maritime areas, due to the growing tourism and freight transport by these means. Telemetry applications for machine-to-machine or IoT communications (e.g., high precision agriculture, intelligent water control, air quality surveillance, etc.), which may be in sparsely populated areas, can also be better resolved by satellites. Finally, the satellites can be used to build up ad hoc tactical or public safety networks. Offering continuity in services is closely related to ubiquitous connectivity. When a user enters an area with poor coverage, the service is interrupted. The

6 Satellite Communications Toward a Sustainable 3D Wireless Network

171

integration of terrestrial and satellite communications would solve these continuity holes. Therefore, when future communication networks know how to integrate the space segment by collecting all its casuistry, then we can talk about 3D networks that offer the right connection in a sustainable way at all times and in any place. For example, the satellite may be easier to integrate and operate by local communities to extend services to remote rural areas. In general, they will be more scalable networks, which will reduce costs.

6.1.2.2

The Digitization of Networks: SDN Y NFV

One of the aspects that 5G has brought, unlike its predecessors (2G, 3G, 4G), is a high digitization that allows controlling and building by “software” various virtual communications networks on the same infrastructure. In this way, all the resources offered by an infrastructure are used more profitable by mapping them into multiple services. These are the so-called software-defined networks or SDN. To do this, even the radio frequency signal (the one closest to the antenna) has to be sampled. In this way, all the functions and directives of the network can be defined in “software” (NFV), becoming a virtual network on an infrastructure. In other words, by sharing the same infrastructure, different services can be offered and superimposed. This high digitization has been possible thanks to the increase in the calculation and storage capacities of general purpose processors, which can act as routers of the packets within the network, or as base stations, or as access points, for example. Therefore, specific processors are no longer necessary, thus reducing the cost and increasing the densification of communication nodes in the network, which may have more and more distributed intelligence. In this way, flexibility is given to the network. Very short reaction times are achieved, resources are used to the maximum, and a truly sustainable and scalable network is obtained, allowing coexistence and adaptation to very different SLAs. All of this is practically a reality today thanks to 5G standardization. We speak then of the network in the cloud, controlled by large “data centers.” A modem no longer has to be a physical device but can be made up of a set of SW functions in the cloud. An example of all that has been said is Microsoft’s Azure Orbital platform (Azure Orbital n.d.). Azure Orbital is a fully “software” ground station in the cloud that Microsoft offers as a service to satellite operators to transmit via satellite and process data from their satellites or spacecraft. Its model is pay-per-use, without the operator having to create its own ground stations for communication with satellites. With the digitization of space and the network in general, the location of gateways will be an educated decision between places with good satellite vision and places close to cloud servers. Developments that take days today will happen in seconds thanks to SDN and NFV. Digitization will create huge amounts of data, which can be managed with AI. The “softwarization” will allow expanding and contracting the network geographically as needed, without the need to increase the deployment of the infrastructure or the HW. An infrastructure with that technology will enable a 3D network, sharing Earth and space in their different

172

A. I. Pérez Neira

orbits. The network will be scalable via “software,” with all the sustainability and cost reduction that this entails. The SDNs and virtualization described is what has truly made the different agents of satellite communications (operators and manufacturers) see integration with the terrestrial segment in an easy and low-cost way; it does not require large investments in HW, but it is development of the “software.” By collecting very diverse infrastructures (radio, satellite, fiber optics, etc.), it is said that 5G is a network of networks. In fact, unlike 4G, 5G has not involved the development of any new technology (thus following the implicit rule that it is the even generations of communications that represent technological advances). What it has meant is a great advance in the integration of services in very different sectors (automotive, health, etc.) within the communications network. The work being carried out at the 5G standardization level for the inclusion of the satellite segment (3GPP TR 38.811 2018; 3GPP TR 22.822 2018) shows that your industry has realized that it is facing a unique opportunity to reduce operating costs of your network, as well as those of the user terminals, using the compatibility that guarantees standardization. Until now, interoperability between different manufacturers of satellite equipment was almost non-existent, and the market was highly fragmented (EMEA Satellite Operators Association 2020). The opportunities that 5G has opened up, combined with the increasing view of space that defense organizations in different countries have, make it realistic this time to foresee a new leap in the commercialization of satellite communications. Projects such as Blackjack (DARPA 2020) cannot be ignored, the objective of which is to create low-orbit military satellite constellations from the technologies developed for commercial satellites. Later we will return to this topic of SW networks before we will present the basic characteristics of a satellite channel.

6.2 GEO or No GEO? A Second Space Revolution Typically, mobile operators use the connection to geostationary or GEO satellites when they must serve rural or semi-rural areas, and it is not profitable to deploy fiber optics. In parallel to large capacity GEO satellites (HTS), new constellations of MEO and LEO satellites have emerged in recent years. Above all, they are attractive because they offer very low latency. The satellites are connected to each other through ISL links, generally optical, which reduces the number of base stations on the ground to route traffic. These new LEO constellations promise to also be high capacity HTS and operate with native IP protocols (Van der Breggen 2017). The LEO Starlink constellation, operated by SpaceX, offers low-cost broadband internet, low latency, and global coverage. Regulatory requirements were completed in 2017 to launch nearly 12,000 satellites. The first 60 satellites were launched in May 2019 and commercial operations are expected to begin shortly, offering 100 Mb/s and 20 ms latency.

6 Satellite Communications Toward a Sustainable 3D Wireless Network

173

Large producers such as Lockheed Martin are developing new aerospace platforms that integrate AI, “software” and metal 3D printing technology, to build both satellites and launch rockets. This promises a reduction in production time to days instead of years and cost by 50%. These perspectives have led to the emergence of new players who have opted for non-GEO mega constellations, mainly, SpaceX, OneWeb, Google, Amazon, Relativity Space, Telesat, Hiber, Capella Space, and ST Engineering, in addition to O3b that was bought by the large operator SES. The lower costs and the emergence of numerous new companies and “start-ups” has created a unique situation, and there is talk of the second space revolution, which will create a rich space ecosystem in different orbits, between which there will be diversity and communication. Many analysts believe that the non-GEO satellite communications market has gone from a speculation phase to an affirmation phase (Suresh n.d.). The outlook is boiling and promising, although it is important that new services do not take long to get up and running, since it is risky to predict demand for many years to come. While waiting for these new constellations to operate in a stable manner, communications with GEO satellites are the ones that prevail to date. The greatest gains obtained with communications from GEO satellites have come from taking advantage of their great coverage and broadcasting capacity, where the increase in users did not imply an increase in cost on the part of the operator. These gains decrease by two orders of magnitude when moving to broadband communications services, which compete directly with land mobile communications and fiber communications. Mobile operators only use satellites to cover rural or semi-rural areas where it is expensive to deploy fiber. Therefore, the GEO satellite industry must evolve and, among other things, invest part of its high profits in researching new technologies, as the terrestrial segment has done, for example, to make user terminals cheaper. Most probably, a good strategy is to consider a multi-layer approach to deliver new innovations in global, mobile satellite communications and transform the capabilities offered to Internet of Things (IoT) and mobility customers for years to come. That is, to consider a combination of GEO and NGEO satellites that allows to redefine connectivity at scale with the highest capacity for mobility worldwide and at hot spots, as well as the fastest average speeds and the lowest average latency of any network, planned or in existence. In this way, new use cases can be considered such as (i) urban air mobility (e.g., autonomous flying taxis and personal air transport); (ii) large-scale IoT deployments that can integrate, manage, and monitor disparate sensors and devices via a single cloud environment; (iii) smart cruise ships; (iv) tactical private networks; and (v) public safety networks. An important issue in the integration with terrestrial communications networks is how to share the radio frequency spectrum between the satellite segment and the terrestrial segment. Satellite communications are severely limited by noise and the presence of any interference can seriously degrade them. This has led to the fact that frequencies much higher than those used in land mobile communications have always been used in the satellite segment, that is, to say, the Q\V\W bands. In fact, today there is a great activity of research and development of technology to establish

174

A. I. Pérez Neira

communications by optical satellites. These are endorsed by companies with great prestige in their career in telecommunications, such as Tesat (https://www.tesat.de/) or the DLR (German Space Agency). Communications in THz (100–300 GHz) are also the subject of study (Saqlain 2019). At these frequencies the link capacity shoots up, hence its appeal. However, the atmospheric attenuations are also much greater, and this represents a great technological challenge. If the interference generated by sharing the spectrum is known, advanced techniques can be used to mitigate and control such interference (Artiga et al. 2018).

6.3 The Architecture and the Satellite Communication Channel Today there are approximately 534 operational satellites in GEO orbit (https://www. satsig.net/sslist.htm) and more than 10,000 in other orbits. Each one plays a different role in communications.

6.3.1 Satellite Orbits The altitude of a satellite orbit, which is the distance from the satellite to the Earth’s surface, determines the orbital speed of the satellite around it. The satellite orbit also depends on the eccentricity of said orbit and its inclination. There are three types of orbits depending on their height. The geostationary orbit (GEO) is the one that has an altitude of 35,786 km and has the same orbital period as the Earth, that is, it takes 24 hours to go around the Earth. For an observer who is in this, the satellite appears fixed in a certain longitude, although it can drift towards the north or south. Medium Earth orbits (MEO) (Fig. 6.4) are those that are between 2000 km from Earth to geostationary orbit. The associated orbital periods vary between 2 and 24 hours. Low Earth (LEO) orbits (Fig. 6.5) lie between 160 km and 2000 km from Earth. LEO satellites move fast around it and have a period between 1.5 and 2 h. Because a LEO satellite can quickly change its position to an observer on Earth, the satellite is only visible to him for a few minutes. There are other possible orbits such as polar or synchronous orbits with the sun. In this case, they always pass over a geographical point at the same time. These orbits are used for satellites not for communications but for Earth observation, as they can thus compare the images they take from the same site. A satellite in an orbit synchronized with the sun is usually at an altitude between 600 and 800 km. If you are at 800 km, you will travel at a speed of about 7.5 km/s.

6 Satellite Communications Toward a Sustainable 3D Wireless Network

175

Fig. 6.4 MEO orbits, example of the constellation Galileo for positioning. (© ESA P. Carril)

Fig. 6.5 LEO orbits. (© ESA –L. Boldt-Christmas)

6.3.2 The Basic Architecture As we have advanced before (Fig. 6.3), a basic satellite communication system consists of the following components: • The satellite: contains a bus to control satellite operations (power, thermal control, altitude control, etc.) and a payload for communications (antennas, transponders and microprocessors in the most advanced).

176

A. I. Pérez Neira

• User terminal. • Gateway: base station that, on the one hand, controls the operation of the satellite with telemetry and, on the other, connects it with the rest of the terrestrial network. • “Feeder link”: communication link between the “Gateway” and the satellite. • “Service link”: communication link between the satellite and the user terminal. Each link or “link” operates at different frequencies, which also varies depending on whether the link is upstream or downstream to the satellite. As low frequencies suffer less propagation losses, they are used for satellite transmission, which is more power-limited. Depending on the functionality implemented in the communications load carried by the satellite, two types can be considered: transparent or regenerative. With the first, the satellite receives the signals from the Earth, amplifies them, and retransmits them after having changed them to other more suitable carrier frequencies. Basically, the higher the frequency, the more it is attenuated, and therefore the signals will go on some carriers or others depending on the power availability of the transmitter. In the case of a regenerative satellite, it performs onboard processing to demodulate and decode the received signals and thus clean them of noise for later retransmission. Although this last type of satellite is more desired, its development cost and power consumption mean that it has only recently begun to be manufactured, with the repeater satellite being the most common. The latest generation satellites use multi-beam technology to cover different areas of the Earth with more resolution and flexibility. Due to their importance, I will dedicate a chapter exclusively to them later. The footprint of a beam is generally elliptical and is generally considered equivalent to what a cell would be in mobile land communications. For non-geostationary satellites (MEO, LEO), the beam tracks can move and sweep the Earth thus following the movement of the satellite, or, alternatively, they can be designed to remain fixed on the Earth. For this, it is necessary to develop beam pointing mechanisms that compensate for the movement of the satellite. The radius of the beam depends on the satellite and can range from a few tens of kilometers to a few hundred of them.

6.3.3 Satellite Channel Characteristics The main characteristics are described below to understand how they can impact the integration of satellite communications with terrestrial communications.

6.3.3.1

High Propagation Losses: Communication Dominated by Noise

Due to the large distances to be covered between a terrestrial terminal and a satellite, satellite communications introduce high propagation losses. Being direct

6 Satellite Communications Toward a Sustainable 3D Wireless Network

177

vision communications, the received power, Pr , is formulated as a function of the transmitted power, Pt , through Friis law, as:  Pr = Pt Gr Gt

.

λ 4π d

2 [W]

(6.2)

where Gr y Gt are the gains of the receiving and transmitting antennas, respectively, λ is the carrier wavelength, and d is the distance between the satellite and the terrestrial terminal. To compensate for these losses, satellites need power amplifiers, which tend to work in saturation, creatingunwanted nonlinear distortions. We note that Friis law implies that losses (i.e., . PPrt in the isotropic case (i.e., Gr = Gt = 1) increase inversely to λ2 . This implies that, in the absence of directional antennas, satellite communications experience more attenuation the higher the frequency, that is, the lower λ. However, given antenna aperture dimensions, Gr , Gt increase with λ−1 . Therefore, these gains largely offset the frequency losses. On the other hand, communication must traverse around 20 km of atmosphere and, therefore, experiences high attenuations due to molecular absorption, which are increased if there is rain or clouds. These losses are notable above 10 GHz. Due to all this, and because the satellites have limited power, the satellite links have been conditioned mainly by noise and a precise pointing between the transmitting and receiving antennas is necessary to achieve that Gr y Gt reach their maximum value. It is usual to work with SNRs of 15 dB or less, while in land mobile communications the usual SNR is around 20 or 25 dB.

6.3.3.2

Communications with a Significant Line-of-Sight Component

To compensate the propagation losses, it is of outmost importance a precise transmitting and receiving antennas pointing. Therefore, the communication channel model must have a strong direct vision component. If they were not direct vision communications, the attenuation due to blocking obstacles would make reception with the adequate signal level very difficult. For example, at satellite transmission frequencies, materials such as brick can attenuate between 40 and 80 dB. The human body itself attenuates between 20 and 35 dBs. In this sense, they are more vulnerable than land mobile communications, where d is not as high as there are numerous base stations, which, on the other hand, do not have available power problems. On the other hand, the absence of “scatters” around the satellite also contrasts with terrestrial mobile communications, where the large number of objects creates a communication full of multiple paths or “scatters.” In short, the channels of one or the other communication follow different models.

178

6.3.3.3

A. I. Pérez Neira

Satellite Transmission Frequencies

The selection of the frequency is determined by different aspects. Some of them have already been commented: coverage and size of the beam, atmospheric conditions in the region to be served, availability of a robust ecosystem on land that allows controlling the satellite, etc. In general, frequencies must be high to be able to pass through the atmosphere and not be reflected by it. For example, current GEO satellites use the Ka band (26.5–40 GHz), which, on the other hand, is less congested than the C (3.7–6.4 GHz) and Ku (12–18 GHz) bands. This means that for fixed communications satellite services, frequencies ranging from 19.7 to 21.2 GHz are used for transmissions from the satellite to the terrestrial terminal and from 29.5 to 31 GHz in the link from the user to the satellite. In the case of mobile satellite communications, the L band (1.5–2.5 GHz) is used, as it suffers less attenuation and requires lower antenna gain in the user terminal. However, if the mobile is an airplane or a ship, which allows parabolic antennas of considerable gain, the Ka band is the one used. A consequence of direct vision communication is the importance of using highly directive antennas, which is achieved by transmitting at frequencies above 10 GHz, since the higher the frequency, the smaller the antenna size has to be to achieve high directivity. At these frequencies, in a transmission by direct vision, the medium generally does not change the polarization of the transmission, and this represents advantages when designing certain modulations (Henarejos and Pérez-Neira 2015). It is interesting to mention that these direct vision communication characteristics, high propagation losses, and conservation of polarization in transmission are shared by communications in THz, which are beginning to be studied for future generation mobile communications. It would be good if, in this regard, these new systems take advantage of their synergies.

6.3.3.4

User Terminal

As the signal-to-noise ratio is low, the user terminal must have high sensitivity (that is, little thermal noise) and high-gain antennas (this justifies the parabolic antennas that we see on balconies or roofs). If the satellite is also not GEO, the user terminal should continuously monitor said satellite while it is in its field of vision. This way your antenna can continuously point it at its maximum gain.

6.3.3.5

Limited Processing

The restrictions of power consumption, weight, and cost of the system place high limitations on the processing capacity on board the satellite.

6 Satellite Communications Toward a Sustainable 3D Wireless Network

6.3.3.6

179

Time Variant Coverage

The coverage of a GEO satellite is static, with little change in the pointing of the satellite beam to compensate for the movement of the satellite. In contrast, the movements of non-GEO satellites, especially LEO satellites, lead to coverage that varies in time and space. A typical LEO satellite is visible from the same point on Earth for only a few minutes. This means that even in a LEO system with fixed beams, the satellites that are visible from the same point keep changing. This variability has a great impact on the management of the mobility of the user terminal within a communication system.

6.3.3.7

Propagation Delay

In land mobile communications, rapid interaction between a user terminal and the base station is possible due to the short distances, which are usually on the order of 1 ms. This is in contrast to the high delays in satellite communications, since the distances are much greater. The propagation time of a radio wave between the Earth and a GEO satellite is 250 ms. For a LEO satellite, the time is much shorter as it has a lower altitude than the GEO, and it may range from 2 to 14 ms. On the Internet, the delay suffered by an information packet depends on its routing path and supports many services with low delay sensitivity, called “delay tolerant,” (file transfer, video transmission, etc.). These applications are suitable for the satellite to be used as another node in the network; however, for this the standard must be designed considering the maximum delays that may exist between two nodes. It is worth commenting that, although terrestrial networks introduce small delays, these are not generally controllable, as it depends on the routing of information through the complex network of nodes. On the contrary, working independently of the terrestrial network, the delay introduced by a satellite is high but known, and this can be useful for certain applications.

6.3.3.8

Doppler

The Doppler is the change in frequency that a wave undergoes due to the relative movement between the transmitter and the receiver. Its absolute value depends on the value of the carrier frequency. In non-GEO satellite communications, this relative motion is high and introduces an important component to take into account, which is non-existent in land mobile communications. The Doppler is especially pronounced in communications with LEO satellites, and it may vary over time, which requires the introduction of techniques that adequately compensate for them. For instance, at a frequency of 3 GHz and 600 km altitude, the Doppler varies ±48 kHz. This contrasts with only the ±900 Hz that a 5G system can support.

180

A. I. Pérez Neira

In the case of GEO satellite communications, it can be considered that the Doppler is non-existent.

6.3.3.9

Long Development Time and Validation Phase

In general, satellite solutions take a long time to develop and validate before they can be put into orbit and the satellite is operational. For example, from its conception a GEO satellite has required 10–15 years until it is in orbit. Not in vain, afterwards they are asked for a useful life of about 15 years. This is in great contrast to terrestrial communications, which are much easier and simpler to manufacture and test. Among other things, this landscape is changing thanks to the 3D manufacturing and the decrease in payload complexity and weight of future LEO satellites, although it is difficult to imagine a total convergence since the necessary certification phase to be able to be put into orbit does not exist in mobile communications.

6.3.3.10

Satellite Communication Standards

Everything discussed highlights the existing challenges in these communications that require solutions that are generally different from those of terrestrial communications. Currently the standards are used DVB-S2X (Digital Video Broadcasting (DVB) 2009), for broadband communications, and BGAN, for interactive mobile communications. Both are perfectly suited in wave form and in access mode to the GEO satellite communication channel, among other characteristics: they have long channel codes that allow better protection of communication from noise. The price to pay is a greater delay in communication, which in this case is imperceptible with respect to the physical delay introduced by the channel. Its transmission modes (modulations and channel code rate) take advantage of the broadcasting properties of the satellite and adapt to changes in the channel with ACM techniques. The slow variations of the GEO channel make it easier to adapt than in the case of land mobile communications. Finally, they are FDD standards, which use different frequencies to transmit from or to the satellite. For this reason, unlike terrestrial radio communications, reciprocity cannot be applied between the downlink to the satellite and the uplink, and different techniques must be used to train the receivers to estimate the channel. Given these characteristics, talking about 3D networks in which the terrestrial and satellite segments are considered equally and are big words, since it implies combining in the same standard very different propagation channels, types of terminals, power restrictions, vision windows for communication, configuration of communication cells, and, ultimately, management of radio resources that offer continuity of service to the user. 5G is making the first attempt but more at the level of high application and service layers than of physical layers, with the corresponding inefficiency that this may entail.

6 Satellite Communications Toward a Sustainable 3D Wireless Network

181

6.4 A New Communication Paradigm: 3D Joint Computing and Communication Networks Studies carried out on radio communications show that the demand for connectivity is exponential, and it is expected that in 2030 the volume of communications traffic will have grown more than 600 times compared to the traffic in 2010 (ITU-R M.2370-0 2015). Machine-to-machine communications and the Internet of Things, which started with 5G, will contribute greatly to this, along with augmented reality applications and other new services. Therefore, a new global communications system will be necessary to respond to this increased demand for connectivity. This system will use terrestrial and non-terrestrial base stations in an agile way, that is, it will integrate the vertical dimension at its various altitudes, as shown in Fig. 6.6. Communications through satellites of various orbits must accompany this increased demand. The inclusion of new dimensions in terms of altitude will create a 3D connectivity considerably different from the current 2D. Future networks will most likely integrate communications with computing and continuously measure the radio medium to react quickly, efficiently, and automatically to its changes of state (Chowdhury et al. 2020). Example of this joint communication, computing, and sensing is when, while a set of sensors transmits its measurements by radio to a data fusion center, it is possible to take advantage of the combination or sum that this medium makes of the signals. In this way, the calculation and aggregation

Fig. 6.6 3 Dimensional (3D) networks

182

A. I. Pérez Neira

of data to be done by the fusion center is accelerated. Another example is using waveforms that are useful not only for communications but also for radar. In this way, both functions, that of communications and that of taking measurements of the environment, can gain efficiency by using the same HW and the same frequency band. Only with a paradigm shift can you take advantage of all the degrees of freedom offered by a 3D network. Satellite space networks can play an important role in this new paradigm and benefit from it. Note that Earth observation is leading many sensing technologies and multifunction sensing devices (SAR, optics, etc.). In fact, it is contemplated that, in the future, satellite operators will be so for both communications and Earth observation data. However, this requires many intermediate steps and will depend on who the end customer is going to be. What is certain is that the new low-orbit constellations are a fundamental actor in this paradigm shift and will allow us to observe the Earth in a flexible and low-cost way, with more and more different types of sensors and technologies. Among other things, the large amount of new data available will make it possible to increase the spatial and temporal resolution of the images. At the same time, however, it will be necessary to have a whole series of tools to adequately develop future applications. For example, AI tools that allow the fusion with very diverse additional information, political, social, etc. Also, good communication management will be necessary to offer the required quality of service. If the Internet of Things collects measurements made both from Earth and from space, this will improve response time in emergency situations or the efficiency of certain applications. These can include, for example, precision mapping to build 3D maps in real time. In general, if we keep in mind the “Green Deal” (https://ec.europa.eu/info/strategy/priorities-2019-2024/ european-green-deal_en), undoubtedly, satellite communication plays a key role in improving the sustainability of our oceans and, in general, of our planet. The Internet is a communications network that supports many applications that tolerate delays (“delay tolerant networks”) and where information is stored in different nodes of the network. We are talking about the internet of space (https://www.teldat.com/ blog/es/internet-desde-el-espacio-satelite/) and its integration into the internet that we know today. Regarding the communications part, it may seem that this future vision of a full interaction with space is too daring if we consider what has been exposed in previous chapters. The satellite transmission channel presents greater attenuations, delay, and Doppler than the terrestrial one. On the other hand, the qualification of any device to be able to be shipped on a satellite is a long process. However, as we mentioned at the beginning, access to space has been democratized thanks to the reduction of time and the cost of developing smaller satellites. Specifically, nanosatellites and CubeSats (see Fig. 6.7) have become suitable platforms that offer a good compromise between performance and cost of production, launch, and maintenance.

6 Satellite Communications Toward a Sustainable 3D Wireless Network

183

Fig. 6.7 3 CubeSats units: 10 × 10 × 30 cm with dropdown solar panel or 24 W of power. OPSSAT contains a computer 10 times more powerful than any other ESA satellite. (https://www.esa. int/Enabling_Support/Operations/OPS-SAT) © ESA - M. Pedoussaut

6.4.1 Fundamentals to Be Revisited C. Shannon studied the maximum possible transmission of information given a communication channel. With this he established the foundations of the Information Theory, which considers communications in their asymptotic regime, that is, when the dimensions of the system go to infinity. In this way, the geometric complexity that finite-dimensional spaces can present is avoided. Practice has proven that the most efficient way to do this is to increase, as necessary, the time duration of the frame by adding redundancy to the information to be transmitted. This redundancy is what protects the transmission of errors. Shannon’s basic scheme was of a point-to-point communication degraded only by additive Gaussian noise. In this transmission, it is optimal to follow the source-channel separation theorem, that is, first encode the source information (analog or digital) to compress it to bits and then introduce the channel encoding that protects said bits against errors. It is this separate encoding that has motivated most current designs that encode digitally in bits. However, when the transmission is not point-to-point, but distributed, as it may be the case in the 3D complex networks, other strategies may be optimal. Therefore, it is useful to know other possible approaches that are not the asymptotic of Information Theory. H. Witsenhausen showed in a simple way the difficulty of solving a control/communication distributed system. Subsequently, M. Gastpar applied these control results to solve communication in a distributed sensor network. In Gastpar (2008), it is shown that in this network the optimal thing is not to transmit the signal measured by the sensors in bits. If the measurements that they take of a certain physical phenomenon are independent, and the objective is to recover with the least mean square error a certain function of these measurements, the optimal communication strategy by each sensor is to directly transmit the observed measurement. Therefore, the separate source and channel encoding that a communications engineer would have performed is not optimal. This would have converted the measurements to bits, with a source encoder, and then protected the transmission of those bits from noise with a channel encoder. In his study of sensor networks, M. Gastpar was the

184

A. I. Pérez Neira

first to conceive communication and computational problems in combination, the so-called “Over-the-air computing.” The not always binary nature that presents the optimal solution in the joint measurement, communication, and computing paradigm motivates the study of analog coding techniques. In Perez-Neira (2019), Diaz-Vilor et al. (2019), and Perez-Neira and Lagunas (2020), the authors introduced “radio frequency coding” (RFC) with the aim of expanding these studies. RFC is a methodology that consists of replicating or mimicking channel coding techniques from radio frequency or from the complex components of the baseband signal (phase and quadrature). To do this, it incorporates redundancy appropriately, just as channel coding does.

6.4.2 Enabling PHY Technologies from the Satellite Perspective A disruptive new physical layer that comprises waveforms, numerology, and coding is needed; let us call it radio frequency coding and should allow: • Joint communication and computing, as it is not always optimal to transmit the sensed data into bits Gastpar (2008). • Further evolution of network coding for a truly over-the-air computing, where AI can help toward this cross-layer design. • Integrated sensing and communications, for high precision spatial awareness without the need for feedback from the receiver, with high quality in time varying systems (e.g., Orthogonal Time Frequency Space (OTFS) Modulation, THz communications, etc.) • Going up from 50 GHz to THz: simple non-coherent modulations and polarization mod. In going up in frequency, satellite communications meet Free Space Optics (FSO). In fact, they are very suitable for Inter Satellite Links, as they can provide lower delay than over fiber communications, thus, presenting a very good alternative to terrestrial BS connection. FSO offers present excellent power efficiency and high data security. Due to this, optical electronics are being developed, with the corresponding reduced power consumption. However, challenges must be solved as atmospheric effects (i.e., requiring GS diversity) and precise beam pointing and control.

6.5 Conclusions This chapter has presented how 5G engineering will evolve toward 3D networks in where global access to the internet for all will be an important pillar, being

6 Satellite Communications Toward a Sustainable 3D Wireless Network

185

the satellite segment is key to achieve this. Future space networks will integrate communication, computing and sensing to meet the stringent throughput/latency and sustainability requirements with channel awareness and fast autonomous adaptation. Earth observation is leading many sensing technologies and multifunction sensing devices (SAR, optics, etc.), and also THz are already being used in astronomy. Therefore, the satellite is a very well-prepared ecosystem for this change in paradigm.

References 3GPP TR 22.822 (2018) Study on using satellite access in 5G, V15.0.0. Available at http:/ /www.3gpp.org/ftp//Specs/archive/22_series/22.822/22822-g00.zip. Ultimo acceso en el 28 Nov 2020 3GPP TR 38.811 (2018) Study on New Radio (NR) to support non-terrestrial networks, V15.0.0. Available at http://www.3gpp.org/ftp//Specs/archive/38_series/38.811/38811-f00.zip. Ultimo acceso en el 28 Nov 2020 Artiga X, Pérez-Neira AI, Baranda J, Lagunas E, Chatzinotas S, Zetik R, Gorski P, Ntougias K, Pérez D, Ziaragkas G (2018) Shared access satellite-terrestrial reconfigurable backhaul network enabled by smart antennas at mm-wave band. IEEE Netw Mag 32(5):46–53 Azure Orbital. https://azure.microsoft.com/es-es/services/orbital/ Chowdhury MZ, Shahjalal M, Ahmed S, Jang YM (2020) 6G wireless communication systems: applications, requirements, technologies, challenges, and research directions. IEEE Open J Commun Soc 1:957–975 DARPA 2020 Blackjack. https://www.darpa.mil/program/blackjack Diaz-Vilor C, Perez-Neira AI, Lagunas MA (2019) RSBA-resource sharing beamforming access for 5G-mMTC. In: 2019 IEEE globecom workshops (GC Wkshps), Waikoloa, HI, USA, pp 1–6 Digital Video Broadcasting (DVB) (2009) Second generation framing structure, channel coding and modulation systems for broadcasting, interactive services, news gathering and other broadband satellite applications, part II: S2-extensions (DVB-S2X). European Telecommunications Standards Institute Standard ETSI EN 302:307–302 EMEA Satellite Operators Association (2020) ESOA satellite action plan for standards. White paper. Available at https://esoa.net/cms-data/positions/ 5G 1771%20ESOA%205G%20standards.pdf. Ultimo acceso en el 28 Nov 2020 ® Fragouli C, Soljanin E (2008) Network coding applications. Found Trends Netw 2(2):135–269 Gastpar M (2008) Uncoded transmission is exactly optimal for a simple Gaussian “sensor” network. IEEE Trans Inf Theory 54(11):5247–5251 Henarejos P, Pérez-Neira AI (2015) Dual polarized modulation and reception for next generation mobile satellite communications. IEEE Trans Commun 63(10):3803–3812 https://ec.europa.eu/info/strategy/priorities-2019-2024/european-green-deal_en https://www.teldat.com/blog/es/internet-desde-el-espacio-satelite/ https://www.tesat.de/ ITU-R M.2370-0 (2015) IMT traffic estimates for the years 2020 to 2030 Lin X et al (2019) 5G new radio: unveiling the essentials of the next generation wireless access technology. IEEE Commun Stand Mag 3(3):30–37 Perez-Neira AI (2019) Radio frequency coding. ICEAA IEEE APWC 2019, Granada, 7–9 September 2019 Perez-Neira A, Lagunas MA (2020) Slicing at physical layer. Accepted at ICASSP2021, http:// arxiv.org/abs/2007.07957

186

A. I. Pérez Neira

Perez-Neira AI, Vazquez MA, Shankar MRB, Maleki S, Chatzinotas S (2019) Signal processing for high-throughput satellites: challenges in new interference-limited scenarios. IEEE Signal Process Mag 36(4):112–131 Saqlain M (2019) Feasibility analysis of opto-electronic THz Earth-satellite links in the low- and mid-latitude regions. Appl Opt 58(25):6762–6769 Space Exploration Holdings [Spa20], LLC Application for Approval for Orbital Deployment and Operating Authority for the SpaceX V-band NGSO Satellite System. FCC File Number: SATLOA-20170301-00027. Available at http://licensing.fcc.gov/cgi-bin/ws.exe/prod/ib/forms/ reports/swr031b.hts?q_set=V_SITE_ANTENNA_FREQ.file_numberC/File+Number/%3D/ SATLOA2017030100027&prepare=&column=V_SITE_ANTENNA_FREQ.file_numberC/ File+Number Shannon CE (1949) Communication in the presence of noise. Proc IEEE 86(2):447–458, February 199 (Reprinted from Proc. of IRE, Vol. 37, No.1, pp.10–21 Suresh V Non-GEO satcom from transition to affirmation. https://www.nsr.com/non-geo-satcomfrom-transition-to-affirmation Van der Breggen R (2017) Why leo are key for telecom operators in search of growth. SatelliteEvolution ptWorldVu Satellites Limited OneWeb Ka-band NGSO constellation FCC File Number: SAT-MOD-20180319-00022. Availexpansion, able at http://licensing.fcc.gov/cgi-bin/ws.exe/prod/ib/forms/reports/ swr031b.hts?q_set=V_SITE_ANTENNA_FREQ.file_numberC/File+Number/%3D/ SATMOD2018031900022&prepare=&column=V_SITE_ANTENNA_FREQ.file_numberC/ File+Number. Ultimo acceso en el 28 Nov 2020 Yeung RW, Zhang Z (1999) Distributed source coding for satellite communications. IEEE Trans Inf Theory 45(4):1111–1120

Ana Pérez-Neira ([email protected]) is full professor at Universitat Politècnica de Catalunya in the Signal Theory and Communication department since 2006 and was Vice rector for Research (2010–14). Currently, she is the Director of Centre Tecnològic de Telecomunicacions de Catalunya, Spain. Her research is in signal processing for communications, focused on satellite communications. She has more than 60 journal papers and 300 conference papers. She is co-author of seven books. She has leaded more than 20 projects and holds eight patents. She is the coordinator of the Networks of Excellence on satellite communications, financed by the European Space Agency: SatnexIV-V. She has been associate editor of the IEEE TSP and EURASIP SP and ASP. Currently she is senior area editor of IEEE OJSP. She is a member of the BoG of the IEEE SPS and Vice-President for conferences (2021–23). She is a IEEE Fellow, a EURASIP Fellow, and a member of the Real Academy of Science and Arts of Barcelona (RACAB). She is recipient for the 2018 EURASIP Society Award, and she has been the general chair of IEEE ICASSP’20 (the first big IEEE virtual conference held by IEEE with more than 15,000 attendees). In 2020, she has been awarded the ICREA Academia distinction by the Catalan government. You can find some highlights of her bio in the interview at: https://signalprocessingsociety.org/newsletter/2019/10/serieshighlight-women-signal-processing-ana-isabel-perez-neira

Chapter 7

Integrating AI into Radar System Design: Next-Generation Cognitive Radars Sevgi Z. Gurbuz, Kristine L. Bell, and Maria S. Greco

7.1 Introduction Most radar systems today operate by transmitting a fixed, pre-defined waveform via a fixed antenna beam pattern regardless of any dynamic behavior of the target or the environment. This would be akin to a person constantly gazing in the same direction without regard to any information acquired from those observations that would warrant a change in tactic. Put simply, the motivation for the design of cognitive radar systems that incorporate artificial intelligence (AI) in their sensing process is to make our own man-made sensing systems operate more like the way we as humans use our own senses—via a process that uses feedback from current observations to guide and optimize how we sense and interpret what we sense in the future. AI is essential to this process as it provides the basis for emulating the human cognitive processes that form the basis for decision-making and reasoning. The cognitive neuroscientist Dr. Joaquin Fuster (2003) has posited that there are five essential cognitive processes: (1) The perception–action cycle (PAC): The PAC constitutes a dynamic, closed feedback loop, where the radar seeks to optimally control the transmissions

S. Z. Gurbuz () University of Alabama, Tuscaloosa, AL, USA e-mail: [email protected] K. L. Bell Metron, Reston, VA, USA e-mail: [email protected] M. S. Greco University of Pisa, Pisa, Italy e-mail: [email protected] © The Author(s), under exclusive license to Springer Nature Switzerland AG 2023 M. S. Greco et al. (eds.), Women in Telecommunications, Women in Engineering and Science, https://doi.org/10.1007/978-3-031-21975-7_7

187

188

(2)

(3)

(4)

(5)

S. Z. Gurbuz et al.

interrogating environment, while the receiver continuously learns information critical to control and perception via interaction with the environment. Memory: Memory plays a critical role in human cognition and cognitive radar systems, as it is needed to enable learning from prior knowledge and experiences. Examples of prior knowledge include physics-based models, maps of scene-specific information, and target signatures, either measured or learned from prior data. Memory relates to not just the physical structures upon which data is recorded, but an intelligent way of accessing and reprocessing the data when needed. Human memory has served as inspiration here too, where mechanisms for short-term versus long-term memory have been proposed (Haykin 2012). Attention: A capability that closely relates to memory and learning is attention; namely, the ability to focus on or enhance a smaller, but important, subset of the data relative to other aspects of the data that may not be as necessary for the decision-making process. Intelligence: If learning is defined as the process of acquiring new knowledge from the environment, intelligence may be best framed as the ability to learn and apply suitable techniques for solving problems. Although what truly embodies both learning and intelligence may be subject to debate, with some describing this loftily as emulation of human intelligence and other quite literally equating these to optimization, the former represents a long-term vision, while the latter merely states the nature of current execution. Fundamentally, intelligence requires goal-directed behavior that leads to problem solving. The solutions to well-defined problems are more easily formulated, whereas open-ended complex problems may require more complex processes. Language: Human language is complex construct for communication of essential information; in cognitive radar systems, the analogue would be a means by which the system could communicate with the operator through a human– machine interface. An operator (presumably human) needs to effectively relay objectives and mission requirements, while the cognitive radar system provides the necessary information to justify the decisions that the radar system takes; otherwise, the operator will not trust the radar. Communications may also be needed between the radar system and other sensor systems and platforms. Networked cognitive systems can potentially offer great benefits in terms of robustness as well as improved performance.

While Fuster’s perspective on human cognitive process modeling has been perhaps the most influential in current work pertaining to cognitive radar design, the definitions of cognitive processes can vary, and the reader is encouraged to read broader works on cognitive psychology for more insight and inspiration. Formally, in the IEEE Standards, cognitive radar is defined as “a radar system that in some sense displays intelligence, adapting its operation and processing in response to a changing environment and target scene. In comparison to active radar, cognitive

7 Integrating AI into Radar System Design: Next-Generation Cognitive Radars

189

radar learns to adapt operating parameters as well as processing parameters and may do so over extended time periods” (Gurbuz et al. 2019). But within the scope of this definition, one may specify varying degrees of cognition, just as humans possess varying degrees of intelligence. Bloom’s taxonomy is one way of visualizing a hierarchical description of human cognitive capabilities (Bloom et al. 1956), defining six different levels as Representation (the lowest), to Understanding, Applying, Analyzing, Evaluating, and Creating (the highest). This hierarchy closely parallels that of automation in the related field of robotics. In fact, one of the earliest expressions of a vision for cognitive radar was given in 2003, with the introduction of the “Sensors as Robots” (Wicks 2006) concept: “As more knowledgeable and proven techniques are obtained, radar systems will begin to function as robots. . . the final step will be autonomous operation of these sensors under the intelligent robot paradigm.” Over the years, several hierarchical frameworks for cognitive process modeling have emerged. Rasmussen (1983) described human behavior in terms of three levels: skill-based, rule-based, and knowledge-based. Skill-based behavior described subconscious yet efficient PACs, which, according to Brüggenwirth et al. (2019), map to basic signal processing and generation units in a radar system. Rule-based behavior is applied by humans in familiar situations, and although consciously controlled, the action is reactive and thus results in procedures that have been learned over time. In radar, the parallel operation would be the procedures that have been pre-stored or hard-coded, based upon offline simulations and the analysis of prior experience. The highest layer is the knowledge-based layer, in which solutions to problems that arise in unfamiliar situations are derived using knowledge-based deliberation. A more fine-scale characterization of radar cognition has recently been proposed (Horne et al. 2018) based on assessment of degree of sophistication of three properties: planning, decision mechanism, and memory. Planning sophistication is divided into two levels, myopic-adaptive and non-myopic. For example, a system that plans only one step into the future is adaptive, but myopic because it involves prediction over a short time scale. As a radar is able to predict further and further into the future, non-myopic planning sophistication is achieved. Memory sophistication is divided into three levels: fixed (based on an internal knowledge database), dynamic (as updated by an external agent), or online learning capable. Finally, the degree of decision mechanism sophistication is also divided into three levels: rule-based, heuristic (which provides more flexible responses than simple rule implementation), or full optimization, which aims to compute the optimal response given all available information. Based on these levels defined in a three-dimensional cognitive modeling space, ten different levels of cognition can thus be defined, ranging from minimally fully adaptive (Level 1) to fully cognitive (Level 10), as listed in Fig. 7.1. Based on this hierarchy, an air traffic control system having a look as part of its decision-making process would be categorized in Level 1. In contrast, the leading cognitive radar experimental testbed, CREW (Smith et al. 2015), is capable of adapting radar parameters in real

190

S. Z. Gurbuz et al.

Fig. 7.1 Levels of cognition (Horne et al. 2018)

time to maintain a specified signal-to-noise ratio (SNR) on target. This system is more advanced in learning and decision-making, but limited to adaptive planning, resulting in one assessment as planning Level 2, learning Level 4, and decisionmaking Level 7 (Horne et al. 2018). Over the past 20 years, there has been a wide range of research on a variety of techniques to increase the level of cognition in practical radar systems. The proposed techniques draw on prior advancements in Bayesian decision theory, information theory, decision theoretic approaches (including fuzzy logic, rule-based systems, metaheuristic algorithms, and Markov decision processes), dynamic programming, optimization (e.g., maximization of SNR), convex optimization, minimization of the Cramer–Rao Lower Bound (CRLB) on estimator variance), deep learning, and game theory. The applications for which cognitive radar has been proposed are similarly quite broad, ranging from detection, estimation, localization, tracking and radar resource management to radar networks, passive radar, SAR imaging, spectrum sharing, command and control, over-the-horizon radar, electronic warfare, remote sensing, health, and transportation. While a comphrehensive examination of all aspects of cognitive radar techniques and applications is beyond the scope of this chapter, we would like to refer the readers to some books and review papers Guerci (2010a), Greco et al. (2018), Gurbuz et al. (2019) that have been recently published for more detailed listing and assessment of recent literature. Instead, this chapter will present foundational concepts pertaining to cognitive radar architectures and design (Sect. 7.2), focusing on the enabling hardware components and real-time signal processing requirements of actual cognitive radar systems, two of the most heavily utilized approaches to cognitive process modelling, namely the Bayesian approach (Sect. 7.3) and deep learning (Sect. 7.4), and two of the most studied cognitive radar applications (Sect. 7.5), namely multitarget tracking/classification and multitarget detection.

7 Integrating AI into Radar System Design: Next-Generation Cognitive Radars

191

7.2 Cognitive Radar Architectures The PAC cycle described by Fuster has become widely adopted in the radar community and serves as the basis for many proposed cognitive architectures. Haykin (2006) described a cognitive radar system that implemented a PAC as having three essential components: (1) intelligent signal processing, which builds on learning through interactions of the radar with the surrounding environment, (2) feedback from the receiver to the transmitter, and (3) preservation of the information content of the radar returns by a radar scene analyzer. Thus, any PAC implementation should include a dynamic closed feedback loop so that the transmit waveform and antenna can be optimally controlled according to a given task. A PAC whose key components have been labeled with their analogs in the radar perception problem is given in Fig. 7.2. The fundamental challenge to radar perception is the dynamic nature of real-world environments, where target behavior, potential occlusions of target returns, radio frequency (RF) interference due to a congested spectrum, and ground clutter all result in factors that may degrade radar performance. While a conventional, fixed transmission radar may only transmit a waveform optimal for one environmental scenario, a cognitive radar can adapt its transmissions to optimize performance based on perceived dynamics. The first essential block of “perception” in the PAC corresponds to the physical process of acquiring sensor measurements, in this case, the radar backscatter from the environment (and target). After the radar returns have been initially processed via radar signal processing, a variety of learning algorithms may be applied to extract needed information from the data. Learning algorithms can exploit not just prior measurements, but also any knowledge that has been stored. The information computed may relate to metrics indicative of the dynamic state of the environment, target behavior, or other indicators relevant to the decision-making process. A cognitive radar must decide the optimal action to take based on the task at hand. The degree of adaptivity of the transceiver and antenna hardware itself determines the scope of potential actions. One possible cognitive radar system architecture is depicted in Fig. 7.2, where both the transceiver and receiver are able to adapt to any desired transmission, and adaptivity at the antenna is enabled via multi-functional reconfigurable antenna (MRA) or other similar controllable antenna. Thus, posssible actions may include changing the antenna beam pattern, polarization, frequency, bandwidth, or waveform of the transmissions and accordingly may require the use of adaptive transceiver components, such as adaptive filters or amplifiers, and may be further facilitated with the use of piezo-electric materials and metamaterials. While this architecture separately shows the functions of the decision center, learning algorithms, and radar signal processor, the implementation of these functions may be co-located on an appropriate digital signal processor equipped with field programmable gate arrays (FPGAs) or graphical processing units (GPU) on an edge computing device. It is also important to note that the cognitive radar system architecture also enables important innovations in multi-modal sensing, where signal-level sensor integration can be made possible by exploiting the data

192

S. Z. Gurbuz et al. Perception – Action Cycle

Cognitive Radar System Architecture Waveform Generator/Database

Sensor Measurements Moving targets Occlusion Interference Clutter Congested Spectrum

Dynamic Environment

Perception

Record “knowledge” Passive Sensors CONVENTIONAL

Learning Store and Retrieve Knowledge

COGNITIVE

Action New Transmit Waveform, Antenna Beam Pattern, Change Sensor Position, etc.

Resource Management; Control of Adaptive Hardware

Transmitter

Controller

Signal Processing Data Science

MRA

Controller Decision Center Controller

(If possive sensor users MRA)

Learning Algorithms

Feedback

Radar Signal Processor

Learning Algorithms

Receiver

MRA

(Number of TX and RX nodes can be extended to any multistotic topology desired)

Database

Fig. 7.2 A perception–action cycle (left) and related cognitive radar system architecture (right)

from not just the cognitive radar, but also from other sensors to be used as inputs to the decision-making process guiding control of the cognitive radar sub-systems.

7.2.1 Real-Time Processing and Online Machine Learning A fundamental requirement of any cognitive system is real-time processing and machine learning supporting its decision processes to enable closing-the-loop to the transceiver. There are three main components representing a bottleneck in latency: (1) the computation time required for initial processing of the raw data, converting it into a suitable data representation so as to enable application of machine learning algorithms on the data; (2) the computation time required for making predictions with a deep neural network (DNN) or other machine learning algorithm; and (3) the computation time required for large-scale, high-dimensional optimization. Experimental studies have shown (Adeoluwa et al. 2021) that in most cases it is the initial processing of the data, which may involve tasks such as the computation of time–frequency representations or other Fourier transform-based analysis, that takes up the most time. The prediction time for DNNs is relatively short once the model has been trained. Consequently, this has led to an increased interest in end-toend learning solutions, which aim to jointly optimize the data to decision pipeline. Several different approaches can be taken toward this aim: (1) the development of complex neural networks and activation functions to enable the development of DNNs that can directly take as input the raw radar complex I/Q data, thereby skipping over the need for computationally complex pre-processing algorithms prior to input to a DNN; (2) the development of DNNs that can functionally replace the role of Fourier transforms in radar signal processing, allowing for joint training and optimization of a multi-stage DNN that accomplishes both preprocessing and prediction; and (3) joint optimization of the sensing, reconstruction, and classification steps via compressed sensing. The idea of replacing fundamental signal processing steps with DNNs is not a new idea, but faces practical challenges in realistic radar sensing scenarios, where there is typically significant clutter and

7 Integrating AI into Radar System Design: Next-Generation Cognitive Radars

193

the resulting SNR is low. Thus, the challenge of real-time processing and machine learning remains an open challenge that will require advances not only in edge computing hardware and high-performance computing, but also end-to-end learning to be adequately addressed. Even if the latency due to initial processing is addressed, however, there remains an issue of latency for model predictions. Although this computation time is much shorter than that of initial processing, a more fundamental challenge exists: the prediction time is only short for pre-trained models. If a model must be updated, the time required for batch re-training is prohibitively high. Incremental online learning must be utilized to update models with new knowledge learned from prior measurements. Indeed, this process is necessary not just for closing the sensing loop, but also to ensure true learning. However, there is an inherent trade-off between prior knowledge and new data, known as the plasticity–stability trade-off (De Lange et al. 2022). The initial model is batch trained based on prior knowledge. This can be based on a database of prior measurements or synthetic samples generated from physics-based models. Data used in batch training are typically well-designed datasets carefully selected to represent as best as possible expected target signatures. The problem, however, is that our models or prior data may not match well with the target as acquired in the current environmental conditions, which are often dynamic in nature. The new data acquired provide an opportunity to learn current, environment-specific aspects of the data to potentially improve the model. But as more and more new data are incorporated into the incrementally updated model, the training based on prior data is forgotten. This phenomenon is known as catastrophic forgetting. On the other hand, how reliable are the new data? Preservation of certain aspects of batch training may be essential to the overall stability of the system and protecting the generalizability of the model. Ultimately, while online incremental learning is an essential approach to enable model update in a computationally feasible way, these trade-offs present a challenge to the long-term efficacy and performance when utilized for cognitive radar, and hence remain another open research area. Finally, most decision-making processes rely on optimization theory and techniques, which may require searching in multi-dimensional, continuous spaces. Better techniques for reducing dimensionality to smaller, discrete searches are needed. New algorithms suitable for the optimization of radar parameters and radar resources over large systems models are necessary, including techniques that can provide near-optimal solutions on the timeline of a typical radar processing interval.

7.2.2 Wideband and Tunable RF Components An important enabling capability of cognitive radars is the ability to adaptive change transmit frequency and bandwidth. This is especially important in the context of spectrum sharing, where the congested nature of the spectrum requires spectrum agility and tunability to avoid performance degradation due to RF interference.

194

S. Z. Gurbuz et al.

Spectrum agility is a term that refers to the ability to tune the cognitive radar’s center frequency and instantaneous bandwidth, while tunability refers to the system’s ability to shape its frequency response. Thus, increasing the bandwidth of current wideband mixers, filters and amplifiers will add to the capabilities of cognitive radar and their ability to effectively respond to adverse dynamics. Not just the transceiver, but also the antenna operating range must fully cover the range of operation. While filter banks could be used to switch to a suitable passband at any given time, true frequency agility would require an enormous amount of filters to cover several GHz operating ranges with varying instantaneous bandwidth. Thus, continued improvement of tunable filters, amplifiers, and antennas will be essential to obtain frequency agility while maintaining power efficiency, low-noise figure, and maximum suppression of out-of-band interference. The speed with which components can respond to match the desired instantaneous spectrum, as well as compactness and efficiency, must also be improved.

7.2.3 Adaptable Radar Antenna Arrays Innovations in antenna array design are also paving the way for the additional capabilities needed for cognitive radar. In particular, advances in multi-functional reconfigurable antennas and all-digital radar arrays have significance for cognitive radar systems. Currently, antenna element properties (i.e., operation frequency bandwidth, polarization, and radiation pattern) are fixed in the initial design and cannot be changed, which limits the optimal design space that could be explored for significant performance improvement. In contrast, multi-functional reconfigurable antennas enable a single antenna to perform multiple functions, where the frequency, polarization, and radiation pattern can be dynamically modified, thereby introducing important additional degrees of freedom in the cognitive system (Gurbuz et al. 2020a). Multiple MRAs used in an array fashion constitute a new class of antenna array, i.e., MRA array (MRAA) (Towfiq et al. 2018), which enables the selection of the optimum antenna properties (frequency, polarization, and radiation pattern) in response to the time-varying channel conditions, and helps retain the best systemlevel performances at all times. MRAAs can potentially bring the controllability of a large, expensive, military-grade phased-array antenna (PAA) systems, but with a small size and low power consumption. This is important because although PAAs, when used in conjunction with adaptive signaling, are capable of steering their beam toward desired directions, while simultaneously placing nulls toward undesired directions of interferers, this capability comes with high complexity and cost associated with a large number of single-function legacy antennas each having a phasing network. Due to the increased size, complexity, and high price, the use of PAAs has thus far been limited to sophisticated military and space systems. In contrast, MRAAs can alleviate the practical drawbacks of the current PAA technology, while being sufficient small and compact (see Fig. 7.3) to be effective alternatives in applications of communications, automotive sensing, and

7 Integrating AI into Radar System Design: Next-Generation Cognitive Radars

195

Fig. 7.3 Picture of actual MRA (top left), examples of MRA modes (bottom left, and architecture of an MRAA). Images courtesy of Dr. Bedri Cetiner (Utah State University)

other civilian radar applications, such as remote health monitoring and human– computer interaction. All-digital antenna arrays possess the capability for independent waveform synthesis and analog-to-digital conversion at every element of the array. Although all-digital arrays generate a large amount of data and require a significant amount of control, the potential for agility and performance is unmatched, opening up a wide design space for cognitive radar. For example, each aperture can be divided into sub-apertures that transmit their own unique waveform in an independently defined direction or operate in a MIMO fashion. Since every element receives the reflected signals due to all waveforms (assuming the signals are within the instantaneous sampling bandwidth of the digitizers), full-gain beams can be formed even if the transmit beams were spoiled or low-gain due to using a fraction of the aperture. Because element-level processing and subsequent beamforming are digital, it can be reconfigured and optimized for different applications, offering maximum flexibility with unprecedented dynamic range. This, in turn, opens up the possibility for unique space–time waveform concepts, multiple transmit beams, and full digital beamforming on receive, rendering the adaptive transmit potential for digital arrays nearly limitless. Several all-digital arrays are being developed that will become operation or available as testbeds in the near future. Some challenges to this technology include the development of suitable calibration algorithms, which are essential to achieving full performance, mitigating the dependence on the implementations of software systems, and handling the big data deluge that will result from all-digital arrays that could potentially generate terabytes of data per second. Real-time cognitive radar using all-digital arrays will require a massive amount of processing power to analyze and learn from the data, make a decision

196

S. Z. Gurbuz et al.

on optimal transmission, and then upload the desired waveforms to the digital-toanalog converter on each element. The next sections discuss the Bayesian and deep-learning-based techniques critical to the learning and decision-making processes of cognitive radar.

7.3 Bayesian Approach to Enacting Perception–Action Cycle 7.3.1 Theoretical Framework The Bayesian approach relies on prior distributions and knowledge-aided models of the environment as derived from past measurements. More fundamentally, the Bayesian approach is a form of stochastic optimization, which is a broad term for techniques that perform decision-making under uncertainty. Stochastic optimization techniques are currently deployed in a wide range of applications, and there are many communities who have established a wide variety of algorithmic strategies (Powell 2019). Stochastic optimization methods seek a policy that exploits models to map from all available information at the current time into an optimized action. As this policy is essentially a perception–action cycle, cognitive radar problems can be classed as types of stochastic optimization problems, and algorithmic strategies for finding policies from the stochastic optimization field can be applied to the design of perception–actions for cognitive radar (Charlish et al. 2020).

7.3.1.1

General Stochastic Optimization Problem Components

Problems addressed by stochastic optimization all have the fundamental components described below. System State We are interested in the state of a dynamic system, which is modeled as a random vector Xk for decision step k with realization xk ∈ X , where X is the system state space. Actions and Action Space We can select an action or actions at each decision step k, which influences the transition of the system state between time step k and k + 1. The realization of an action for decision step k is denoted ak ∈ A, where A is the action space. Exogenous Information New information is obtained at each sequential decision step. The information is modeled as a random vector Z k with realization zk . For completely observable problems, the exogenous information is the system state. For partially observable problems, the exogenous information is a noisy measurement of the system state.

7 Integrating AI into Radar System Design: Next-Generation Cognitive Radars

197

State Transition Function Between decision steps, the system evolves according to a transition function xk+1 = fX (xk , ak , wk ), where wk is a realization of the state transition noise. Due to the state transition noise, the transition can be described by the transition probability density function (PDF) p(xk+1 |xk ; ak ). Objective Function At each decision step, a reward or cost is encountered, which is described by the function ro (xk , ak , zk+1 ).

7.3.1.2

Partial Observability

A common aspect of cognitive radar problems is that the system state is only partially observable through noisy measurements. Therefore, uncertainty is present not only in uncertain state transitions but also through uncertain measurements. Consequently, the components described above can be extended and adapted to the more specific partially observable case. Measurements and Measurement Space The exogenous information can now be characterized as a noisy observation of the system state. The random vector Z k can be defined more specifically as a measurement with realization zk ∈ Z, where Z is the measurement space. Measurement Likelihood Function Measurements are related to the system state through the measurement function zk = h(xk , ak , vk ), where vk is a realization of the measurement noise. Due to the measurement noise, the measurement process can be described by the measurement likelihood function, which is the conditional PDF denoted by L(xk |zk , ak ) ≡ p(zk |xk ; ak ). Information State Since the system state is not observable, it is necessary to decide on an action based on the information state. The information state is the set of actions and measurements that have occurred prior to the current decision step. The information state for decision step k is denoted Ik = (a0 , z1 , . . . , ak−1 , zk ). This information state grows with each time step, i.e., Ik = Ik−1 ∪ (ak−1 , zk ). Belief State Since the cardinality of the information state grows with each time step, it is generally undesirable to be used as the perception upon which actions are decided. Instead, decisions can be based on a belief state. The belief state is a set of parameters with fixed cardinality that characterize the posterior PDF of the system state. The belief state at decision step k is modeled as a random vector B k with realization bk . The belief state is a statistic of the information state; ideally, it would be a sufficient statistic, i.e., p(xk |Ik ) ≡ p(xk |bk ). For example, under linear Gaussian assumptions, a sufficient statistic of the information state is the mean and covariance of the posterior PDF. Typical belief states are parameters of a Gaussian distribution, a Gaussian sum distribution, or a set of particles that characterize the posterior PDF. Belief State Transition Function Analogous to the system state transition function, it is necessary to define a transition function for belief state. This transition

198

S. Z. Gurbuz et al.

function is denoted bk+1 = fB (bk , ak , zk+1 ). As the belief state can be thought of as parameters of the posterior PDF p(xk |bk ), the belief state transition function represents the standard Bayesian prediction and update steps. Objective Function Since an optimizer for a partially observable problem deals in belief states, it is necessary to specify an objective function as a function of the belief state, i.e., r(bk , ak , zk+1 ). This is the reward that is associated with the generation of a measurement zk+1 when the belief state was bk and action ak was taken.

7.3.1.3

Finding the Policy

The general objective is to find a policy that determines a feasible action based on the belief state. The policy is denoted ak = Aπ (bk ), where π carries information about the type of function and its parameters. As the belief state represents a perception of the system state, the policy can be thought of as the perception–action cycle for the system. We wish to maximize rewards or minimize costs over a time horizon comprising H future decision steps. The expected reward achievable over the current and future decision steps that originate from the current belief state is termed the value of the belief state and is denoted as VHπ (bk ). It is the expected value of the summed rewards with respect to the set of future measurements {Z k+1 , . . . , Z k+H }, conditioned on the belief state bk . The optimal policy is found by maximizing the value of the belief state over all possible actions. Solving for the optimal policy function is generally intractable; therefore, the majority of stochastic optimization approaches focus on approximate solutions to the optimal policy function. There are simplifications that can drastically reduce the complexity of the problem but result in a solution that does not fully consider the uncertainty in the problem. If the time horizon is taken as a single step, i.e., H = 1, then the problem of evaluating the impact of the action on expected future rewards is removed. This approach is known as myopic or greedy as it focuses on the immediate expected reward and ignores the impact of potential future rewards. A further common simplification is to perform myopic optimization and to evaluate the myopic reward by replacing the random variables in the problem with their expectations. Where myopic optimization ignores the propagation of uncertainty into the future, replacing the random variables with their expectations ignores the uncertainty altogether. However, the optimization problem can now be treated as a deterministic optimization problem that can be easier to solve.

7.3.2 Application to Cognitive Radar A representative set of cognitive radar problems for different applications including target detection, tracking, and classification can be found in van Keuk and Blackman (1993), Kershaw and Evans (1997), Chong et al. (2008), Sira et al. (2009), Haykin

7 Integrating AI into Radar System Design: Next-Generation Cognitive Radars

199

(2012), Charlish and Hoffmann (2017), Mitchell et al. (2018), John-Baptiste et al. (2022), Guerci (2010b), Aubry et al. (2014), Stinco et al. (2016), Martone et al. (2018), Goodman et al. (2007), Bell et al. (2019), Kreucher et al. (2005), Bell et al. (2015), Charlish and Katsilieris (2017), Bell et al. (2018) and Bell et al. (2021). These problems can be characterized as stochastic optimization problems that possess the framework components described above.

7.3.2.1

Framework Components

In target detection, the system state is the state of the clutter, interference, and noise environment, while in target tracking and classification, the system state is the target state and may also include the environment state. In the case of target tracking (van Keuk and Blackman 1993; Kershaw and Evans 1997; Chong et al. 2008; Sira et al. 2009; Haykin 2012; Charlish and Hoffmann 2017; Mitchell et al. 2018; John-Baptiste et al. 2022), the belief state characterizes a posterior PDF defined on the system state space. Typical belief states are the mean and covariance matrix of the distribution or a set of particles. The belief state transition function incorporates the Bayesian prediction and update processes. Often, the likelihood function is a Gaussian approximation of the true measurement errors. Adaptive tracking (van Keuk and Blackman 1993; Charlish and Hoffmann 2017) methods select actions in the form of revisit interval times as well as the waveform energy for the next measurement, in order to minimize resource usage while maintaining track. An early approach (van Keuk and Blackman 1993) was to use a function that mapped measurement and track accuracies, and Singer maneuver parameters to a revisit interval time. This can be thought of as an empirically derived policy function approximation. Another strand of work has focused on waveform selection and adaptation (Kershaw and Evans 1997; Sira et al. 2009; Haykin 2012), whereby the action space comprised different waveform modulations that were selected in order to minimize track RMSE. The framework components are easy to identify for tracking problems because the framework is essentially an extension to the standard Bayesian tracking process. However, other radar functions and applications can also be cast into the framework. In target detection (Guerci 2010b; Aubry et al. 2014; Stinco et al. 2016; Martone et al. 2018), typical belief states include the clutter, interference, and noise covariance matrix or a posterior distribution on a spectrum occupancy state. For imaging and classification (Goodman et al. 2007; Bell et al. 2019), the belief state characterizes a posterior probability mass function. Typical belief states are the pairwise likelihood ratios or the posterior probabilities themselves. Some works also consider a combination of radar functions (Kreucher et al. 2005; Bell et al. 2015; Charlish and Katsilieris 2017; Bell et al. 2018, 2021), and the system and belief states are combinations of the detection, tracking, and classification states. Generally, the action space is some set of parameters that characterize the radar transmission and reception, including transmit and receive sensor selection and scheduling, transmit frequency, bandwidth, time, duration, power, and waveform

200

S. Z. Gurbuz et al.

design. The exogenous information is some noisy function of the system state that maps to radar measurements; thus the system state is partially observable. Generally, objective functions differ widely.

7.3.2.2

Objective Functions for Cognitive Radar

The exact form of the objective function r(bk , ak , zk+1 ) is crucial, as it must accurately represent the physical problem to be solved in order to achieve a good performance. Specification of objective functions for cognitive radar is an ongoing research area, with existing approaches loosely fitting into the categories of task, information, or utility [quality-of-service (QoS)]-based approaches; however, the separation between the categories is not always distinct and existing approaches form more of a continuum. Task-Based Objective Functions Task-based objective functions calculate the cost or reward of an action in terms of a measure that is specific to the task being performed. Relevant task-based metrics include radar timeline or spectrum usage, probability of target detection, detection range for an undetected target density, tracking root mean square error (RMSE), track sharpness, track purity, track continuity, and probability of correct target classification, to name a few. Each taskbased metric can be regarded as some function q(bk , ak , zk+1 ) that is combined in some way to produce a scalar objective function. It is often the case that a desired task-based metric is difficult to calculate and is replaced by a surrogate metric such as signal-to-interference plus noise ratio (SINR) or an information-theoretic metric. Information-Theoretic Objective Functions A second class of objective functions used in cognitive radar and related fields is based on information theory. Broadly speaking, an information-theoretic objective function gauges the relative merit of a sensing action in terms of the information flow it provides. A primary motivation for information-based objective functions is the ability to compare actions that generate different types of knowledge (e.g., knowledge about a target class versus knowledge about target position) using a common measuring stick. A review of the history of information metrics in this context is provided in Hero et al. (2007). Here, we highlight some of the most commonly used objective functions. The most basic information-theoretic objective function is the posterior Shannon entropy. A related approach computes the information gain between the prior and posterior densities rather than just the information contained in the posterior density. The most popular approach uses the Kullback–Leibler divergence (KLD), which has several desirable properties including its connection to mutual information. A third approach specific to parameter estimation is the Fisher information matrix (FIM) and related Bayesian information matrix (BIM) (Van Trees and Bell 2007), which characterize the amount of information that a distribution contains about individual parameters (such as target position or velocity). The inverse of the BIM is the Bayesian Cramer–Rao Lower Bound (BCRLB), which quantifies the uncertainty in the parameter estimates. The (square root of the) BCRLB has the property that it

7 Integrating AI into Radar System Design: Next-Generation Cognitive Radars

201

is in the units of the parameter being estimated and is a lower bound on the RMSE. Thus it is often used as a surrogate for the RMSE and categorized as a task-based metric. The BIM also has a connection to mutual information. Utility- and QoS-Based Objective Functions QoS approaches (Charlish and Hoffmann 2017) differ from task- or information-based objective functions in that they optimize the user or operator satisfaction that is derived from a task. A utility function is defined on the task quality space that should accurately describe the satisfaction that is derived from the different possible task quality levels. This approach is very valuable in the context of radar resource management as it enables a radar with limited resources to optimize multiple tasks based on the task quality levels that are required by the mission. Mapping the quality levels of differing radar tasks into the common utility space enables trade-offs between tasks evaluated using differing quality metrics. The global utility across the multiple tasks is typically formed by taking a weighted sum of task utilities. When considering the resource usage, a resource function g(bk , ak ) can be combined with the utility function to produce the final objective function. This QoS conceptual approach can also be identified under different names (Mitchell et al. 2018).

7.3.2.3

Solution Methodologies

The majority of the reference works formulate myopic optimization problems; some further simplify the problem by using deterministic optimization. The cognitive radar methodologies in the reference works generally attempt to solve an optimization problem online by performing numerical optimizations or searches over the action space. In John-Baptiste et al. (2022), a neural network is used to learn the policy function that an optimizer with more complexity would generate.

7.4 Neural-Network-Based Cognitive Process Modeling Deep neural networks (DNNs) are a topic that has recently garnered great attention due to its success in computer vision and natural language processing, which have allowed it to achieve previously unattainable advancements for challenging problems. Inspired by the structure of the human neural system, neural networks consist of layered connections of artificial neurons called nodes. Each node receives information from its incoming connections, forms a weighted linear combination its input and a bias, and afterward applies a non-linear transformation through utilization of an activation function. In a feed-forward neural network, information moves in just one direction: from the inputs, through the nodes, to the outputs. The width of the network is determined by the number of nodes in a layer, while the depth is determined by the number of layers in the network. This interconnections provided by DNNs have proven quite effective in capturing quite

202

S. Z. Gurbuz et al.

complex representations between inputs and outputs, as informed by the data provided to the network. The process of training the neural network is accomplished by finding the optimal set of weights that minimizes a pre-defined loss function. Thus, DNNs are an important method for automatically learning features from data, and using these features to understand newly acquired data samples in terms of conceptual categorization. The ability of DNNs to learn relationships based on data alone renders it a powerful tool in the context of cognitive process modeling. However, it should be noted that deep learning (DL) via neural networks is a subset of machine learning (ML), which is an important component of data science and seeks to use statistical methods to gain insights from data that can drive greater understanding of a problem or drive decision-making within a system. AI is often conflated with ML; but, AI encompasses a much greater vision of biomimetic design, where next-generation algorithms would learn just the way humans learn, eventually achieving equal or even greater accuracy and performance relative to human capabilities. Thus, while ML is important to achieving this vision, utilization of ML or DL in itself does not imply that AI or true cognitive processing has been achieved. In other words, while a cognitive system may utilize DNNs, a system that employs DNNs may not necessarily be cognitive.

7.4.1 Principle Types of Neural Networks While a comprehensive survey of deep learning would be beyond the scope of this chapter, familiarity with some of the more commonly utilized DNNs and the philosophy behind their design is essential when considering the question of how DNNs can be utilized in cognitive process modeling (Smith et al. 2020). Convolutional neural networks (CNNs) are one of the most popular DNN architectures, which use spatially localized convolutional filtering to capture the local features of input images. While basic features, such as lines, edges, and corners, are learned in the initial layers, more abstract features are learned in the deeper layers. CNNs are typically comprised of three types of layers: convolutional, pooling, and fully connected layers. A max pooling layer typically follows each convolutional layer, in which local maxima are used to reduce computational complexity in the forward layers and add translational invariancy to the network. Fully connected layers are formed by a basic interconnection of nodes that compute non-linear combinations of the features extracted from the previous layers. CNNs rely on an extremely large amount of labeled training data for supervised optimization of network weights. The objective function of a CNN is highly non-convex; thus, one of the key challenges in training a CNN is initializing the network such that the local minimum of the loss function is as close as possible to the global minimum. Typically, CNNs are randomly initialized, relying on the large training database to increase the likelihood of convergence to a good, albeit not optimum, solution. However, this can be

7 Integrating AI into Radar System Design: Next-Generation Cognitive Radars

203

especially challenging in sensing applications, such as radar-based sensing, where a large amount of data for each class of interest may not be available. In such cases, one possible solution is the use of autoencoders (AE), which use unsupervised learning to reconstruct the input at the output. Essentially, an autoencoder aims to approximate the identity operation by utilizing a symmetric encoder–decoder structure in its architecture. The weights of both the encoder and decoder are optimized such that the reconstruction error between inputs and outputs is minimized. When the decoder is removed from the network, the remaining encoder components can be fine-tuned in a supervised fashion by adding a softmax classifier after the encoder. In this way, autoencoders can be utilized for unsupervised pre-training of DNNs to minimize the requirements for labeled training data. Convolutional autoencoders (CAEs) apply this same philosophy but include convolutional and deconvolutional layers in the encoder and decoder, respectively. Because unsupervised pre-training starts the weight optimization in CAEs at values closer to the global optimum, CAEs have resulted in better performance on radar datasets, which have orders of magnitude fewer training samples than those available for image processing (several thousand versus millions). Alternatively, knowledge gained from a different domain can be exploited to initialize the weights for a DNN via transfer learning. In transfer learning, the training process is again divided into two steps, but this time the pre-training step is done in a supervised fashion using data from another domain. In radar applications, the most common form of transfer learning has involved initializing network weights using models effective in computer vision applications, such as VGGnet, AlexNet, GoogleNet, or ResNet, which have been training on massive datasets of visual images. Studies on radar datasets have shown that transfer learning can surpass the efficacy of CAEs when only truly meager amounts of data for training are available, e.g., fewer than 50 samples per class (Seyfio˘glu et al. 2017). Recurrent neural networks (RNNs) have been the principle mechanism through which sequential time-series data classification has been approached. RNNs model temporal behavior using connections between nodes that form a directed graph along a sequence. At each time step, an output is produced. Long short-term memory (LSTM) RNNs are able to model longer-term behavior through the inclusion of a memory block that consists of a cell, input, output, and forget gates. On radar datasets, it has become common to combine RNNs with CNNs that take in 2D images, e.g., micro-Doppler signatures, or 3D range-Doppler-time tensors (Wang et al. 2016; Zhang et al. 2018; Ero and Amin 2019). Bi-directional LSTMs (Li et al. 2021) or joint domain multi-input multi-task learning (JD-MIMTL) (Kurto˘glu et al. 2022) have also been proposed for sequential classification. Finally, the development of attention mechanisms (de Santana Correia and Colombini 2021; Ghaffarian et al. 2021) within deep neural networks has been an important topic of research that is especially relevant to cognitive radar design, as attention is a core cognitive process. Attention is an essential element of both human and machine intelligence as it is difficult to process large volumes of data at high rates. Attention mechanisms facilitate rapid processing of sensing inputs by selecting and focusing on the information that is more relevant to behavior

204

S. Z. Gurbuz et al.

or task to be optimized. In neural networks, attention mechanisms enable the dynamic management of information flow, features extracted from the data, and the available resources so as to optimize task-cognizant learning. There are a variety of ways to introduce attention in a DNN. In soft deterministic attention, the network considers all inputs to compute the final context vector—a high-dimensional vector representation of the sequences of inputs. In this way, the approach aims to add more context to the calculations. Hard deterministic attention, also known as stochastic attention, is similar except now the inputs included in the final context vector are randomly selected. An alternative, more often used approach in radar applications, is the use of attention mechanisms that take as input sets of features based on item-wise or location-wise relationships. Item-wise attention refers to case where the inputs are known to the model, whereas location-wise attention does not necessarily have known inputs, and hence, the model needs to deal with inputs that are difficult to distinguish. In radar applications, the most common form of attention is obtained through the development of suitable input representations (Gurbuz 2020) so that the most important information is better revealed in the images. Single-input and multiple-input attention models can be developed so that the network can perform better in images having a complex background, e.g., clutter, by focusing on targeted areas. For example, the radar raw I/Q data stream is typically converted into micro-Doppler, range-Doppler, or range-angle images, which can then be supplied a multi-input model. Various approaches for additional processing have been proposed, such as frequency-warped cepstral heatmaps (Erol et al. 2018). Other types of attention mechanisms include concatenating local-feature maps with attention feature maps that capture global dependencies (Campbell and Ahmad 2020). While a complete survey of attention mechanisms proposed for radar datasets is beyond the scope of this chapter, we should underscore that this topic remains an active research area with highly relevant implications for cognitive radar.

7.4.2 Incorporation of Domain Knowledge into DNNs: Physics-Aware DL One of the great benefits of DL is that it allows for data-driven learning, which can capture environment-specific, sensor-specific, or even target-specific properties. However, this approach also neglects the large body of knowledge that has been gained over the years from physics, modeling, and simulation, which has formed the basis for many effective radar signal processing techniques to date. Physicsbased models represent an important source of domain knowledge, especially because they represent information that can constrain DNNs to avoid making physically impossible decisions and improve the accuracy of predictions. They can also capture phenomenological factors integral to the sensing scenario as well as sensor properties. However, physics-based models are less adept at capturing the nuances of environment-specific, sensor-specific, or subject-specific properties, which lie at the heart of data-driven dynamic adaptive systems (DDDASs) such as

7 Integrating AI into Radar System Design: Next-Generation Cognitive Radars

205

cognitive radar. Here, deep learning can provide tremendous insight through datadriven learning. Unfortunately, in sensing problems, it is not common to have a lot of data. The limitations in training sample support ultimately also limit the accuracy and efficacy of DL in sensing. Moreover, no model is perfect—while more complex models could surely be developed to improve accuracy, the dynamic nature of the sensing environment ensures that there will always be some part of the signal that is unknown. This is where leveraging data-driven DL can provide a powerful tool when used in tandem with physics-based models. The resulting hybrid approach, physics-aware machine learning (PhML) (Fig. 7.4), combines the strengths of DL and physics-based modeling to optimize trade-offs between prior versus new knowledge, models vs. data, uncertainty, complexity, and computation time, for greater accuracy and robustness. Much of current literature involving PhML (Willard et al. 2022) has focused on the solution of ordinary differential equations (ODEs), data-driven discovery of physical laws, uncertainty quantification, and data generation—both to synthesize data for validation on simulated data in cases where acquiring real measurements is not feasible and for physics-guided initialization to pre-train deep models. The question of whether Generative Adversarial Network (GAN)-generated samples conform to physical constraints has recently been raised in the context of turbulent flow simulation (Wang et al. 2020), where both deterministic constraints (conservation laws) and statistical constraints (energy spectrum of turbulent flows) have been proposed for incorporation into the loss function. These constraints were shown to yield improvements in performance relative to that attained by standard GANs because the synthetic samples more faithfully emulated certain physical properties of the system, while also significantly reducing (by up to 80%) the training time. Domain knowledge and physics-based models can be incorporated into any step within an ML approach, starting from the way RF data is presented to a DNN, to how the DNN is trained and structured, and the design of the cost function to be minimized. One of the most applications of PhML is the generation of synthetic radar datasets. Data synthesis is important not only for the training of deep models, but also because of its potential role in the testing and evaluation (T&E) of cognitive radar systems. It is often not feasible to test cognitive radar across all possible operational scenarios it may encounter. This is because in real-world applications,

Fig. 7.4 Physics-aware machine learning trade-off

206

S. Z. Gurbuz et al.

target behavior and environmental factors are dynamic. In contrast, simulations would provide the ability to fully validate ATR algorithms prior to deployment in operational settings for which real data is not obtainable. This requires not only accurately representing expected target signatures, but also site-specific clutter. Currently, there remains a significant gap between measured radar signatures and synthetic datasets, which precludes to the use of simulations for T&E. Advancement of physics-aware synthetic data generation techniques could close this gap and thus fill an important need critical to the advancement of cognitive radar design and development.

7.4.3 Physics-Aware Generative Adversarial Networks Model-based training data synthesis has been an effective method for simulating the expected radar returns from a variety of targets for any desired antenna-target geometry. While there are models for simulating various clutter profiles, because interference sources may be device or environment-specific, data-driven methods for data synthesis such as adversarial learning are well-suited to account for such factors. Adversarial learning can be exploited in several different ways to learn and transfer knowledge in offline model training, as illustrated in Fig. 7.5; for example: • To improve realism of synthetic data generated from physics-based models • To adapt data from a different source to resemble data from the target do-main • To directly synthesize both target and clutter components of measured RF data The main benefit of using adversarial learning to improve the realism of synthetic images generated from physics-based models is that it preserves the properties of the target signature that are bound by physical and kinematic constraints, while using the adversarial neural network to learn features in the data unrelated to the target model, e.g., sensor artifacts and clutter. The goal for improving realism is to generate training images that better capture the characteristics of each class and thus improve the resulting test accuracy. However, as the goal of the refiner is merely to improve its similarity to real data, a one-to-one correspondence is maintained

Fig. 7.5 Adversarial training data generation approaches

7 Integrating AI into Radar System Design: Next-Generation Cognitive Radars

207

between synthetic and refined samples. In other words, however, much data we have at the outset, as generated by physics-based models, is the same amount of data that we have after the refinement process—no additional data is synthesized. Alternatively, the data from a source domain may be adapted to resemble real data acquired in the target domain; then, the adapted data is used for network initialization. In this approach, the source domain can be real data acquired using a different RF sensor with different transmit parameters (frequency, bandwidth, pulse repetition interval), while the target domain is that which is to be classified. For example, consider the problem of radar micro-Doppler signature classification. Suppose we wish to classify data acquired from a 77 GHz frequency modulated continuous wave (FMCW) automotive radar, but there is insufficient data to adequately train a DNN. One approach could be to possibly utilize data from a publicly released dataset, or data from a different RF sensor. Suppose we have ample real data from two other RF sensors—a 10 GHz ultra-wideband impulse radar and a 24 GHz FMCW radar. Although the data from these three RF sensors will be similar for the same activity, there are sufficient differences in the micro-Doppler that direct transfer learning suffers from great performance degradation. While the classification accuracy of 77 GHz data with training data from the same sensor can be as high as 91%, the accuracy attained when trained on 24 and 10 GHz data is just 27% and 20% (Gurbuz et al. 2020b), respectively. This represents over 65% poorer accuracy. On the other hand, when adversarial domain adaptation is applied to first transform the 10 and 24 GHz data to resemble that of the target 77 GHz data, classification performance that surpasses that of training with just real target data can be achieved. Effective data synthesis requires being able to represent both target and clutter components. While model-based methods effectively capture target kinematics, they do not capture environmental factors. Conversely, while GANs can capture sensor artifacts and clutter, they face challenges in accurately representing target kinematics. In radar applications, the fidelity of target representations in GANsynthesized data cannot be evaluated by considering image quality alone. A critical challenge is the possibility of generating misleading synthetic signatures. For example, micro-Doppler signature characteristics are constrained not only by the physics of electromagnetic scattering, but also by human kinematics. The skeleton physically constrains the possible variations of the spectrogram corresponding to a given class. But, GANs have no knowledge of these constraints. It is thus possible for GANs to generate synthetic samples that may appear visually similar but are in fact incompatible with possible human motion. An example of erroneous data generation is given in Fig. 7.6 for an auxiliaryconditional GAN (ACGAN), which visually shows some of the physically impossible or out-of-class samples synthesized by GANs. In fact, classification accuracy greatly increases when such fallacious synthetic samples are identified and discarded from the training data. For example, when an ACGAN was used to synthesize 40,000 samples, 9000 samples were identified as kinematic outliers and discarded from the training dataset, but in doing so the classification accuracy increased by 10% (Erol et al. 2020).

208

S. Z. Gurbuz et al.

Physics-aware GAN design (Rahman et al. 2021; Gurbuz 2022) aims at preventing the generation of kinematically flawed data that is inconsistent with possible target behavior by integrating physics-based models and domain knowledge into data-driven DL. This can be done through modifying the GAN architecture, the way the GAN is trained, and by developing more meaningful loss functions. Current GAN architectures typically use distance measures rooted in statistics, such as cross-entropy or Earth mover’s distance, to derive a loss metric reflective of the degree of discrepancy between true and predicted samples. Statistical distance metrics, however, do not reflect discrepancies in the underlying kinematics of actual target motion versus predicted motion. However, target kinematics are physically constrained. One way of informing a GAN of such constraints is through the design of a physics-based loss term that is added to the standard statistical loss term: Loss = Losscritic +LossGP +Lossphysics , where the first term is the critic loss and the second term is gradient penalty (GP). In this way, a penalty will now be incurred if the resulting signature exhibits deviant physical properties. In the case of microDoppler, one property of special significance is the envelopes of the signature, as they form a physical upper bound of the radial velocity incurred during motion. Thus, physics-based loss metrics reflective of consistency in envelope between synthetic and real samples are proposed for micro-Doppler signature classification. Envelope information is provided to the GAN in two different ways: (1) through the addition of one or two auxiliary branches in the discriminator, which take as input the upper and lower envelope of the micro-Doppler signature, as shown in Fig. 7.6 of the resulting Multi-Branch GAN (MBGAN); and (2) physics-based loss metrics such as the Dynamic Time Warping (DTW) distance. In time-series analysis, DTW is an algorithm for measuring the similarity between two temporal sequences that may vary in time or speed. Utilizing a convolutional autoencoder trained on GAN-synthesized data for a 15-class activity recognition problem, we found that use of physics-aware loss in the MBGAN architecture resulting a 9% improvement in classification accuracy over the use of real data alone, and 6% improvement over that of a Wasserstein GAN with gradient penalty. For ambulatory data, the use of just a single auxiliary branch taking the upper envelope worked best. But

Fig. 7.6 (Left) Kinematic flaws in ACGAN micro-Doppler signatures (μDS) and (right) physicsaware MBGAN architecture

7 Integrating AI into Radar System Design: Next-Generation Cognitive Radars

209

in classification of 100 words of sign language (Rahman et al. 2022), the dual branch MBGAN worked best. This illustrates the degree to which understanding the kinematics of the problem can affect results: for ambulation toward the radar, the dominant micro-Doppler are positive and reflected in the upper envelope. But in signing, both positive and negative frequencies hold great significance. For both datasets (ambulatory and signing), we found that direct synthesis of micro-Doppler signature outperformed domain adaptation from other RF data sources.

7.4.4 Reinforcement Learning Reinforcement learning (RL) is a branch of ML that is of growing interest within the cognitive radar community as it can form the basis for the decision process governing the radar’s interaction with its environment. RL generally solves problems expressed as Markov Decision Processes (MDP), formally defined by a tuple (S, A, T , R, γ ), where S is a set of states, A is a set of actions, T is a set of conditional transition probabilities between states, R : S × A × S → R is the reward function, and γ ∈ [0, 1] is the discount factor. An action a ∈ A transitions the environment from state s to some follow-up state s  with probability T (S  |s, a), while the reward R is obtained. The optimal is a strategy π ∗ : S → A ∞ decision t that maximizes the expected reward E[ t=0 γ · rt ]. The discount factor γ allows adjustment of the impact of the near-term vs. long-term reward. As shown in Fig. 7.7, the optimal decision policy π ∗ contains a mapping from each state to an optimal action for the cognitive radar agent to take. If all the transition probabilities and rewards were known, then algorithms such as value or policy iteration can be used to compute π ∗ . RL allows to explore these parameters “on the fly,” e.g., without any a priori knowledge, but by executing actions in Fig. 7.7 Reinforcement learning agent for cognitive radar

210

S. Z. Gurbuz et al.

the environment and observing the follow-on state and receiving some reward (model-free, unsupervised learning approach). The decision policy naturally reflects the structure of the perception–action cycle. MDPs or partially observable MDPs (POMDPs) have also been used for radar tracking (Charlish and Hoffmann 2015) or classification problems (Brüggenwirth et al. 2019). Q-Learning, SARSA, or Double-Q-Learning are some well-known algorithms that solve the RL problem. In these approaches, the Q-Value Q : S × A → R is computed as an estimate for the expect reward in the current state S taking action A. The -value defines the amount of exploration (e.g., trying out new actions) vs. exploitation (e.g., using the current optimal policy). Unfortunately, the curse of dimensionality (Goodfellow et al. 2016) often prevents these approaches to scale to larger problem sizes. The concept of Deep Q-Learning (DQN) (Mnih et al. 2015) revolutionized the field in 2015, by learning human-level control policies on a variety of different Atari 2600 games simply by watching the screen and executing different control actions. In this approach, a deep neural net (e.g., CNN) is used to approximate the Q-Value for the state and action spaces in a neural net. In Double-Q-Learning (DDQN), a different network is used for value evaluation than for the selection of the next action. This improves stability and the learning rate in noisy environments. Policy gradients algorithms such as DDPG directly optimize in the policy space. They are generally believed to be applicable to a wider range of problems but can suffer from instability during training. In Ak and Brüggenwirth (2020), DQN and DDQN have been used for a cognitive radar problem for jammer or interference mitigation. The objective for a steppedfrequency phased-array radar was to avoid co-channel interference in the frequency and spatial domains, e.g., the radar had to choose the proper waveform that omits the frequency bands used by the interferer or steer its electronic beam accordingly for surveillance tasks. For each successful transmission, the cognitive radar agent received a reward. The simulation results show that the cognitive radar approach using RL outperforms the use of random frequency or beam position selection. Also, it was shown that a configuration using DQN and LSTM networks performed best. For cognitive radar applications, it remains to be determined how much training is done beforehand, e.g., using simulations to find π ∗ , and how much exploration (i.e., improvement to the policy) is done in later in the real radar environment. As such, the design of RL techniques to implement cognitive decision processes remains an active research area.

7.5 Case Studies 7.5.1 Multitarget Tracking and Classification Using the Bayesian Approach This example demonstrates how the Bayesian approach to cognitive radar is used to select radar waveforms to perform simultaneous tracking and classification for

7 Integrating AI into Radar System Design: Next-Generation Cognitive Radars

211

Fig. 7.8 Tracking and classification simulation scenario (Bell et al. 2021)

three targets (Bell et al. 2021). Our simulation includes three moving targets and a moving sensor as illustrated in Fig. 7.8. The targets are assumed to be from one of five classes, which remain fixed throughout the simulation. The measuring platform can adaptively select between a “tracking” mode and a “classification” mode. When in tracking mode, it can choose a pulse repetition frequency (PRF), bandwidth (BW), and pulse count (Np ) from a list of possibilities. The measurement process results in detection-level data of target range, range rate, azimuth, and elevation. Tracking measurements are made with a probability of detection PD , which is a function of signal-to-noise ratio (SNR), which is in turn a function of the waveform parameters. When in classification mode, the radar can select among modes which trade probability of correct classification (pcc ) with dwell time (T ). When this measurement modality is selected, the observation is a classification call. The available waveforms, their parameters, and dwell times are listed in Table 7.1. Also included is the “do nothing” waveform #0. Targets are tracked using detection data coupled with an Extended Kalman Filter and classified using classification calls and an exact Bayesian recursion on the target class distribution. The tracking and classification update cycle is every 100 ms. We assume 10 ms is allocated for the purposes of tracking and classification of the three known targets, and the remaining 90 ms is used for other radar functions. The resource allocation algorithm may elect to measure each target during the 10 ms dwell or any subset of the targets as long as the total measurement time fits into the allocated time budget. For each target, the sensor may select from the following options:

212 Table 7.1 Waveform parameters and dwell times

S. Z. Gurbuz et al. Waveform

BW (MHz)

PRF (kHz)

Np

T (ms)

0 1,2,3 4,5,6 7,8,9 10,11,12 13,14,15 16,17,18 19,20,21 22,23,24

1,5,10 1,5,10 1,5,10 1,5,10 1,5,10 1,5,10 1,5,10 1,5,10

N/A 20 10 20 10 20 10 20 10

1 1 10 10 20 20 50 50

0.0 0.05 0.1 0.5 1.0 1.0 2.0 2.5 5.0

Waveform 25 26 27

pcc 0.3 0.6 0.75

T (ms) 1.0 2.5 5.0

• Do nothing Choose waveform #0. This takes zero time and generates zero utility. It frees up the timeline to dedicate extra dwell time to other targets. • Perform a track dwell. Choose from waveforms #1–24. This takes variable time given by Np /PRF and provides variable utility depending on the waveform parameters. • Perform a classification dwell. Choose from waveforms #25–27. This takes variable time and provides variable utility. The predicted utility of each sensing action is scored using an informationtheoretic metric, the mutual information. This is a convenient metric that incorporates information about the target kinematic state and class and is straightforward to compute. The objective function is then maximized subject to the timeline. Figure 7.9 shows the time sequence of measurement modalities (no measurement, track measurement, or classification measurement) selected for each of the three targets over 100 Monte Carlo trials. It also shows the probability of choosing the correct class and the track root mean square error (RMSE). The trials have the sensor and target trajectories and classes fixed, but a random realization of the measurements is drawn each time. This, in turn, affects the cognitive radar waveform selection calculations leading to different waveform selections, and therefore a different classification probability and track RMSE each time. Note that the target class is static so once the class is identified with high probability, no further classification measurements are needed. Broadly speaking, we find that algorithm interleaves track and classification dwells during the first portion of the simulation to learn about the target class while maintaining track accuracy. This example has demonstrated that the Bayesian approach to implementing the cognitive radar perception–action cycle can effectively allocate radar system resources among tracking and classification tasks for multiple targets, thus making efficient use of the radar timeline to optimize system performance.

7 Integrating AI into Radar System Design: Next-Generation Cognitive Radars

213

Fig. 7.9 Tracking and classification results. Top: Mode selection for the three targets; Bottom: classification probability, tracking position, and velocity RMSE (Bell et al. 2021)

7.5.2 Multitarget Detection in Massive MIMO Radar Using Reinforcement Learning This example demonstrates the use of an RL-based algorithm for cognitive multitarget detection in the presence of unknown disturbance characteristics (Ahmed et al. 2021). RL is a learning approach addressing model-free problems by using software-defined agents, which learn from the observations collected from the environment, and take the best possible actions according to a reward function. The radar acts as an agent that continuously senses the unknown environment (i.e., targets and disturbance) and consequently optimizes transmitted waveforms in order to maximize the probability of detection (PD ) by focusing the energy in specific range-angle cells (i.e., beamforming). We assume no prior information about the environment; in particular, no assumptions are made about the number of targets or about the statistical model of the disturbance. In these simulations, the environment changes, and the performance of our algorithm is analyzed such that the radar agent capability to adapt to those changes is tested. The number of total time steps is 100, and the results are averaged over 1000 Monte Carlo runs. The performance of the algorithm is analyzed for a massive multiple-input multiple-output (MMIMO) regime where the number of transmit and receive elements is N = 104 and the detection threshold is set so that the probability of false alarm is PF A = 10−4 . We consider a uniform linear array (ULA) at the transmitter and receiver each with half-wavelength interelement spacing. The angular locations are represented in terms of the spatial frequency ν ∈ [−0.5, 0.5). The angle grid is divided into L = 20 angle bins with spacing ν = 0.05.

214

S. Z. Gurbuz et al.

Fig. 7.10 Disturbance PSD (blue) with target angle locations (red dashed lines for initial values and black dashed lines for final values) (Ahmed et al. 2021)

We assume the existence of four targets at various spatial frequency locations with SNR = −5 dB, −8 dB, −10 dB, −9 dB, respectively. In this scenario, the targets’ spatial frequencies are changed after 50 time steps. Initially, ν = [−0.2, 0, 0.2, 0.3], and finally, ν = [−0.05, 0.05, 0.25, 0.35], while their respective SNR remains the same. The initial spatial frequencies are depicted in dashed red in Fig. 7.10, and the final spatial frequencies are depicted in dashed black. The disturbance model is chosen to mask the target angles, where the disturbance power is spread all over the spatial frequency range. The disturbance power spectral density (PSD) is shown in Fig. 7.10 in blue. Note that the disturbance PSD has multiple peaks. We compare the performance of our RL-based beamformer to a standard omnidirectional equal power allocation with no RL. Here, the antennas emit orthogonal waveforms, and the power is divided equally across all antennas. Figure 7.11 depicts the difference between our proposed algorithm and omnidirectional MIMO. In order to obtain those figures, we counted the number of detections in each angle bin, averaged across all Monte Carlo runs. Figure 7.11a demonstrates better detection performance for all targets, even the ones with low SNR. Conversely, in the omnidirectional approach in Fig. 7.11b, targets 2, 3, and 4 with lower SNR are mostly masked under the disturbance peaks. Figure 7.12 shows the reward behavior across time and averaged over the Monte Carlo runs. The reward in Fig. 7.12 and the PD in Fig. 7.11 show convergence after the first T = 20 time steps. Then, when the environment is changed by changing the angles, the agent sensed that change through exploration. Hence, the drop in the reward and PD is seen after T = 50 time steps, where the agent starts relearning the changes, and then after 10 time steps, the reward and PD converge again. As highlighted by the simulation results, our RL-based beamformer outperforms the omnidirectional beamformer in terms of target detection performance under environmentally harsh conditions such as low SNR, heavy-tailed disturbance, and rapidly changing scenarios.

7 Integrating AI into Radar System Design: Next-Generation Cognitive Radars

215

Fig. 7.11 Detection performance of RL-based beamforming versus omnidirectional with equal power allocation for dynamic environment: changing angles at T = 50. (a) RL-based beamforming. (b) Omnidirectional (Ahmed et al. 2021) Fig. 7.12 RL reward in dynamic environments: changing angles at T = 50 (Ahmed et al. 2021)

7.6 Challenges The application of deep learning in cognitive radar faces several challenges, including data representation, suitable network architectures, bias in acquired data, lack of sufficient training data, and real-time computation on embedded processing platforms. In this section, we will discuss each of these challenges in turn.

7.6.1 Data Representation and Network Architectures The data acquired by radar systems is typically acquired through quadrature receivers that supply complex sampled data, zn , so that the phase can be uniquely ascertained as zn . The received signal supplied to the radar processor is inherently a complex time series. Unfortunately, the predominance of neural network architectures presumes inputs of 1D or 2D real data, and current DNN architectures cannot process data in a fashion compatible with the original form of the radar measurements. Pre-processing is therefore needed to convert data. Examples include synthetic aperture radar imaging, high-range resolution profiling, or time–frequency analysis. While these outputs are complex, due to the reliance on real inputs, only the magnitude is provided to the DNN. While results can be obtained in this fashion, they are sub-optimal.

216

S. Z. Gurbuz et al.

7.6.2 Data Bias Bias in a neural network or machine learning algorithm’s output can be present when the correct pre-processing of the data has not occurred. Pre-processing and standardization are common for machine learning approaches in order to remove bias and process data.

7.6.3 Low Training Sample Support Radar data is typically time-consuming and costly to acquire, resulting in limited availability. Techniques such as unsupervised pre-training and transfer learning (TL) from other domains have been proposed to overcome the problem of low sample support in RF sensing. While TL has been helpful, it does suffer from losses due to the differences in phenomenology between source and target domains. To minimize these differences, synthetic data generation techniques using modeling of target motion and generative adversarial learning have been proposed. However, these methods are applicable to the synthesis of images, which, in the case of radar, often represent a snapshot of the data in time. In contrast, cognitive radars represent dynamic systems, which can adapt on both transmit and receive as time progresses. Thus, a critical need is to develop DNNs that operate effectively on time series. While RNNs provide one such mechanism, RL has become a focus of greater interest as a means to dynamically train networks. RL algorithms mitigate the challenge of limited data/training data. Training data is replaced with an environment model and a custom reward function. Randomized training examples are created using the environment model. The reward function is used to evaluate the neural network’s performance against the example and to provide an input to the network weight adjustment algorithm. RL includes the idea of “trial and error”-based learning. Each training run is a trial. The error, quantified by the reward function, is used to learn. Not only can RL mitigate the problem of low training sample support, but it can also provide a mechanism for addressing the additional challenges presented by open-set problems and can exploit contextual awareness.

7.6.4 Real-Time Embedded Implementations Lack of computational resources can also be a bottleneck to machine learning algorithms and approaches due to higher-order processing needed to train machine learning models. The use of advanced computing systems, graphical processing units (GPUs), parallel processing tools, and cloud infrastructures enables machine learning algorithms such as neural networks to be computationally efficient and feasible. However, enabling real-time implementations of DNNs is still an active

7 Integrating AI into Radar System Design: Next-Generation Cognitive Radars

217

research area and is of critical importance to cognitive radars. For efficient perception–action cycle feedback, the lag time in analysis and decision-making must be as short as possible.

7.7 Conclusion This chapter has provided an overview of the fundamental motivations and principles guiding the design of next-generation cognitive radar systems. The essential elements of a cognitive radar, as represented through the modeling of cognitive processes, especially the perception–action cycle, have been presented. Cognitive radar architectures and enabling RF hardware and antenna design, as well as Bayesian and deep-learning-based approaches to developing decision processes for fully adaptive transmissions, have been discussed. The chapter concludes with examples of two case studies: (1) multi-target tracking and classification using the Bayesian approach and (2) multi-target detection in massive MIMO radar using reinforcement learning. The challenges in the application of deep learning to cognitive radar are discussed to provide insights into directions for future research needed to continue to develop more advanced cognitive radar systems.

References Adeoluwa O, Kearney SJ, Kurtoglu E, Connors C, Gurbuz SZ (2021) Near real-time ASL recognition using a millimeter wave radar. In: Defense + commercial sensing Ahmed A, Ahmad AA, Fortunati S, Sezgin A, Greco MS, Gini F (2021) A reinforcement learning based approach for multitarget detection in massive mimo radar. IEEE Trans Aerosp Electron Syst 57(5):2622–2636 Ak S, Brüggenwirth S (2020) Avoiding jammers: a reinforcement learning approach. In: 2020 IEEE international radar conference (RADAR), pp 321–326 Aubry A, De Maio A, Piezzo M, Farina A (2014) Radar waveform design in a spectrally crowded environment via nonconvex quadratic optimization. IEEE Trans Aerosp Electron Syst 50(2):1138–1152 Bell K, Baker C, Smith G, Johnson J, Rangaswamy M (2015) Cognitive radar framework for target detection and tracking. IEEE J Sel Topics Signal Process 9(8):1427–1439 Bell K, Smith GE, Mitchell AE, Rangaswamy M (2018) Multiple task fully adaptive radar. In: Proceedings of the 52nd Asilomar conference on signals, systems, and computerse Bell K, Smith G, Mitchell A, Rangaswamy M (2019) Fully adaptive radar for target classification. In: Proceedings of the 2019 IEEE radar conference Bell K, Kreucher C, Rangaswamy M (2021) An evaluation of task and information driven approaches for radar resource allocation. In: Proceedings of the 2021 IEEE radar conference Bloom B, Krathwohl D, Masia B (1956) Taxonomy of educational objectives: the classification of educational goals. David McKay Company, New York Brüggenwirth S, Warnke M, Wagner S, Barth K (2019) Cognitive radar for classification. IEEE Aerosp Electron Syst Mag 34(12), 30–38 Campbell C, Ahmad F (2020) Attention-augmented convolutional autoencoder for radar-based human activity recognition. In: 2020 IEEE international radar conference (RADAR), pp 990– 995

218

S. Z. Gurbuz et al.

Charlish A, Hoffmann F (2015) Anticipation in cognitive radar using stochastic control. In: 2015 IEEE radar conference (RadarCon), pp 1692–1697 Charlish A, Hoffmann F (2017) Cognitive radar management. In: Novel radar techniques and applications volume 2: waveform diversity and cognitive radar, and target tracking and data fusion, institution of engineering and technology, pp 157–193 Charlish A, Katsilieris F (2017) Array radar resource management. In: Novel radar techniques and applications: real aperture array radar, imaging radar, and passive and multistatic radar, institution of engineering and technology, pp 135–171 Charlish A, Bell K, Kreucher K (2020) Implementing perception-action cycles using stochastic optimization. In: Proceedings of the 2020 IEEE radar conference Chong E, Kreucher C, Hero AO (2008) Monto-carlo-based partially observable markov decision process approximations for adaptive sensing. In: Proceedings of the 9th international workshop on discrete event systems, pp 173–180 De Lange M, Aljundi R, Masana M, Parisot S, Jia X, Leonardis A, Slabaugh G, Tuytelaars T (2022) A continual learning survey: Defying forgetting in classification tasks. IEEE Trans Pattern Analy Mach Intell 44(7):3366–3385 de Santana Correia A, Colombini EL (2021) Attention, please! A survey of neural attention models in deep learning. CoRR abs/2103.167751. [Online]. Available: https://arxiv.org/abs/ 2103.16775 Erol B, Amin MG (2019) Radar data cube processing for human activity recognition using multisubspace learning. IEEE Trans Aerosp Electron Syst 55(6):3617–3628 Erol B, Gurbuz SZ, Amin MG (2018) Frequency-warped cepstral heatmaps for deep learning of human motion signatures. In: 2018 52nd Asilomar conference on signals, systems, and computers, pp 1234–1238 Erol B, Gurbuz SZ, Amin MG (2020) Motion classification using kinematically sifted acgansynthesized radar micro-doppler signatures. IEEE Trans Aerosp Electron Syst 56(4):3197– 3213 Fuster J (2003) Cortex and mind: unifying cognition. Oxford University Press, Oxford Ghaffarian S, Valente J, van der Voort M, Tekinerdogan B (2021) Effect of attention mechanism in deep learning-based remote sensing image processing: a systematic literature review. Remote Sensing 13(15). [Online]. Available: https://www.mdpi.com/2072-4292/13/15/2965 Goodfellow I, Bengio Y, Courville A (eds) (2016) Deep learning. MIT Press, Cambridge Goodman NA, Venkata PR, Neifeld MA (2007) Adaptive waveform design and sequential hypothesis testing for target recognition with active sensors. IEEE J Sel Topics Signal Process 1(1):105–113 Greco MS, Gini F, Stinco P, Bell K (2018) Cognitive radars: on the road to reality: progress thus far and possibilities for the future. IEEE Signal Process Mag 35(4):112–125 Guerci J (2010a) Cognitive radar: the knowledge-aided fully adaptive approach. Artech House, Norwood Guerci JR (2010b) Cognitive radar: the knowledge-aided fully adaptive approach. Artech House, Norwood Gurbuz S (ed) (2020) Deep neural network design for radar applications. IET/SciTech Publishers, London Gurbuz S (2022) Physics-aware machine learning for dynamic data-driven radar target recognition. In: Proceedings of the dynamic data driven application systems (DDDAS) conference Gurbuz SZ, Griffiths HD, Charlish A, Rangaswamy M, Greco MS, Bell K (2019) An overview of cognitive radar: Past, present, and future. IEEE Aerospace Electron Syst Mag 34(12):6–18 Gurbuz AC, Mdrafi R, Cetiner BA (2020a) Cognitive radar target detection and tracking with multifunctional reconfigurable antennas. IEEE Aerosp Electron Syst Mag 35(6):64–76 Gurbuz S, Rahman M, Kurtoglu E, Macks T, Fioranelli F (2020b) Cross-frequency training with adversarial learning for radar micro-doppler signature classification. Proce SPIE 11408:1–11 Haykin S (2006) Cognitive radar: a way of the future. IEEE Signal Process Mag 23(1):30–40 Haykin S (2012) Cognitive dynamic systems: Perception-action cycle, radar and radio. Cambridge University Press, Cambridge

7 Integrating AI into Radar System Design: Next-Generation Cognitive Radars

219

Hero AO, Castanon DA, Cochran D, Kastella K (2007) Foundations and applications of sensor management. Springer, Berlin Horne C, Ritchie M, Griffiths H (2018) Proposed ontology for cognitive radar systems. IET Radar, Sonar Navig 12(12), 1363–1370 John-Baptiste P, Johnson JT, Smith GE (2022) Neural network-based control of an adaptive radar. IEEE Trans Aerosp Electron Syst 58(1):168–179 Kershaw D, Evans R (1997) Waveform selective probabilistic data association. IEEE Trans Aerosp Electron Syst 33(4):1180–1188 Kreucher C, Hero A, Kastella K (2005) A comparison of task driven and information driven sensor management for target tracking. In: Proceedings of the 44th IEEE conference on decision and control, pp 4004–4009 Kurto˘glu E, Gurbuz AC, Malaia EA, Griffin D, Crawford C, Gurbuz SZ (2022) Asl trigger recognition in mixed activity/signing sequences for rf sensor-based user interfaces. IEEE Trans Human-Mach Syst 52(4):699–712 Li H, Mehul A, Le Kernec J, Gurbuz SZ, Fioranelli F (2021) Sequential human gait classification with distributed radar sensor fusion. IEEE Sensors J 21(6):7590–7603 Martone AF, Ranney KI, Sherbondy K, Gallagher KA, Blunt SD (2018) Spectrum allocation for noncooperative radar coexistence. IEEE Trans Aerosp Electron Syst 54(1):90–105 Mitchell AE, Smith GE, Bell KL, Duly AJ, Rangaswamy M (2018) Cost function design for the fully adaptive radar framework. IET Radar Sonar Navig 12(12):1380–1389 Mnih V, Kavukcuoglu K, Silver D, Rusu AA, Veness J, Bellemare MG, Graves A, Riedmiller MA, Fidjeland A, Ostrovski G, Petersen S, Beattie C, Sadik A, Antonoglou I, King H, Kumaran D, Wierstra D, Legg S, Hassabis D (2015) Human-level control through deep reinforcement learning. Nature 518:529–533 Powell W (2019) A unified framework for stochastic optimization. Euro J Oper Res 275(3):795– 821 Rahman MM, Gurbuz SZ, Amin MG (2021) Physics-aware design of multi-branch gan for human rf micro-doppler signature synthesis. In: 2021 IEEE radar conference (RadarConf21), pp 1–6 Rahman MM, Malaia EA, Gurbuz AC, Griffin DJ, Crawford C, Gurbuz SZ (2022) Effect of kinematics and fluency in adversarial synthetic data generation for asl recognition with rf sensors. IEEE Trans Aerosp Electron Syst 58(4):2732–2745 Rasmussen J (1983) Skills, rules, and knowledge; signals, signs, and symbols, and other distinctions in human performance models. IEEE Trans Syst Man Cyber 13(3):257–266 Seyfio˘glu MS, Gürbüz SZ (2017) Deep neural network initialization methods for microdoppler classification with low training sample support. IEEE Geosci Remote Sensing Lett 14(12):2462–2466 Sira SP, Li Y, Papandreou-Suppappola A, Morrell D, Cochran D, Rangaswamy M (2009) Waveform-agile sensing for tracking. IEEE Signal Process Mag 26(1):53–64 Smith GE, Cammenga Z, Mitchell A, Bell KL, Rangaswamy M, Johnson JT, Baker CJ (2015) Experiments with cognitive radar. In: 2015 IEEE 6th international workshop on computational advances in multi-sensor adaptive processing (CAMSAP), pp 293–296 Smith GE, Gurbuz SZ, Brüggenwirth, S, John-Baptiste P (2020) Neural networks & machine learning in cognitive radar. In: 2020 IEEE radar conference (RadarConf20), pp 1–6 Stinco P, Greco M, Gini F, Himed B (2016) Cognitive radars in spectrally dense environments. IEEE Aerosp Electron Syst Mag 31(10):20–27 Towfiq MA, Bahceci I, Blanch S, Romeu J, Jofre L, Cetiner BA (2018) A reconfigurable antenna with beam steering and beamwidth variability for wireless communications. IEEE Trans Antenn Propag 66(10):5052–5063 van Keuk G, Blackman S (1993) On phased-array radar tracking and parameter control. IEEE Trans Aerosp Electron Syst 29(1):186–194 Van Trees HL, Bell KL (2007) Bayesian bounds for parameter estimation and nonlinear filtering/tracking. Wiley, Hoboken Wang S, Song J, Lien J, Poupyrev I, Hilliges O (2016) Interacting with soli: Exploring finegrained dynamic gesture recognition in the radio-frequency spectrum. In: Proceedings of the 29th annual symposium on user interface software and technology. ACM, New York, pp 851– 860

220

S. Z. Gurbuz et al.

Wang R, Kashinath K, Mustafa M, Albert A, Yu R (2020) Towards physics-informed deep learning for turbulent flow prediction. In: Proceedings of the 26th ACM SIGKDD international conference on knowledge discovery & data mining, pp 1457–1466 Wicks MC (2006) Sensors as robots. In: 2006 international waveform diversity & design conference, pp 1–7 Willard JD, Jia X, Xu S, Steinbach MS, Kumar V (2022) Integrating scientific knowledge with machine learning for engineering and environmental systems. ACM computing surveys 55:1– 37 Zhang Z, Tian Z, Zhou M (2018) Latern: dynamic continuous hand gesture recognition using fmcw radar sensor. IEEE Sensors J 18(8):3278–3289

Sevgi Z. Gurbuz received the B.S. degree in electrical engineering with a minor in mechanical engineering and the M.E. degree in electrical engineering and computer science from the Massachusetts Institute of Technology, Cambridge, MA, USA, in 1998 and 2000, respectively, and the Ph.D. degree in electrical and computer engineering from Georgia Institute of Technology, Atlanta, GA, USA, in 2009. From February 2000 to January 2004, she worked as a Radar Signal Processing Research Engineer with the U.S. Air Force Research Laboratory, Sensors Directorate, Rome, NY, USA. Formerly as an Assistant Professor in the Department of Electrical and Electronics Engineering at TOBB University of Economics and Technology, Ankara, Turkey, and Senior Research Scientist with the TUBITAK Space Technologies Research Institute, Ankara, Turkey, she is currently an Assistant Professor of Electrical and Computer Engineering at the University of Alabama, Tuscaloosa, AL, where she directs the Laboratory for Computational Intelligence in Radar (CI4R). Her current research interests include radar signal processing, statistical signal processing, AI/ML for radar perception, and radar-enabled cyber-physical systems, especially for biomedical, human-computer interaction (HCI), and automotive autonomy applications. She was granted a US Patent related to radar-based sign language recognition in 2022 and is the author of over 90 conference, 25 journal publications, and eight book chapters. She is the editor of a book entitled “Deep Neural Network Design for Radar Applications” that was published by IET in 2020. Currently, Dr. Gurbuz serves as a member of the IEEE Radar Systems Panel and its Education, Publications and Civilian Radar committees, and as an Associate Editor of the IEEE Transactions of Aerospace and Electronic Systems, IEEE Transactions on Radar Systems, and IET Radar, Sonar, and Navigation. Dr. Gurbuz is a recipient of a 2023 NSF Career award, the 2022 American Association of University Women Research Publication Grant in Medicine and Biology, IEEE Harry Rowe Mimno Award for 2019, 2020 SPIE Rising Researcher Award, EU Marie Curie Research Fellowship, and the 2010 IEEE Radar Conference Best Student Paper Award. She is a Senior Member of the IEEE, and a member of the SPIE and ACM.

7 Integrating AI into Radar System Design: Next-Generation Cognitive Radars

221

Kristine L. Bell (M’88-S’91-M’96-SM’01-F’15) received the B.S. in Electrical Engineering from Rice University in 1985, and the M.S. in Electrical Engineering and the Ph.D. in Information Technology from George Mason University (GMU) in 1990 and 1995, respectively. She was an Associate/Assistant Professor with the Statistics Department and C4I Center, GMU, from 1996 to 2009. She is currently a Senior Fellow with Metron, Inc. and also holds an Affiliate Faculty position in the Statistics Department, GMU. Her technical expertise is in the area of statistical signal processing and multi-target tracking with applications in radar and sonar. Her research interests include cognitive sensing and sensor management. Dr. Bell has served on the IEEE Dennis J. Picard Radar Technologies Medal Selection Committee, the IEEE Jack S. Kilby Signal Processing Medal Selection Committee, the IEEE Aerospace and Electronic Systems Society (AESS) Fellow Evaluation Committee, and the AESS Radar Systems Panel, where she was the chair of the Student Paper Competition Committee. She was an Associate Editor of the IEEE Transactions on Signal Processing and the Chair of the IEEE Signal Processing Society’s Sensor Array and Multichannel (SAM) Technical Committee. Dr. Bell was a recipient of the GMU Volgenau School of Engineering Outstanding Alumnus Award in 2009 and the IEEE AESS Harry Rowe Mimno Best Magazine Paper Award in 2021.

Maria Sabrina Greco graduated in Electronic Engineering in 1993 and received the Ph.D. degree in Telecommunication Engineering in 1998, from University of Pisa, Italy. From December 1997 to May 1998 she joined the Georgia Tech Research Institute, Atlanta, USA as a visiting research scholar where she carried on research activity in the field of radar detection in non-Gaussian background. In 1993 she joined the Dept. of Information Engineering of the University of Pisa, where she is Full Professor since Dec. 2016. She is an IEEE fellow since Jan. 2011 and she was co-recipient of the 2001 and 2012 IEEE Aerospace and Electronic Systems Society’s Barry Carlton Awards for Best Paper and recipient of the 2008 Fred Nathanson Young Engineer of the Year award for contributions to signal processing, estimation, and detection theory. In May and June 2015 she visited as invited Professor the Université Paris-Sud, CentraleSupélec, Paris, France. She has been general-chair, technical program chair and organizing committee member of many international conferences over the last 10 years. She has been lead guest editor of the special issue on "Advanced Signal Processing for Radar Applications" of the IEEE Journal on Special Topics of Signal Processing, December 2015, guest co-editor of the special issue of the Journal of the IEEE Signal Processing Society on Special Topics in Signal Processing on "Adaptive Waveform Design for Agile Sensing and Communication," published in June 2007 and lead guest editor of the special issue of International Journal of Navigation and Observation on" Modelling and Processing of Radar Signals for Earth Observation published in August 2008. She is an Associate Editor of IET Proceedings - Sonar,

222

S. Z. Gurbuz et al. Radar and Navigation, Editor-in-Chief of the IEEE Aerospace and Electronic Systems Magazine, member of the Editorial Board of the Springer Journal of Advances in Signal Processing (JASP), and Senior Editorial board member of IEEE Journal on Selected Topics of Signal Processing (J-STSP). She is also member of the IEEE AES and IEEE SP Board of Governors and Past Chair of the IEEE AESS Radar Panel. She has been as well SP Distinguished Lecturer for the years 2014–2015, and now she is AESS Distinguished Lecturer for the years 2015–2017 and member of the IEEE Fellow Committee. Elected President of the IEEE Aerospace Electronic Systems Society (AESS), an association of the IEEE (Institute of Electrical and Electronics Engineers) which has about 5000 members among Electronic Engineers, Telecommunications, Systems, Aerospace, and whose area of interest concerns complex space, maritime and land-based systems, including radar, and its various applications. Professor Greco will hold the position of President-Elect for the next two years and will be the President of the Society for the other following two years. Her general interests are in the areas of statistical signal processing, estimation and detection theory. In particular, her research interests include clutter models, coherent and incoherent detection in non-Gaussian clutter, CFAR techniques, radar waveform diversity and bistatic/multistatic active and passive radars, cognitive radars. She co-authored many book chapters and more than 190 journal and conference papers.

Chapter 8

Passive Radar: A Challenge Where Resourcefulness Is the Key to Success Francesca Filippini and Fabiola Colone

8.1 Introduction Nowadays, radar systems are employed in a wide variety of applications that range from military and defense purposes to civil sectors such as automotive advanced driver assistance systems and smart-home features. However, the history of these sensors has roots that go back more than a century, to the first experimental demonstrations made by Christian Hülsmeyer, in 1904, and Guglielmo Marconi, in 1920 (Gordon 1985; Griffiths et al. 2019; Griffiths 2013; Marconi 1922). However, it was not until World War II that radar technology underwent significant development, primarily driven by military and security applications. Later, radar has become a key sensor also for civil applications such as urban and indoor monitoring as well as air, maritime, and ground traffic control. To this day, the radar technology can detect the presence of a target, accurately locate it in a two- or three-dimensional space, estimate its motion parameters, create images, and extract features that enable its classification. The rapid development of this technology and its now pervasive use in different areas of our daily lives have been motivated by several advantages that are offered by these sensors. These include, among others, its ability to work both day and night; the capability to penetrate clouds, fogs, mist, and snow; and therefore to work properly under several weather conditions. However, it is also possible to identify some of the major limitations of these systems. Among them, it is worth mentioning that conventional active radars are typically characterized by high cost of realization and maintenance and have a non-negligible impact on the environment where they are installed. By emitting electromagnetic energy, they might cause interference with preexisting

F. Filippini () · F. Colone Department of Information Engineering, Electronics and Telecommunications (DIET), SAPIENZA University of Rome, Rome, Italy e-mail: [email protected]; [email protected] © The Author(s), under exclusive license to Springer Nature Switzerland AG 2023 M. S. Greco et al. (eds.), Women in Telecommunications, Women in Engineering and Science, https://doi.org/10.1007/978-3-031-21975-7_8

223

224

F. Filippini and F. Colone

systems, and their installation is typically not welcome in populated areas. Finally, the increasing demand of spectral resources by the continuously proliferating radio frequency (RF) technologies is causing a crowded spectrum (Griffiths et al. 2015), where sensing applications must compete with communications ones. In this chapter, we take this perspective, and we introduce an alternative radar sensor to the conventional ones that can mitigate the identified limitations and can be more easily employed in a variety of applications that will be described and shown throughout this chapter. The purpose of the chapter is not to provide a detailed description of each considered application and required techniques but rather to provide an overview of the latest and most fascinating challenges in this field. For each application scenario, we refer the interested reader to the referenced literature for a deeper understanding. The reminder of the chapter is organized as follows. In Sect. 8.2, the passive radar concept will be introduced along with the main advantages and limitations of this technology. Section 8.3 will focus on two of the conventional applications of this technology, namely, the aerial and maritime surveillance. In Sect. 8.4, instead, we will describe some more recent applications of these sensors and those we think may be the most interesting development frontiers in the upcoming years. Throughout the chapter, the description of the fields of application of this technology will be supported by several experimental results that demonstrate its effectiveness in practical operations.

8.2 Passive Radar Radar systems perform target detection and localization by effectively processing the reflections of a transmitted signal on an object of interest. Passive radar (PR), also referred to as passive coherent location (PCL) system or passive covert radar (PCR), is a receive-only bistatic radar system that does not transmit electromagnetic energy on its own, instead it exploits signals emitted by preexisting noncooperative sources of opportunity, usually referred to as illuminators of opportunity (IOs), to perform target detection and localization (Baker et al. 2005; Griffiths and Baker 2005, 2014, 2017). By relying on a third-party transmitter, the exploited waveform is not known to the receiver. Therefore, passive radars are generally equipped with at least two receiving antennas, one pointed toward the area to be monitored, referred to as surveillance antenna, and the other, referred to as reference antenna, steered toward the transmitter of opportunity to collect a clean copy of the transmitted signal; see the geometry sketched in Fig. 8.1. Figure 8.1 also shows a block diagram of the basic signal processing steps foreseen in a PCL system, briefly detailed in the following. We refer the interested reader to the extensive technical literature referenced throughout this chapter for detailed insights and explanations. • Disturbance cancellation: From the geometry in Fig. 8.1, it is evident that each surveillance channel collects the target echoes but also receives a fraction of the

8 Passive Radar: A Challenge Where Resourcefulness Is the Key to Success

225

Fig. 8.1 Illustrative scenario of a passive radar system for aerial surveillance

direct signal emitted by the exploited IO (see the dashed red lines) as well as clutter and multipath contributions coming from the stationary scene. Therefore, the purpose of this processing stage is to avoid the mentioned undesired contributions mask low power target echoes. Usually, this stage is performed by subtracting the undesired interference signal, estimated by summing up delayed and weighted replicas of the reference signal, from the signal collected by each surveillance antenna (Colone et al. 2009, 2016; Lombardo and Colone 2012). • Range-Doppler map evaluation: Once the interfering contributions have been removed, the key step in the PR processing chain is based on the evaluation of the two-dimensional cross-correlation function between the surveillance signal and Doppler shifted replicas of the reference signal. Provided that the latter is a good copy of the transmitted signal, this stage approximates a bank of matched filters (MF) tuned to different Doppler of interests (Griffiths and Baker 2014; Howland et al. 2005; Lombardo and Colone 2012). By collecting the outputs of the cross-correlations at different Doppler filter, a bistatic range-Doppler (or bistatic range-velocity) map is built, where a potential target would appear as a peak at a given bin. The number of employed Doppler filters is proportional to the selected coherent processing interval (CPI) which is usually quite large in passive radars, both to extract the low power target echoes from the competing disturbance and to guarantee a high Doppler frequency resolution. • Automatic target detection: The target detection stage is a decision process and based on a comparison between a given cell under test (CUT) of the bistatic range-Doppler map and a threshold, properly selected to guarantee a desired

226

F. Filippini and F. Colone

false alarm probability (Pfa). A common solution is to use a varying threshold that adapts itself to the local disturbance characteristics to guarantee a constant false alarm rate (CFAR) (Lombardo and Colone 2012). Once a target echo peak has exceeded the selected threshold, an estimate of its bistatic range and bistatic Doppler shift will be available. • Target localization and tracking: Based on the availability of multiple surveillance channels that separately underwent the previous signal processing stage, also a direction of arrival (DoA) estimate can be extracted (Colone et al. 2013b; Filippini et al. 2018). This would allow the localization of the target peak in the Cartesian plane. Finally, target detections collected at multiple consecutive CPIs can undergo a tracking stage aimed at identified potential tracks and further reduce the resulting number of false alarms. Target tracking can be performed either in the bistatic range-Doppler domain or in the Cartesian plane. The transmitters that can be exploited as sources of opportunity are many and range from the broadcast services for public utility to the local area wireless network up to satellite-based transmitters to communication and radio navigation. The choice will strictly depend on the characteristics required by the considered application. Typically, terrestrial transmitters for audio and video broadcasting services, such as analog FM radio, digital audio broadcasting (DAB), or digital video broadcasting – terrestrial (DVB-T), have been preferred, thanks to the high level of transmitted power which guarantees wide coverage and, in turn, enables the surveillance of wide areas. For shorter-range surveillance, local area networks signals, e.g., 5G, IEEE 802.11 WiFi, and LTE, represent a possibility, especially thanks to their wide bandwidth, which results in a good resolution. Satellite transmitters for communication and navigation, e.g., Global Navigation Satellite System (GNSS) or digital video broadcasting – satellite (DVB-S) signals, instead, represent a good candidate that could potentially offer global coverage, also in areas that are not covered by terrestrial transmitters, such as the open sea. Finally, also an existing radar could be used as transmitter of opportunity, benefitting from the good characteristics of waveform that has been designed for radar purposes. The parasitic nature of these PR sensors offers several advantages. Below, we recall the main ones: • PR operability does not require dedicated spectrum allocation. This characteristic overcomes some of the limitations of conventional active radars described in the introduction, such as the spectral resources demand and the risk of interference with other RF technologies. • The only need to design, implement, and manage the receiving part of the system means that PR systems require a reduced cost of construction and operation as well as facilitated maintenance. • The lack of emitted radiations and the reduced size of these sensors make them “eco-friendly” or “green” systems. In fact, they are especially suitable for coastal and urban areas monitoring and for preserving both the landscape integrity and the electromagnetic health of the population.

8 Passive Radar: A Challenge Where Resourcefulness Is the Key to Success

227

• The low vulnerability to electronic countermeasures, together with the above characteristics, also enables covert operations, making such systems extremely appealing for military purposes. Most of the above advantages stem from the lack of a dedicated cooperative transmitter, which in turn also represents one of the main drawbacks of these systems. In fact, PCL systems inherently employ waveforms that are not tailored for radar purposes without any possibility to control the radiative properties of the transmitters of opportunity. The largest development of these sensors is due to the purpose of using them in conventional surveillance applications, i.e., monitoring of aircraft and maritime targets. More recently, new development frontiers have been considered, aiming at short-range applications of targets characterized by less regular motions, e.g., drones, or for surveillance applications from moving platforms but also for imaging (Klemm et al. 2017; Malanowski 2019; Palmer et al. 2015). In this chapter, we will cover some of these applications, where the authors have been actively contributing, we will elaborate on the peculiarities of each one, and try to provide the reader with the right tools to understand the future of this fascinating technology. Based on this short introduction on passive radars, it is clear that the main challenge for those working with these sensors is to bring out the best of what already exists in order to achieve something for which it was not programmed. To achieve this goal, the key to success is a good dose of resourcefulness, a quality that fortunately we as women possess.

8.3 Passive Radar for Conventional Surveillance Applications This section of the chapter is devoted to the description of two well-established applications of the passive radar technology that significantly motivated the development and progress made in the past 20 years, namely, aerial and maritime surveillance, respectively, addressed in Sects. 8.3.1 and 8.3.2.

8.3.1 Passive Radar for Air Traffic Control One of the first and main application areas where passive radar sensors have been envisioned to be employed is air traffic control (ATC). The two main reasons for using these systems with this purpose are the need to guarantee a backup to conventional surveillance systems as well as to act as a gap-filler in case of lack of coverage of active radars. This certainly applies to defense scenarios where ATC systems could be the subject of targeted attacks aimed at disabling them but also to the blind areas of civilian ATC systems, e.g., at low altitudes.

228

F. Filippini and F. Colone

In long-range surveillance applications as the mentioned one, extensive coverage and great continuity should be guaranteed for PRs to be effectively employed. In this regard, the most effective transmitters to be employed as illuminators of opportunity are broadcast emitters, such as FM radio (Colone et al. 2013a, b; Howland et al. 2005; Saini and Cherniakov 2005; Zaimbashi 2016), digital audio, or video broadcasting services (DAB or DVB-T) (Berger et al. 2010; Coleman and Yardley 2008; Colone et al. 2014a; Howland et al. 2005; Palmer et al. 2013; Poullin 2005; Saini and Cherniakov 2005; Yardley 2007). In addition, this application requires high reliability and robustness to the conditions of the operating environment. Specifically, the performance must be kept reasonably constant regardless of changes in the propagation channel conditions or the presence of other transmitters in the area. However, by relying on third-part transmissions, such characteristics might be at risk despite the effectiveness of the developed processing techniques, and the corresponding requirements must be carefully addressed. One effective way to overcome these limitations is to exploit the information diversity gained in one or different domains. In this regard, one possibility is to exploit signals simultaneously transmitted by the same IO at different carrier frequencies. This is typically possible when using broadcast transmitters, for instance, FM radio transmitters, that usually emit on different channels at the same time. Strategies along this line have been proposed in (Martelli et al. 2016, 2017b, 2018, 2020c; Milani et al. 2018, 2020, 2021; Olsen and Woodbridge 2012a, 2012b), and they were shown to yield increased robustness with respect to the time-varying characteristics of the employed waveform as well as the propagation channel conditions. Another possible strategy is to resort to a proper combination of signals collected using differently polarized receiving antennas. The rationale behind the use of multi-polarimetric passive radars is the use of the polarimetric diversity to effectively counteract the disturbance contributions, e.g., interfering transmissions, while providing robustness against the target echo fading which results from its variable polarization. Several approaches have been considered for a proper exploitation of the polarimetric information to improve the target detection task, see, e.g., (Colone and Lombardo 2015; Conti et al. 2016; Filippini and Colone 2020b, 2021; Filippini et al. 2017) and the references therein. Finally, another possibility is represented by the joint use of the two information diversity sources, with the aim of efficiently leveraging signals collected by differently polarized antennas at different frequency channels (Colone and Lombardo 2016; Filippini and Colone 2021). An example of the results that can be obtained when properly exploiting the information diversity in a FM radio-based passive radar system for aerial surveillance is reported in Fig. 8.2. The employed system was equipped with two dual-polarized antennas (see Fig. 8.2a), one steered toward the area to be monitored and the other steered toward the transmitter. Moreover, the system enabled the simultaneous collection of FM radio signals on four different frequency channels. For the results shown in Fig. 8.2, the collected data were subdivided in coherent processing intervals (CPI) of 1 s each and processed according to the scheme reported in Filippini and Colone (2020a, 2021).

8 Passive Radar: A Challenge Where Resourcefulness Is the Key to Success

229

SURV-V

SURV-H REF-V REF-H

(a)

500

400

400

300

300

200

200

Bistatic Velocity [m/s]

Bistatic Velocity [m/s]

500

100 0 -100

100 0 -100

-200

-200

-300

-300

-400

-400

-500

-500 0

50 100 Relative Bistatic Range [km]

(b)

150

0

50 100 Relative Bistatic Range [km]

150

(c)

Fig. 8.2 Results with a FM radio-based passive radar for aerial surveillance: (a) dual-polarimetric reference and surveillance antennas; target detection results over 50 consecutive datafiles obtained when (b) using conventional single-channel scheme (c) exploiting both the polarimetric and frequency information diversity. © [2020] IEEE. (Reprinted, with permission from Ref. Filippini and Colone (2020a))

In both Fig. 8.2b, c, we report the target detection plots, namely, the points exceeding a given detection threshold, obtained for 50 consecutive datafiles, covering a time span of approx. 1 minute, on the same bistatic range-velocity plane. Specifically, continuous lines represent the true trajectories of the aircraft that were present in the monitored area at the time of the observation, while dots represent the target detection obtained with the passive radar that were correctly associated with the air-truth or marked as false alarms. The result shown in Fig. 8.2b is obtained when using a conventional passive radar system exploiting a single polarimetric and frequency channel among the available ones, while in Fig. 8.2c, we have properly exploited the information diversity offered by the available four frequency channels and two polarimetric channels. Observing Fig. 8.2b, c, we note that when using the conventional single channel processing, the system is able to detect targets with good continuity only up to 50 km. Note that this is the best performing combination of single frequency and single polarimetric channel among the available ones. The number of correctly detected targets significantly increases in Fig. 8.2c, thanks to the proper exploitation of both the polarimetric and temporal information. In particular, several additional detections are obtained on the trajectories of farthest aircraft.

230

F. Filippini and F. Colone

8.3.2 Passive Radar for Maritime Surveillance Along with the ATC, the other long-range surveillance application that motivated a lot of the development of passive radars in recent years is the maritime traffic monitoring. As mentioned in Sect. 8.2, one major advantage of passive radar is its low environmental impact, due to both its small size and the absence of a dedicated transmitter and therefore of electromagnetic pollution. Moreover, continuous illumination such as that provided by PCL systems exploiting broadcast transmissions enables the use of long coherent integration times on vessels moving at low speeds aiming at further increasing the coverage without significant migration loss. These features make this system an ideal candidate to be used for the surveillance of coastal areas, where the protection of the environment is a necessary aspect to be taken into account when installing new sensors and the accurate long-range surveillance is a required capability. Finally, it is possible to consider the use of a network of passive radar sensors, aimed at protecting long coastal areas. This scenario is qualitatively shown in Fig. 8.3. In a maritime surveillance application, the primary requirements are (i) good coverage, enabling the detection of targets several kilometers from the coast; (ii) good range resolution; and (iii) the ability to observe both large vessels such as container ships at long range and small vessels such as rubber or fishing boat closer to coast, e.g., approaching a harbor. For all these reasons, the waveform that has been considered the most suitable for the purpose is digital terrestrial television. DVB-T transmitters are, in fact, typically many and scattered on the territory, thus offering a remarkable coverage, which can be further increased if signals simultaneously transmitted by the same IO at multiple frequency channels are properly combined (Martelli et al. 2016, 2017a). Furthermore, the wide horizontal beam and high-power

Fig. 8.3 Illustrative scenario of a passive radar system for maritime surveillance

8 Passive Radar: A Challenge Where Resourcefulness Is the Key to Success

Surveillance Antennas

231

Reference Antenna

Targets of opportunity

(a)

(b)

30

2

1

20

Passive Radar receiver

10 Y (km)

Bistatic Velocity (m/s)

0

0

-1

-2

-10 -3 -20

-4

-30 0

2

4

6

8

10

-5

-5

-4

-3

-2

-1

Relative Bistatic Range (km)

X (km)

(c)

(d)

0

1

2

Fig. 8.4 Results of the DVB-T-based passive radar for short-range maritime surveillance: (a) Acquisition geometry; (b) cooperative target; target detection results obtained along 125 consecutive scans reported in (c) bistatic range-velocity domain and (d) X-Y domain

level of the broadcast transmitters enable their use even for long-range surveillance. Moreover, the low carrier frequencies of the considered source of opportunity offer the possibility of employing DVB-T-based PCL systems for over-the-horizon (OTH) maritime surveillance (Langellotti et al. 2014). Finally, the possibility of using long CPIs and managing very different target dynamics has been proved to be possible (Martelli et al. 2018, 2020b, c; Pignol et al. 2018). An example of results to show the effectiveness of this technology is reported in Fig. 8.4, for data collected in Riva di Traiano, Italy, with a DVB-T-based passive radar system equipped with three receiving antennas, steered toward the sea (see Fig. 8.4a). Figure 8.4 shows the target detection results obtained for 125 consecutive datafiles reported on the same plane, after a tracking stage. During the observed time, several boats where present (see, for instance, the circles in Fig. 8.4a), including a cooperative sailing boat (see Fig. 8.4b) that was approaching the coast (see the longest track). In the reported figure, the different tracks associated with the detected boats are reported on the bistatic range velocity plane (see Fig.

232

F. Filippini and F. Colone

8.4c). Then, based on the availability of more than one receiving antenna, also an estimation of the azimuth angle is provided enabling the same tracks to be projected onto the X-Y plane (see Fig. 8.4d), centered in the passive radar receiver location. To prove the long-range detection capability of the DVB-T-based maritime passive radar, we report in Fig. 8.5 the results obtained for a test conducted in Leghorn, Italy (Langellotti et al. 2014). The DVB-T-based PCL receiver installation is sketched in Fig. 8.5a, while the target detection results for two consecutive tests of 7 minutes each are shown in Fig. 8.5b. Continuous tracks identify the maritime traffic as provided by the AIS live registration, while circles indicate the PCL results after a standard track initiation procedure is applied against the raw detections obtained over consecutive scans for the two considered bursts, respectively. Remarkably, we are able to detect and correctly localize the target on the X-Y plane even beyond the standard radar horizon.

8.4 Advanced Applications of Passive Radar Once we have covered the most conventional applications of passive radar sensors, in this section, we will go through some of the more advanced applications of passive radar that are expected to drive further development on this technology in the next few years. In particular, Sect. 8.3.1 will focus on the use of passive radar sensors on moving platform; Sect. 8.4.2 will address the exploitation of PCL systems for drone surveillance based on different IOs; and Sect. 8.4.3 will cover the most recent application of passive radar for monitoring of human targets in indoor scenarios.

8.4.1 Passive Radar Onboard Moving Platform Among the most interesting and challenging recent applications of passive radar systems is the use of PR sensors mounted on airborne or ground moving platforms. Passive radar on moving platform offers a number of advantages with respect to stationary ground-based applications and enables advanced imaging and surveillance functionalities (Blasone et al. 2020, 2021; Brown et al. 2012; Dawidowicz et al. 2012a, b; Palmer et al. 2017; Tan et al. 2014; Wojaczek et al. 2019, 2021; Wu et al. 2016; Yang et al. 2017). The development in this field is also fostered by the progress in small and powerful hardware, as well as the adoption of commercial-off-the-shelf (COTS) hardware components for the receiving system which led to reductions of weight, energy consumption, and cost. That makes it feasible to install mobile multichannel passive radars on small vehicles and boats, light aircrafts, or unmanned aerial vehicles (UAVs), representing an appealing solution for covert monitoring operations over wide areas. A qualitative representation of this application is shown in

8 Passive Radar: A Challenge Where Resourcefulness Is the Key to Success

233

Surveillance antenna array

Reference antenna

RX & processing

(a) 20 10 0

Y (km)

-10 TGT B

-20

TGT E

-30 -40 -50 TGT A TGT C

-60

TGT D -70 -80 -100

-90

-80

-70

-60

-50 -40 X (km)

-30

-20

-10

0

(b) Fig. 8.5 Results of the DVB-T-based passive radar for long-range maritime surveillance: (a) Acquisition geometry and installation (b) target detection results in the X-Y plane for two consecutive tests of 7 minutes each. © [2014] IEEE. (Reprinted, with permission from Ref. Langellotti et al. (2014))

234

F. Filippini and F. Colone

Fig. 8.6 Illustrative scenario of passive radar systems onboard moving platforms

Fig. 8.6. In addition to conventional military surveillance applications, for which these sensors are particularly well suited given their ability to perform covert monitoring, this technology might also have civilian applications. For example, these sensors could be integrated in the aim to facilitate navigation or assisted driving on board of vehicles, either ground based or maritime. In this purpose, the PRs could be easily employed without suffering from interference nor increasing the energy consumption on board of the vehicle. Along with the benefits of passive radar on moving platforms, this application also brings a number of challenges, mainly due to the presence of Doppler distortions on the received signals, induced by the relative motion of the receiver with respect to the stationary scene. This drawback might severely affect the performance of the system since the detection of targets characterized by small radial velocity components is hindered by the Doppler spread of clutter echoes. These problems have been addressed in the recent literature by resorting to an appropriate exploitation of space-time adaptive processing (STAP) applied to the signals collected by multiple receiving channels (Blasone et al. 2020, 2021; Wojaczek et al. 2019). The simplest approach is based on the displaced phase center antenna (DPCA) concept and requires the availability of a pair of antennas on receive. It operates by performing a subtraction of radar echoes received by the two channels at the time that their two-way phase centers occupy the same spatial position (Klemm et al. 2017; Klemm 1998, 2004). This approach has been shown to be effective in active pulsed radar, provided that the internal clutter motion is limited. However, when exploiting time-varying waveforms of opportunity, e.g., DVB-T signals, the effectiveness of this approach is limited. In fact, one can no longer rely on the paradigm of equivalent observations of the scene from

8 Passive Radar: A Challenge Where Resourcefulness Is the Key to Success

235

Fig. 8.7 Results of a DVB-T based passive radar on moving platform: (a) The location of the measurement campaign (b) the boat used as PCL-receiver equipped with the Parasol system; Range-Doppler map obtained for experimental data after: (c) MF; (d) STAP after RpF © [2019] IEEE. (Reprinted, with permission from Ref. Wojaczek et al. (2019))

the two receiving antennas since the employed waveform changes on subsequent observations. This effect limits the capability of filtering out the echoes from the stationary scene thus jeopardizing the detection of moving target echoes. A solution to this problem has been presented in (Wojaczek et al. 2019), by resorting to the STAP approach after reciprocal filter (RpF) and its effectiveness is shown on experimental data in Fig. 8.7d with respect to the conventional matched filter solution in Fig. 8.7c. The data were collected by Fraunhofer FHR and the Norwegian Defence Research Establishment (FFI) in Oslo fjord (Fig. 8.7a). The DVB-T-based receiving system, equipped with two surveillance antennas, was mounted on the boat shown in Fig. 8.7b. Figure 8.7c shows the range-Doppler map obtained at a single surveillance channel for a CPI of 0.57 s, with a matched filter, along with the enlarged view of range-Doppler regions where possible targets are present. The targets included in the two regions very close to the strong clutter returns cannot be discriminated, while the third target is sufficiently isolated and

236

F. Filippini and F. Colone

can be reasonably distinguished, even if with a local signal-to-clutter plus noise ratio (SCNR) lower than 18 dB. Figure 8.7d, on the other hand, shows the result of the RpF-based STAP approach. This strategy allows filtering out both the stationary scatterers main peaks and their sidelobes structures so that all three targets now appear as isolated peaks and they can be easily detected against the residual background.

8.4.2 Passive Radar for Drone Surveillance In recent years, drones and UAVs have become widespread and have been used for a variety of applications, such as search and rescue operations, environmental monitoring, as well as aerial photography and video. Beyond their harmless applications, the widespread use of civilian drones has also caused several personal privacy and public safety issues (De Cubber 2019; Ritchie et al. 2017). For instance, these objects are posing a serious threat to aviation safety and have caused entire airports to be shut down several times due to aircraft-drone collisions or UAVs hovering above the airport area. Therefore, accurate detection, tracking, and classification of these objects have become key requirements for surveillance systems aimed at providing accurate and reliable monitoring of critical infrastructures and protected areas; see, for instance, Fig. 8.8. For this reason, different anti-drone sensors have been developed, including

Fig. 8.8 Illustrative application for passive radar systems for drone surveillance

8 Passive Radar: A Challenge Where Resourcefulness Is the Key to Success

237

radars. However, there are several restrictions for radar installation in populated areas and/or any other site where electromagnetic emissions are limited by current regulations. In addition, it is necessary to limit the risk of interference with other RF systems that typically operate in the above environments. In such scenarios, passive radar technology is a compelling alternative to the use of conventional active radar systems. The potentialities of passive radar in counter drone operations, based on different RF waveforms of opportunity, have been investigated in several works recently appeared in the technical literature; see, e.g., (Blasone et al. 2020; Cabrera et al. 2020; Ilioudis et al. 2020; Jarabo-Amores et al. 2018, 2021; Martelli et al. 2020b; Milani et al. 2018, 2020, 2021; Schüpbach et al. 2017; Ummenhofer et al. 2020). The small size of the target and its unpredictable motion set specific requirements on the achievable performance and, in turn, on the employed IO. For instance, to achieve a very high range resolution, waveforms of opportunity characterized by large frequency bandwidths must be preferred. On the other hand, if the capability of detecting targets is required at longer ranges, it is necessary to operate with IOs that offer good coverage and high received power levels. For this reason, the idea of a multisensory passive radar system was proposed in (Lombardo et al. 2021). The motivating idea is to leverage the good characteristics of different sources of opportunity, using each subsystem for a different purpose. A DVB-T-based passive radar, which offers the widest coverage but the worst range and Doppler resolution, could detect the drone at longer ranges and alert shortrange sensors, based on WiFi and on DVB-S, which provide a target localization with much better accuracy than the DVB-T-based sensor. Additionally, WiFi-based sensors can also exploit the transmissions of the drone itself, which typically emit signals in the WiFi band, enabling the joint use of passive radar and device-based passive location approaches (Milani et al. 2018, 2020, 2021). A qualitative sketch of this configuration is sketched in Fig. 8.8. To demonstrate the feasibility of drone detection using PR, the authors along with their research group carried out several experimental trials. As an example, we consider in Fig. 8.9 the data collected during three different acquisition campaigns, whose geometries are illustrated in Fig. 8.9a, c, e, where either one or two drones were used as cooperative targets and different waveforms of opportunity were used. Figure 8.9b, d, f, instead, report the drone detection results in the bistatic rangeDoppler (or range-velocity) plane. Specifically, Fig. 8.9b is obtained with the DVBT-based AULOS system from Leonardo S.p.A. and shows that two different small UAVs can be detected up to medium ranges, e.g., few kilometers (see test area B in Fig. 8.9a) (Martelli et al. 2020b). On the contrary, both the DVB-S- and WiFi-based sensors (Fig. 8.9d (Milani et al. 2021) and Fig. 8.9f) allow detecting drones up to a few dozens of meters but with high accuracy. Additional insights on these results are reported in Chapter 9 of (Blasone et al. 2020).

238

F. Filippini and F. Colone

Fig. 8.9 Examples of drone detections with passive radar based on (a, b) DVB-T © [2019] IET. (Reprinted, with permission Ref. Martelli et al. (2020b)). (c, d) DVB-S © [2020] IEEE. (Reprinted, with permission Ref. Martelli et al. (2020a)). (e, f) WiFi (Martelli et al. 2017b)

8.4.3 Passive Radar for Human Activity Monitoring Nowadays, passive radar technology is becoming an attractive technology also for monitoring of indoor environments, both residential and commercial. In particular, passive radar sensors based on WiFi transmissions have become attractive candidates for wireless sensing. This is especially due to the following main reasons: (i) the availability of WiFi transmitters has been rapidly increasing in modern

8 Passive Radar: A Challenge Where Resourcefulness Is the Key to Success

239

Fig. 8.10 Illustrative application for passive radar systems in indoor scenarios

buildings; therefore, WiFi-based sensors can represent a low-cost solution that is based on existing infrastructures, and (ii) due to its nonintrusive nature, WiFibased passive radars can be employed in the surveillance of an indoor area without introducing any privacy issue or discomfort and without requiring any cooperation from the user such as the need to carry wearable devises an illustrative surveillance scenario is shown in (Fig. 8.10). The range of application of such sensors can vary from the detection and tracking of intruders entering an indoor area to the monitoring of human gestures useful for smart-home applications such as energy saving but can also include ehealthcare scenarios where the collected data can be used aiming at recognizing falls, monitoring the breath activity, etc. The effectiveness of WiFi-based passive radars for indoor monitoring has been demonstrated in the technical literature (Chen et al. 2020; Chetty et al. 2012; Colone 2011, 2017; Colone et al. 2012, 2014b, 2017; Falcone et al. 2012; Li et al. 2021; Milani et al. 2021; Pastina et al. 2015; Sun et al. 2021; Tan et al. 2016). The major limitation to the use of this concept for the aforementioned applications is the resolution that can be achieved in the range domain. In fact, typical WiFi transmissions are characterized by a bandwidth between 20 and 40 MHz. This characteristic corresponds to an equivalent minimum monostatic range resolution of approx. than 8 m. More recent IEEE 802.11 standards include transmissions with a bandwidth of 80 or 160 MHz, which would allow increasing the range resolution. However, even in the wider bandwidth case, the achieved resolution makes the range information not very useful for discriminating targets in scenarios with limited size. Therefore, targets should be discriminated in alternative domains, for instance, using the Doppler (Li et al. 2021) or angle information.

240

F. Filippini and F. Colone

Fig. 8.11 Real target trajectory (a) and WiFi passive radar localization results (b). © [2015] IEEE. (Reprinted, with permission from Ref. Pastina et al. (2015))

In this section, we report two different results obtained against human targets. For the first test, two surveillance antennas were employed to provide angular localization capability in a quasi-monostatic configuration in a large indoor area. In the test reported in Fig. 8.11, two men move along partially overlapping paths (see real trajectories in Fig. 8.11a). In particular, they initially walk on the same direction with a constant separation of about 1 m; then they abruptly change their heading, going toward opposite sides of the room. The localization results are reported in Fig. 8.11b in a Cartesian plane, whose origin is located on the transmitter location and defined with the X axis aligned with the short side of the hall and the Y axis oriented toward its center. Figure 8.11b shows that the system can correctly detect and track the two targets, with only small deviations with respect to their real trajectories. Note that the two targets can be distinguished during the entire acquisition time of approx. 25 s because they are moving toward opposite directions, i.e., they show opposite Doppler frequencies. It would be much more difficult to distinguish them based on the range information because, as mentioned above, the achievable range resolution of the system is quite limited. An example of a Doppler only result is reported in Fig. 8.12, where a single human target is moving in a smaller indoor area (Fig. 8.12b) according to the trajectory reported in Fig. 8.12a. Specifically, the target first gets closer to the RX-TX pair, mounted in a quasi-monostatic configuration, and then it moves away from it. The target signature is observed in the bistatic velocity-time domain (see Fig. 8.12c), extracted from the first range cell and normalized with respect to an estimated background level. The evolution of the target’s bistatic velocity corresponds to the reported trajectory. Furthermore, it can be observed that periodic peaks are present around the main signature, which correspond to the periodic

8 Passive Radar: A Challenge Where Resourcefulness Is the Key to Success

RXTX

0

TX RX

-2

Y(m)

241

-4

-6

-8

-10 -2

-1.5

-1

-0.5

0 X(m)

0.5

1

1.5

2

(a)

(b)

10

5 20 dB

Bistatic velocity (m/s)

30

0

10

-5

-10 0

2

4

6 Time(s)

8

10

12

0

(c) Fig. 8.12 Results of WiFi-based passive radar system in indoor scenario: (a) Human target trajectory (b) acquisition campaign and (c) target Doppler signature

movement of the arms and legs, in addition to the movement of the main target body. The ability to analyze the micro-Doppler effects induced by the additional movement of the body parts enables the classification of the movement and in turn allows more advanced applications such as the monitoring of elderly people.

242

F. Filippini and F. Colone

8.5 Conclusion In this chapter, we have reviewed and described the most recent, innovative, and challenging applications of passive radar technology. We have shown how a parasitic exploitation of third-party illuminators of opportunity can allow the use of these sensors for a wide variety of purposes. In particular, broadcast signal-based passive radar is nowadays a mature technology to be widely employed for long-range surveillance applications, such as aerial and maritime monitoring. Moreover, we have shown how the nice characteristics of PCL sensors could be successfully exploited in several novel applications, ranging from the surveillance of small UAVs to the monitoring of human target in indoor scenario. With this chapter, we hope to have succeeded in showing the readers all the possibilities offered by this fascinating concept. We are excited about which challenge the future holds for this technology, and we are sure that women engineers will be ready to bring out a good dose of resourcefulness and willpower to face it. Acknowledgments The authors would like to gratefully acknowledge the collaboration of many colleagues at Sapienza University of Rome. We are extremely thankful to all the colleagues of the entire Radar Remote Sensing and Navigation research group whose support was essential to the realization of the work reported in this chapter.

References Baker CJ, Griffiths HD, Papoutsis I (2005) Passive coherent location radar systems. Part 2: waveform properties. IEE Proc Radar Sonar Navig 152:160–168 Berger CR, Demissie B, Heckenbach J, Willett P, Zhou S (2010) Signal processing for passive radar using OFDM waveforms. IEEE J Sel Top Sign Proces 4(1):226–238 Blasone GP, Colone F, Lombardo P, Wojaczek P, Cristallini D (2020) Passive radar DPCA schemes with adaptive channel calibration. IEEE Trans Aerosp Electron Syst 56(5):4014–4034 Blasone GP, Colone F, Lombardo P, Wojaczek P, Cristallini D (2021) Passive radar STAP detection and DoA estimation under antenna calibration errors. IEEE Trans Aerosp Electron Syst 57(5):2725–2742 Brown J, Woodbridge K, Griffiths H, Stove A, Watts S (2012) Passive bistatic radar experiments from an airborne platform. IEEE Aerosp Electron Syst Mag 27:50–55 Cabrera O, Bongioanni C, Filippini F, Sarabakha O, Colone F, Lombardo P (2020) Detecting drones and human beings with DVB-S based COTS passive radar for short-range surveillance. In: 2020 IEEE international radar conference, Washington DC Chen Q, Liu Y, Tan B, Woodbridge K, Chetty K (2020) Respiration and activity detection based on passive radio sensing in home environments. IEEE Access 8:12426–12437 Chetty K, Smith GE, Woodbridge K (2012) Through-the-wall sensing of personnel using passive bistatic WiFi radar at standoff distances. IEEE Trans Geosci Remote Sens 50(4):1218–1226 Coleman C, Yardley H (2008) Passive bistatic radar based on target illuminations by digital audio broadcasting. IET Radar Sonar Navig 2(5) Colone F (2011) Ambiguity function analysis of wireless LAN transmissions for passive radar. IEEE Trans Aerosp Electron Syst 47(1):240–264

8 Passive Radar: A Challenge Where Resourcefulness Is the Key to Success

243

Colone F (2017) Short-range passive radar potentialities. In: Klemm R, Nickel U, Gierull C, Lombardo P (eds) Novel radar techniques and applications – volume 1: real aperture array radar, imaging radar, and passive and multistatic radar. IET Publisher Colone F, Lombardo P (2015) Polarimetric passive coherent location. IEEE Trans Aerosp Electron Syst 51(2):1079–1097 Colone F, Lombardo P (2016) Non-coherent adaptive detection in passive radar exploiting polarimetric and frequency diversity. IET Radar Sonar Navig 10(1):15–23 Colone F, Falcone P, Bongioanni C, Lombardo P (2012) WiFi-based passive bistatic radar: data processing schemes and experimental results. IEEE Trans Aerosp Electron Syst 48(2):1061– 1079 Colone F, Bongioanni C, Lombardo P (2013a) Multi-frequency integration in FM radio based passive bistatic radar. Part I: target detection. IEEE Aerosp Electron Syst Mag 28(4):28–39 Colone F, Bongioanni C, Lombardo P (2013b) Multi-frequency integration in FM radio based passive bistatic radar. Part II: direction of arrival estimation. IEEE Aerosp Electron Syst Mag 28(4):40–47 Colone F, Langellotti D, Lombardo P (2014a) DVB-T signal ambiguity function control for passive radars. IEEE Trans Aerosp Electron Syst 50(1):329–347 Colone F, Pastina D, Falcone P, Lombardo P (2014b) WiFi-based passive ISAR for high-resolution cross-range profiling of moving targets. IEEE Trans Geosci Remote Sens 52(6):3486–3501 Colone F, Palmarini C, Martelli T, Tilli E (2016) Sliding extensive cancellation algorithm for disturbance removal in passive radar. IEEE Trans Aerosp Electron Syst 52(3):1309–1326 Colone F, Martelli T, Lombardo P (2017) Quasi-monostatic versus near forward scatter geometry in WiFi-based passive radar sensors. IEEE Sensors J 17(15):4757–4772 Colone F, O’Hagan DW, Lombardo P, Baker CJ (2009) A multistage processing algorithm for disturbance removal and target detection in passive bistatic radar. IEEE Trans Aerosp Electron Syst 45(2):698–722 Conti M, Moscardini C, Capria A (2016) Dual-polarisation DVB-T passive radar: experimental results, In: Proceedings of the IEEE national radar conference 2016, pp 1–5 Dawidowicz B, Kulpa K, Malanowski M, Misiurewicz J, Samczynski P, Smolarczyk M (2012a) DPCA detection of moving targets in airborne passive radar. IEEE Trans Aerosp Electron Syst 48:1347–1357 Dawidowicz B, Samczynski P, Malanowski M, Misiurewicz J, Kulpa KS (2012b) Detection of moving targets with multichannel airborne passive radar. IEEE Aerosp Electron Syst Mag 27:42–49 De Cubber G (2019) Explosive drones: how to deal with this new threat?. In: Proceedings of the 9th international workshop on measurement, prevention, protection and management of CBRN risks, Belgium, pp 1–8 Falcone P, Colone F, Lombardo P (2012) Potentialities and challenges of WiFi-based passive radar. IEEE Aerosp Electron Syst Mag 27(11):15–26 Filippini F, Colone F (2020a) Multi-carrier adaptive detection in polarimetric passive radars. In: 2020 IEEE radar conference (RadarConf20), pp 1–6 Filippini F, Colone F (2020b) Polarimetric passive radar: a practical approach to parametric adaptive detection. IEEE Trans Aerosp Electron Syst 56:4930–4946 Filippini F, Colone F (2021) Multi-carrier and multi-polarimetric model based adaptive target detector for passive radar systems. IET Radar Sonar Navig 15:853–866 Filippini F, Colone F, Cristallini D, Bournaka G (2017) Experimental results of polarimetric detection schemes for DVB-T-based passive radar. IET Radar Sonar Navig 11(6):883–891 Filippini F, Martelli T, Colone F, Cardinali R (2018) Target DoA estimation in passive radar using non-uniform linear arrays and multiple frequency channels. In: 2018 IEEE radar conference (RadarConf18), Oklahoma City, OK, pp 1290–1295 Gordon W (1985) A hundred years of radio propagation. IEEE Trans Antennas Propag 33(2):126– 130 Griffiths H (2013) The German WW2 HF radars elefant and see-elefant. IEEE Aerosp Electron Syst Mag 28(1):4–12

244

F. Filippini and F. Colone

Griffiths H, Baker C (2005) Passive coherent location radar systems. Part 1: performance prediction. IEE Proc Radar Sonar Navig 152:153–159 Griffiths H, Baker C (2014) Passive bistatic radar. In: Melvin WL, Scheer JA (eds) Principles of modern radar, Radar applications, vol III. SciTech Publishing, Edison, pp 499–541 Griffiths HD, Baker CJ (2017) An introduction to passive radar. Norwood, Artech House Griffiths H et al (2015) Radar spectrum engineering and management: technical and regulatory issues. Proc IEEE 103(1):85–102 Griffiths HD, Knott P, Koch W (2019) Christian Hülsmeyer: invention and demonstration of radar, 1904. IEEE Aerosp Electron Syst Mag 34(9):56–60 Howland PE, Maksimiuk D, Reitsma G (2005) FM radio based bistatic radar. IEE Proc Radar Sonar Navig 152:107–115 Ilioudis C, Clemente C, Soraghan J (2020) GNSS-based passive UAV monitoring: a feasibility study. IET Radar Sonar Navig 14(4):516–524 Jarabo-Amores MP et al (2018) Drone detection feasibility with passive radars. In: 2018 15th European radar conference (EuRAD), pp 313–316 Jarabo-Amores M-P et al (2021) DVB-T-based passive radar for silent surveillance of drones. In: Clemente C, Fioranelli F, Colone F, Li G (eds) Radar countermeasures for unmanned aerial vehicles. SciTech Publishing Klemm R (1998) Introduction to space-time adaptive processing. In: IEE colloquium on space-time adaptive processing (Ref. No. 1998/241), pp 1/1–111 Klemm R (2004) Applications of space-time adaptive processing. IEE Publishing Klemm R et al (eds) (2017) Novel radar techniques and applications, Part III: passive and multistatic radar. IET Publisher Langellotti D et al (2014) Over the horizon maritime surveillance capability of DVB-T based passive radar. In: 2014 44th European microwave conference, pp 1812–1815 Li W, Piechocki RJ, Woodbridge K, Tang C, Chetty K (2021) Passive WiFi radar for human sensing using a stand-alone access point. IEEE Trans Geosci Remote Sens 59(3):1986–1998 Lombardo P, Colone F (2012) Advanced processing methods for passive bistatic radar. In: Melvin WL, Scheer JA (eds) Principles of modern radar: advanced radar techniques. SciTech Publishing, Raleigh, pp 739–821 Lombardo P et al (2021) Multiband passive radar for drones detection and localization. In: Clemente C, Fioranelli F, Colone F, Li G (eds) Radar countermeasures for unmanned aerial vehicles. SciTech Publishing Malanowski M (2019) Signal processing for passive bistatic radar. Artech House, Norwood Marconi SG (1922) Radio telegraphy. Proc Inst Radio Eng 10(4):215–238 Martelli T, Colone F, Tilli E, Di Lallo A (2016) Multi-frequency target detection techniques for DVB-T based passive radar sensors. Sensors 16:1594 Martelli T, Colone F, Tilli E, Di Lallo A (2017a) Maritime surveillance via multi-frequency DVB-T based passive radar In: 2017 IEEE radar conference (RadarConf), pp 0540–0545 Martelli T, Murgia F, Colone F, Bongioanni C, Lombardo P (2017b) Detection and 3D localization of ultralight aircrafts and drones with a WiFi-based passive radar. In: Proceedings of the 2017 international conference on radar systems (Radar 2017), Belfast, UK, pp 1–6 Martelli T, Filippini F, Pignol F, Colone F, Cardinali R (2018) Computationally effective range migration compensation in PCL systems for maritime surveillance. In: 2018 IEEE radar conference (RadarConf18), pp 1406–1411 Martelli T, Cabrera O, Colone F, Lombardo P (2020a) Exploitation of long coherent integration times to improve drone detection in DVB-S based passive radar. In: 2020 IEEE radar conference (RadarConf20), pp 1–6 Martelli T, Colone F, Cardinali R (2020b) DVB-T based passive radar for simultaneous counter drone operations and civil air traffic surveillance. IET Radar Sonar Navig 14(4):505–515 Martelli T, Filippini F, Colone F (2020c) Tackling the different target dynamics issues in counter drone operations using passive radar. In 2020 IEEE international radar conference (RADAR), pp 512–517

8 Passive Radar: A Challenge Where Resourcefulness Is the Key to Success

245

Milani I, Colone F, Lombardo P (2018) 2D localization with WiFi passive radar and device-based techniques: an analysis of target measurements accuracy. In: 2018 19th international radar symposium (IRS), pp 1–10 Milani I, Bongioanni C, Colone F, Lombardo P (2020) Fusing active and passive measurements for drone localization. In: 2020 21st international radar symposium (IRS), pp 245–249 Milani I, Bongioanni C, Colone F, Lombardo P (2021) Fusing measurements from Wi-Fi emissionbased and passive radar sensors for short-range surveillance. Remote Sens 13:3556 Olsen KE, Woodbridge K (2012a) Performance of a multiband passive bistatic radar processing scheme—Part I. IEEE Aerosp Electron Syst Mag 27(10):16–25 Olsen KE, Woodbridge K (2012b) Performance of a multiband passive bistatic radar processing scheme-Part II. IEEE Aerosp Electron Syst Mag 27(11):4–14 Palmer JE, Harms HA, Searle SJ, Davis L (2013) DVB-T passive radar signal processing. IEEE Trans Signal Process 61(8):2116–2126 Palmer J, Cristallini D, Kuschel H (2015) Opportunities and current drivers for passive radar research. In: IEEE radar conference, Johannesburg Palmer J et al (2017) Receiver platform motion compensation in passive radar. IET Radar Sonar Navig 11:922–931 Pastina D, Colone F, Martelli T, Falcone P (2015) Parasitic exploitation of Wi-Fi signals for indoor radar surveillance. IEEE Trans Veh Technol 64(4):1401–1415 Pignol F, Colone F, Martelli T (2018) Lagrange-polynomial-interpolation-based keystone transform for a passive radar. IEEE Trans Aerosp Electron Syst 54(3):1151–1167 Poullin D (2005) Passive detection using digital broadcasters (DAB DVB) with COFDM modulation. Proc. Inst. Electr. Eng. Radar Sonar Navig 152:143–152 Ritchie M, Fioranelli F, Borrion H (2017) Micro UAV crime prevention: can we help Princess Leia? In: Savona BL (ed) Crime prevention in the 21st century. Springer, New York, pp 359–376 Saini R, Cherniakov M (2005) DTV signal ambiguity function analysis for radar application. IEE Proc Radar Sonar Navig 152:133–142 Schüpbach C, Patry C, Maasdorp F et al (2017) Micro-UAV detection using DAB-based passive radar. In: IEEE radar conference, Seattle, WA, pp 1037–1040 Sun H, Chia LG, Razul SG (2021) Through-wall human sensing with WiFi passive radar. IEEE Trans Aerosp Electron Syst 57(4):2135–2148 Tan DKP, Lesturgie M, Sun H, Lu Y (2014) Space-time interference analysis and suppression for airborne passive radar using transmissions of opportunity. IET Radar Sonar Navig 8(2):142– 152 Tan B, Woodbridge K, Chetty K (2016) Awireless passive radar system for real-time through-wall movement detection. IEEE Trans Aerosp Electron Syst 52(5):2596–2603 Ummenhofer M, Lavau LC, Cristallini D, O’Hagan D (2020) UAV micro-doppler signature analysis using DVB-S based passive radar. In: 2020 IEEE international radar conference (RADAR), pp 1007–1012 Wojaczek P, Colone F, Cristallini D, Lombardo P (2019) Reciprocal-filter-based STAP for passive radar on moving platforms. IEEE Trans Aerosp Electron Syst 55(2):967–988 Wojaczek P, Cristallini D, O’Hagan DW, Colone F, Blasone GP, Lombardo P (2021) A threestage inter-channel calibration approach for passive radar on moving platforms exploiting the minimum variance power spectrum. Sensors 21(1):69 Wu Q, Zhang YD, Amin MG, Himed B (2016) Space–time adaptive processing and motion parameter estimation in multistatic passive radar using sparse Bayesian learning. IEEE Trans Geosci Remote Sens 54(2):944–957 Yang P-C, Lyu X-D, Chai Z-H, Zhang D, Yue Q, Yang J-M (2017) Clutter cancellation along the clutter ridge for airborne passive radar. IEEE Geosci Remote Sens Lett 14(6):951–955 Yardley HJ (2007) Bistatic radar based on DAB ifluminators: the evolution of a practical system. IEEE Aerosp Electron Syst Mag 22(11):13–16 Zaimbashi A (2016) Multiband FM-based passive bistatic radar: target range resolution improvement. IET Radar Sonar Navig 10(1):174–185

246

F. Filippini and F. Colone Francesca Filippini received her M.Sc. degree (cum Laude) in Communication Engineering and the Ph.D. degree in Radar and Remote Sensing, from Sapienza University of Rome, in 2016 and 2020, respectively. From January to May 2016, she has been working on her Master Thesis with the Passive Radar and Antijamming Techniques Department at Fraunhofer Institute FHR. She is currently a Post-Doctoral Researcher with the Department of Information Engineering, Electronics and Telecommunications at Sapienza University of Rome. Dr. Filippini received the 2020 IEEE AESS Robert T. Hill Best Dissertation Award for her Ph.D. thesis and the 2020 GTTI Best PhD Thesis Award defended at an Italian University in the areas of communications technology. She also received the 2018 Premium Award for the Best Paper in IET Radar, Sonar & Navigation, the Best Paper Award at the 2019 Int. Radar Conference, the second Best Student Paper Award at the 2018 IEEE Radar Conference and the Best Paper Award at the 2017 GTTI Workshop on Radar and Remote Sensing. She is a member of the IEEE Aerospace and Electronic System Society (AESS) Board of Governors, where she is currently serving as Co-Editor in Chief for the IEEE AESS QEB Newsletters and Secretary.

Fabiola Colone received the master degree in Telecommunications Engineering and the Ph.D. degree in Remote Sensing from Sapienza University of Rome, Italy, in 2002 and 2006, respectively. She joined the DIET Dept. of Sapienza University of Rome as a Research Associate in January 2006. From December 2006 to June 2007, she was a Visiting Scientist at the Electronic and Electrical Engineering Dept. of the University College London, London, UK. She is currently a Full Professor at the Faculty of Information Engineering, Informatics, and Statistics of Sapienza University of Rome, where she serves as Chair of the degree programs in Communications Engineering. The majority of Dr. Colone’s research activity is devoted to radar systems and signal processing. She has been involved, with scientific responsibility roles, in research projects funded by the European Commission, the European Defence Agency, the Italian Space Agency, the Italian Ministry of Research, and many radar/ICT companies. Her research has been reported in over 170 publications in international technical journals, book chapters, and conference proceedings. Dr. Colone is co-editor of the book “Radar Countermeasures for Unmanned Aerial Vehicles,” IET Publisher. She has been co-recipient of the 2018 Premium Award for Best Paper in IET Radar, Sonar & Navigation. From 2017 to 2022, she was member of the Board of Governors of the IEEE Aerospace and Electronic System Society (AESS) in which she served as Vice-President for Member Services, and Editor in Chief for the IEEE AESS QEB Newsletters. She is IEEE Senior Member from 2017 and member of the IEEE AESS Radar System Panel from 2019. Dr. Colone is the Associate Editor in Chief for the IEEE Transactions on Radar Systems. She was Associate Editor for the IEEE Transactions on Signal Processing from 2017 to 2020, and she is a member of the Editorial Board of the Int. Journal of Electronics and

8 Passive Radar: A Challenge Where Resourcefulness Is the Key to Success

247

Communications (Elsevier). She was Technical Co-Chair of the IEEE 2021 Radar Conference (Atlanta, USA) and of the European Radar Conference EuRAD 2022 (Milan, Italy), and she served in the organizing committee and in the technical program committee of many international conferences.

Chapter 9

Remote Sensing Through Satellites and Sensor Networks Silvia Liberata Ullo and Afreen Siddiqi

9.1 Remote Sensing: An Overview Remote sensing (RS) in contemporary conception is a discipline, and also refers to a process, focused on technologies for observing objects and their characteristics from a distance through the use of sensors. The sensors may be located on aerial- or spaceborne platforms, towers on the ground, or locations on the Earth’s surface. The distinguishing feature of RS is of acquiring quantitative and qualitative information without direct physical contact with objects of observation. The birth of RS can be traced with the invention and development of photographic techniques. Some of the impetus for developing observation capabilities from high elevations came from military operational needs, and the first air-borne platforms for RS for such purposes were balloons. In parallel, beyond military applications, there was increasing interest (and curiosity) to observe regions from above, from other points of view, and for different purposes. There were great contributions of photographers, who sought high elevations to capture new images and unique perspectives of the world, in advancing use of aerial platforms for RS. With the invention of airplanes, RS was used for studying environmental problems, and applications started to extend beyond simple imagery and Earth observation (EO). Finally, with the birth of the space age, and advent of spacecraft that could be purposefully deployed to orbit the Earth (satellites), it has become

S. L. Ullo () University of Sannio, Benevento, Italy e-mail: [email protected] A. Siddiqi Engineering Systems Laboratory, Department of Aeronautics and Astronautics, Massachusetts Institute of Technology, Cambridge, MA, USA Harvard Kennedy School, Cambridge, MA, USA e-mail: [email protected] © The Author(s), under exclusive license to Springer Nature Switzerland AG 2023 M. S. Greco et al. (eds.), Women in Telecommunications, Women in Engineering and Science, https://doi.org/10.1007/978-3-031-21975-7_9

249

250

S. L. Ullo and A. Siddiqi

possible to acquire detailed images with many possibilities of use including: to study cloud masses (meteorology), trees and plants (forestry and agronomy), the Earth’s surface and sub-surface (geology), the surface water distribution (hydrology), and inhabited areas (urban planning). Many different applications through the analysis of data collected from sensors onboard satellites and from sensors connected in mobile or fixed networks (wired or wireless) have been developed over time. In the next sections, some background will be given on satellite RS and RS through sensor networks, and some applications of data acquired from these systems will be discussed. It is worth highlighting that satellite RS and data collected through sensor networks, on the ground or over the sea, represent two limits in a certain sense. On one end, RS from satellites allows for constructing a macro-scale understanding of a phenomenon, by covering very wide areas, and by collecting specific information based on the characteristics of the sensor onboard. The sensor networks, on the other end, can collect information at a micro-scale with accuracy by measuring environment parameters and other useful data within specific locations. In between, there are RS payloads on air-borne platforms on unmanned aerial vehicles (UAVs) or more simply drones, and manned aircraft, that can observe at the meso-scale. Here, we will primarily focus on RS from the surface (through sensor networks), RS from space (through satellites), and will discuss how the joint use of data from both can help in constructing a better understanding of important phenomena affecting our planet and human societies. Furthermore, the focus will be restricted to environmental monitoring (EM) through satellites and surface sensor networks. EM is conducted for a variety of applications such as for assessing pollution (of land, water, and air), conditions of ecosystems (in forests, mountains, and coastal regions), and consequences of seismic events and natural disasters. RS is playing a critical role in addressing some of the most critical challenges we face today, including understanding the impacts of a changing climate, forecasting and planning for intense weather events, informing adaptation of our cities and agricultural practices, and ensuring protection of endangered ecosystems that enable our wellbeing.

9.2 RS from Sensor Networks Data acquisition for knowledge of specific physical parameters requires collection of information with detail and accuracy at regular time periods. Sensors are being widely used for EM. For instance, sensors are used for monitoring a crop and analyzing its state of health and growth and measuring parameters such as air humidity and temperature and soil dryness. Air quality sensors are used for the study of pollution in confined urban zones (requiring measurements of CO2 levels and other gases). Moreover, sensors are important in studies that must be carried out within enclosed spaces. Several applications in manufacturing facilities require such sensors in production processes and for inspection and quality control. Advances

9 Remote Sensing Through Satellites and Sensor Networks

251

in miniaturization and low-power electronics have led to emergence of wearable sensors such as those used in monitoring human health-related parameters (heart rate, temperature values, etc.). The progress in data communication technologies has converged with sensing technologies, such that sophisticated networks of sensors are now being used for a variety of applications related to environmental monitoring. The following sections provide a brief overview.

9.2.1 Data Collection Through Sensor Networks Sensors (sometimes called smart sensors due to advanced characteristics) have gained widespread use. The collected information, acquired from these sensors, is transferred through wired or wireless networks toward a data processing system in order to perform more challenging tasks and to transform the information in a higher level of knowledge and actions (Mukhopadhyay and Gupta 2008; Mack 2014). The ubiquitous distribution of sensors has resulted in the so-called Internet of Things (IoT), as shown in Fig. 9.1. The different symbols in the figure represent the various monitored things/devices. The Internet represents the backbone for information transmission, communication among the nodes, and for process management. Different sensors are embedded inside devices, measure the parameters of interest, and collect related data. There are several issues related to the limited power and processing capabilities of each node. Interested readers can refer to Chéour et al. (2018) for an example of how these issues are tackled in the case of wireless sensor network (WSN) applications.

Fig. 9.1 An example of IoT paradigm with several devices connected through a network. Each device is a node of the network, and the sensors are embedded inside the nodes

252

S. L. Ullo and A. Siddiqi

9.2.2 Environmental Monitoring Through Sensor Networks A variety of applications, including weather forecasting (Kulkarni and Kute 2016; Kamal 2017), air pollution control (GSMA 2018; Jovanovska and Davcev 2020; Arco et al. 2016), water quality control and monitoring (Elmustafa and Mujtaba 2019; Pavithra 2018; Pathak et al. 2019), and crop damage assessment (Pathak et al. 2019; Sivakannu and Balaji 2017), have been developed through the use of sensor networks and/or IoT. In Ullo and Sinha (2020), a critical review of noteworthy contributions and research studies on Smart Environment Monitoring (SEM) using IoT and sensor networks is presented. An example of how SEM systems can be used for measuring humidity, temperature, radiation, dust, UV signals, etc., is shown in Fig. 9.2. As shown in the figure, the backbone of the system is a WSN that represents the interface between IoT devices and data captured through the various types of smart sensors. Data collected through the WSN are then transmitted and stored on a cloud platform. This is a concept of a “smart city” (Jovanovska and Davcev 2020; Wong et al. 2018; Alharbi and Soh 2019), using a SEM system to enable clean and healthy environments for residents. Some additional examples of methods related to the use of sensor networks for EM are shown in Figs. 9.3, 9.4, and 9.5. Figure 9.3 shows how water contamination can be monitored and its control carried out by using a cloud-based system that connects IoT devices and different sensors. Through the IoT devices, the system can monitor if the water is contaminated or clean since all the devices have embedded AI-based sensors.

Fig. 9.2 SEM system addressing various issues in the environment monitoring using wireless sensor networks (WSNs) and IoT devices. Figure courtesy of Ullo and Sinha (2020)

9 Remote Sensing Through Satellites and Sensor Networks

253

Fig. 9.3 SEM system highlighting water contamination and its monitoring using cloud connecting IoT and sensors. Figure courtesy of Ullo and Sinha (2020)

Fig. 9.4 Smart agriculture monitoring system using IoT devices and sensors. Figure courtesy of Ullo and Sinha (2020)

SEM systems are also playing an important role in sustainable growth and enhanced productivity in agriculture. As shown in Fig. 9.4, enhancements in agricul-

254

S. L. Ullo and A. Siddiqi

Fig. 9.5 Smart sensors for agriculture applications. Figure courtesy of Ullo and Sinha (2021)

ture require monitoring of many factors, including soil condition, moisture analysis, water quality, and water quantity levels. These factors are shown in the scheme represented in Fig. 9.4. A smart agriculture monitoring system is depicted that is making use of suitable IoT devices, which are incorporating sensors able to capture relevant data, and is connected through a WSN. Figure 9.5 highlights the role of smart sensors in agriculture for increasing productivity. The smart sensors and IoT change the conventional agriculture practices into a smart farming paradigm. If it is made affordable and accessible, it can help improve incomes for farmers worldwide and also partly address the problem of overuse of fertilizers and pesticides that harm the natural environment.

9.3 Satellite Remote Sensing Since 1957 (Fig. 9.6), thousands of artificial (human-built) satellites have been launched into space. Some have been designed to take images of the Sun, the Earth, and other planets, and some are for observing deep space in search of black holes, distant stars, and galaxies. There are satellites for communications and meteorological surveys. The first artificial satellite in history, Sputnik 1, was launched in 1957 by the Soviet Union. This simple satellite, essentially a small aluminum ball with four

9 Remote Sensing Through Satellites and Sensor Networks

255

Fig. 9.6 The first artificial satellite, Sputnik 1, launched in 1957. Sputnik 1 is shown in an artist’s rendition here. Figure purchased by SciencePhotoLibrary (2023)

long antennas powered by batteries, marked the beginning of the space age. Modern satellites are far more complicated and use different techniques for RS that are categorized as passive and active. In passive RS, reflected signal from objects that are enlightened by external sources of electromagnetic energy is used for observation. In active RS, an electromagnetic energy source is used on the object, and the reflected signals are observed. Figure 9.7 provides an illustration, where in the case of passive satellite RS the external electromagnetic energy source is represented by the Sun, and in the case of active satellite RS, the same satellite is transmitting the electromagnetic signal and is then receiving the reflected one. Other configurations are possible. For instance, in case of passive RS, sensors onboard the satellites can record the sunlight in the visible and infrared wavelengths’ intervals, reflected by the Earth surface, but also its emitted energy in the thermal wavelengths’ interval. Then, depending on the type of sensor, the collected data will be used for different purposes. Similarly, regarding the active RS, the receiver can be on the same satellite but also on a different satellite, and in a back-scattering or forward-scattering configuration, if the receiver will collect the signal in the same direction of the transmitted one, or in the opposite direction. As shown in Fig. 9.7, data collected by the sensor onboard the satellite are down-streamed to a ground station. There is some onboard data processing within satellites, but most of the observed data are processed later after being received at a ground station. A large amount of EO data and tools have been made openly available, enabling new possibilities for research and enhancing the strategic role of space-based RS. This includes EO data from the Copernicus mission of the European Space Agency (ESA), available since 2014 with the launch of the first Sentinel-1A satellite

256

S. L. Ullo and A. Siddiqi

Fig. 9.7 Satellite data acquisition through passive and active RS systems

(Europe’s Copernicus programme 2023). The National Aeronautics and Space Administration (NASA), of the United States (US), has also made EO data available for free from its LandSat missions (Landsat Science 2023). Beyond making EO data openly available, thereby advancing new applications for its use after the data are received on the ground, there are several advances being made in onboard processing. ESA has promoted an onboard EO mission, the φ-sat1 experiment, which has demonstrated the potential of artificial intelligence (AI) as a reliable and accurate tool for cloud detection onboard the satellite (Giuffrida et al. 2021). Some other studies are being carried out in the same direction (Del Rosso et al. 2021a; Danielsen et al. 2021; Ziaja et al. 2021; Diana et al. 2021; Rapuano et al. 2021). AI onboard is shaping the designs of future spacecraft for RS.

9 Remote Sensing Through Satellites and Sensor Networks

257

As briefly discussed earlier, satellite RS offers many advantages: it provides a view of large regions of our planet with spatially referenced digital information; and it provides important data on land use and land cover (LULC), habitats, landscape, and built-infrastructure. Moreover, most of the remote sensors operate during all seasons, daily, and (mostly) continuously. Based on the type of sensor, the satellite data are used for different purposes. Several applications can be found for civil and environmental engineering, energy engineering, computer science and electronic engineering; land monitoring, observation, and assessment; and for the study of climate change; deforestation; pollution; atmospheric chemistry; coastal regions, seas, oceans; agriculture; disaster risk reduction; economic activity and development; and support for governments for national security.

9.3.1 Environmental Monitoring Through Satellite RS Satellites for RS are often characterized based on their onboard sensors. The main categories of sensors for EO and environmental monitoring include radars, optical imagers, and sounders. Some widely used spacecraft for a variety of EO applications include the Sentinel spacecraft. Sentinel-1A of ESA Copernicus mission, was launched in April 2014. It carries a single C-band synthetic aperture radar (SAR) instrument operating at a center frequency of 5.405 GHz (Sentinel Online 2023). The onboard sensor is an active sensor collecting data regardless of weather conditions and time of day (as it operates also at night time). It is worth highlighting that optical passive sensors are affected due to the presence of clouds over the area of interest. Also, optical sensors need an external source enlightening the area of interest (except for specific applications using thermal and near-infrared bands). Sentinel-1 includes a right-looking phased-array antenna providing fast scanning in elevation and azimuth, a data storage capacity of 1.410 Gb, and a 520 Mbit/s X-band downlink capacity. SENTINEL-1 operates in four exclusive acquisition modes: stripmap (SM) with a geometric resolution of 5 m by 5 m at a swath width of 80 km; interferometric wide swath (IW) with a geometric resolution of 5 m by 20 m at a swath width of 250 km; extra-wide swath (EW) at a geometric resolution of 20 m by 40 m at a swath width of 400 km; and wave mode (WV) at 5 m by 5 m geometric resolution for a vignette 20 km by 20 km. In practice, the SAR system designs include different operational modes that either optimize the spatial resolution (at the expense of the swath, and hence the coverage) or the swath width (at the expense of the resolution). The four acquisition modes are represented in Fig. 9.8. The Sentinel-1 C-SAR instrument supports operation in dual polarization (HH+HV, VV+VH) implemented through one transmit chain (switchable to H or V) and two parallel receive chains for H and V polarization, except for the wave mode, which is a single-polarization mode (selectable between HH and VV).

258

S. L. Ullo and A. Siddiqi

Flight Direction

Sub-Satellite Track

Orbit Height ~700 km 20° m 0k 10 m 0k 10

20°

0k 40 m

5k 37

25°

m

23°

45°

0k

25

36.5° m

Strip Map Mode

Extra Wide Swath Mode

Wave Mode Interferometric Wide Swath Mode

Fig. 9.8 Sentinel-1 acquisition modes. Source ESA Sentinel Online (2023)

To close the list of the Sentinel-1 main characteristics, it is necessary to add that Sentinel-1 is in a near-polar, Sun-synchronous orbit with a 12-day repeat cycle, and if both twin satellites are considered, the revisit time decreases to 6 days (ESA Sentinel Online 2023). It is important to highlight that a SAR is a specific type of radar employed to generate a 2-dimensional (2D) image or a 3-dimensional (3D) model of the terrain swath width [i.e., the digital elevation model (DEM)]. A SAR exploits the radar antenna motion to obtain a better spatial resolution than classical radars (i.e., conventional stationary beam-scanning radars) (NASA 2023). Besides Sentinel-1, many other satellite SAR systems have made their data available for monitoring our planet, and their main list is shown in Fig. 9.9. The list demonstrates the many different characteristics of each satellite and shows that choice of satellite data depends on the specific purpose for which it needs to be used. The details about Sentinel-1 were presented as one example here. There are many other kinds of sensors, and capabilities that currently deployed satellites are providing for RS. Interested readers can refer to Chuvieco (2016), Campbell and Wynne (2011), Born and Wolf (2013), Shankar (2020), Ulaby et al. (1981), Woodhouse (2005) and Cumming and Wong (2005) for more details.

9 Remote Sensing Through Satellites and Sensor Networks

Satellite/Sensor

Band

ERS-1/AMI JERS-1/SAR ERS-2/AMI RADARSAT-1

C L C C

ENVISAT/ASAR

C

ALOS/PALSAR

L

TerraSAR-X

X

RADARSAT-2

C

ALOS-2/PALSAR-2

L

Cosmo-Skymed

X

Sentinel-1

C

GaoFen-3 (GF-3)

C

Spatial resolution (m) 30 18×24 30 10–30 50–100 10–30 150–1000 10–20 100 1–2 3 16 1 3 50–100 1×3 3–10 60–100 1 5–15 30–100 5 5×20 20×40 5 1 3 5 8 10 10 25 25 25 25 50 100 500

Imaging mode StripMap StripMap StripMap StripMap ScanSAR StripMap ScanSAR StripMap ScanSAR SpotLight StripMap ScanSAR SpotLight StripMap ScanSAR SpotLight StripMap ScanSAR StripMap StripMap ScanSAR Stripmap Interferometric wide swath Extra wide swath Wave mode Spot light Ultra-fine strip Fine Strip I Full polarized strip I Fine strip II Wave imaging Standard strip Full polarized strip II Extended low Extended high Narrow scan Wide scan Global

259 Revisiting interval (d) 35 44 35 24 35 46 11

24

14

16

12

29

Polarization mode VV HH VV HH HH, VV, HH+VV, HH+HV, VV+VH HH,VV, HH+HV, VV+VH, HH+HV+VH+VV HH, VV, HH+VV, HH+HV, VV+VH HH,VV, HV, VH HH+HV, VV+VH HH+HV+VH+VV HH, VV, HV,VH HH+HV, VV+VH HH+HV+VH+VV HH, VV HH+VV, HH+HV VV+VH, HH,HV, VH, VV HH+HV,VH+VV,HH,VV

Optional single polarization Optional single polarization Optional dual polarization Full polarization Optional dual polarization Optional dual polarization Optional dual polarization Full polarization Optional dual polarization Optional dual polarization Optional dual polarization Optional dual polarization Optional dual polarization

Fig. 9.9 The main space-borne SAR systems. Figure courtesy of Liu et al. (2019)

9.3.1.1

Selected Applications in EM

Here, some examples of applications in EM using SAR data and use of SAR interferometry (InSAR) and Differential InSAR (DInSAR) techniques are provided. There are extensive literature (including studies presented in Di Martire et al. 2016, Ullo et al. 2018a, De Corso 2020, Ullo et al. 2019a and Ullo et al. 2018b) that show how InSAR and DInSAR can help in measuring ground displacements. These displacements can be due to landslides or earthquakes and are important for strategic infrastructure and structural monitoring. Optical satellite data are useful for many other purposes, including fire detection and monitoring, vegetation monitoring, and urban built-infrastructure classification (Moraguez et al. 2020; Yang et al. 2019; Barbosa et al. 2020; Cicala et al. 2018; Luca et al. 2018; Ghazaryan et al. 2018). Additionally, combination with other data from Light Detection And Ranging (LiDAR) devices, sensor networks, or SAR satellite

260

S. L. Ullo and A. Siddiqi

Fig. 9.10 Landslide susceptibility maps considering also burnt areas. Figure courtesy of Di Napoli et al. (2020)

data (Cicala et al. 2018; Luca et al. 2018; Ullo et al. 2020; Weng 2009; Focareta et al. 2015; Addabbo et al. 2016, 2015) provides enhanced capability for EM and new applications. To give an example, the work presented in Di Napoli et al. (2020) shows how satellite data can be used to carry out post-fire assessment of burnt areas in the Camaldoli and Agnano hill (Naples, Italy), aiming to expand the knowledge of the relationship existing between the triggering of landslides and burnt areas. In the study area, the fires seem to act as a predisposing factor for landslides, while the triggering factor is usually represented by precipitation. Figure 9.10 shows landslide susceptibility maps considering burnt areas. In particular, (a,b) show a susceptibility map considering fires that occurred in 1995–1996 and the rainfalltriggered landslides of 1997 (Camaldoli hill); (c,d) susceptibility maps considering

9 Remote Sensing Through Satellites and Sensor Networks

261

Fig. 9.11 The WorldView-2 image and results of the proposed buildings extraction algorithm over Lioni town. Figure courtesy of Ullo et al. (2020)

the fires that occurred in 1999–2000 and the rainfall-triggered landslides of 2001 (Camaldoli hill); finally, (e,f) susceptibility map considering the fires that occurred in 2017–2018 and the rainfall-triggered landslides of 2019 (Agnano hills). Another important application of satellite data, namely Very High-Resolution (VHR) data, is given in Ullo et al. (2020), where World-View2 optical data are used for building detection and mapping, in combination with LiDAR data and data from other sources. Two case studies are presented and discussed, based on the use of 2-D and 3-D LiDAR data, showing how the data combinations succeed in improving the analysis and monitoring of specific areas of interest, by helping in exploring external environment and extracting building features from urban areas. Figure 9.11 shows final results of the urban classification when several pre-processing algorithms have been employed together with an object-based image analysis (OBIA).

9.3.1.2

New Applications of RS in Sustainable Development

The use of satellite data for RS and EM has expanded rapidly—and will likely continue to accelerate as technologies become cheaper. EO will continue to advance climate change studies, oceans and land analysis, and geo-hazards assessment. And applications in urbanization and transportation will continue to grow in importance. These include applications in updating road maps, monitoring asphalt

262

S. L. Ullo and A. Siddiqi

conditions, wetland delineation, urban planning; impervious surface mapping, fresh water quality monitoring, telecommunication planning; coastal mapping; disaster mapping and monitoring; and damage assessment. The increasing availability of satellite data has attracted attention beyond operational applications (such as for fire and floods monitoring), to longer-term questions related to supporting regional and national economic development (Magliarditi et al. 2019a; Siddiqi et al. 2019a). There is growing use of satellite data and artificial intelligence-supported tools for monitoring progress toward the sustainable development goals (SDGs) that have been adopted by member countries of the United Nations (UN-ARIES 2021). New efforts for quantifying and valuing ecosystem services, and assessments of natural capital stocks (such as forests, fresh water bodies, minerals, and fisheries), are relying on remote sensing data for computing the value of these resources and including them in national accounts (UN-SEEA 2021). As satellite data are becoming increasingly accessible and are collected at shorter time intervals, it is now possible not only to better understand how the Earth is changing, but also to utilize these insights to improve decision-making (Foreman et al. 2018). The emerging expansion of EO activities, together with the growth of “New Space” entrepreneurship worldwide, the so-called New Space services opportunities, is becoming available through low-cost satellite constellations involving micro-, mini-, and small satellites (Reid et al. 2019; Supporting the Sustainable Development Goals 2023; Keola et al. 2015).

9.4 Data Combination and Techniques Synergy Fusion and combination of data acquired from different sources is of increasing interest and often provides better results as compared to using single sensors. Moreover, techniques for data processing or pre-processing are increasingly sophisticated. Not only classical algorithms, but especially those based on advanced operations relying on machine learning (ML) and deep learning (DL) paradigms, are being used in an extensive way, and many examples can be found in the literature Di Napoli et al. (2020), Ullo et al. (2019b), Del Rosso et al. (2021b), Ullo et al. (2021) and Rajendran et al. (2020). Among the ML-based techniques, the technologies involving quantum machine learning (QML) applied to RS are at the frontier. Some research efforts have started making use of such advanced techniques as QML for RS. For instance, quantum computers and convolutional neural networks (CNNs) are considered together for accelerating geospatial data processing (Henderson et al. 2020). Quantum circuitbased neural network classifiers for multispectral land cover classification have been introduced in preliminary proof-of-concept applications (Gawron and Lewinski 2020), and an ensemble of support vector machines running on the D-Wave quantum annealer has been proposed for remote sensing image classification (Cavallaro et al. 2020). In Zaidenberg et al. (2021) and Zaidenberg et al. (2021a), hybrid

9 Remote Sensing Through Satellites and Sensor Networks

263

quantum-classical neural networks for remote sensing applications are discussed, and a proof-of-concept for binary classification, using multispectral optical data, is reported. Lastly, in more recent work Sebastianelli et al. (2022), circuit-based hybrid quantum convolutional neural networks (QCNNs) have been successfully employed as image classifiers in the context of RS. One of the promising trends, in data combination, is those where data obtained at macro-scale and at micro-scales are combined. One such effort in this area is presented in Sebastianelli et al. (2021), showing how a multi-user platform based on AI algorithms and processing heterogeneous data can compute risk levels. The model includes a specific neural network that is first trained to learn the correlations between selected inputs, related to the case of interest: environmental variables (chemical–physical, such as meteorological), human activity (such as traffic and crowding), level of pollution (in particular the concentration of particulate matter), and epidemiological variables related to the evolution of the contagion. The objective is the creation of a new tool to support organizations (such as public agencies) in the implementation of targeted countermeasures, based on quantitative and multi-scale elements, for prevention of public emergencies, such as the current COVID-19 pandemic. In this project, advanced techniques work to process heterogeneous data from multiple sources (satellite, sensor networks, databases, etc.), and an example of a practical application is given. Figures 9.12 and 9.13 show, respectively, the different data sources employed in the analysis and the block diagram of the proposed solution. Other examples include collaboration between RS scientists working on satellite data processing and geo-technicians (Ullo et al. 2019a), where in situ measurements and displacements retrieved through DInSAR technique from SAR Sentinel-1 data have been analyzed and compared for dam monitoring. This chapter illustrates one of the first cases of joint use of satellite data and locally acquired structure data, which will hopefully lead to other such applications in the future.

Fig. 9.12 Different data sources employed in the model for achieving COVID-19 countermeasures. Figure courtesy of Sebastianelli et al. (2021)

264

S. L. Ullo and A. Siddiqi

Fig. 9.13 The block diagram of the solution proposed in the project for COVID-19 countermeasures. Figure courtesy of Sebastianelli et al. (2021)

9.5 Emerging Trends: RS with Small Spacecraft Constellations A recent, and expanding trend in RS for EO, is the emergence of small spacecraft constellations that provide high coverage, resolution, and high frequency of data collection as compared to a single spacecraft. While large satellites historically provided communication and imaging services, there are trends of smaller spacecraft operating in constellations to provide such services. Figure 9.14 shows how the trends in spacecraft mass at launch have changed over the decades since 1990. Each data point corresponds to a spacecraft launched in a particular year for an Earth observation mission. It can be observed that while large spacecraft continue to be launched and used, there are a growing number of small spacecraft being used for EO, and since 2010, there are an increasing number of spacecraft at 100 kilograms and less. New observing strategies, which combine the use of coordinated data collection from space, air, and ground, are also being developed through use of new design tools (LeMoigne et al. 2017; Maciuca et al. 2009), valuation methods (Siddiqi et al. 2020a, 2019b), and technology development. In Marcuccio et al. (2019), smaller satellites and larger constellations are presented, by highlighting the major players that are emerging in the private satellite sector. Moreover, another interesting analysis tracking the transitioning from large to small satellites is presented in Paek et al. (2020), in the case of space-based SAR platforms, that due to new miniaturization technologies (though slower than the optical counterpart) have reached sizes on the order of 100 kg smaller if compared to the traditional ones. Table 9.1 shows a sample selection of commercial SAR smallsats working in the X-band with main characteristics. For small satellites, there has recently also been the launch of a 185 kg C-band satellite (Werner 2021). In addition to the interferometric capabilities that SAR satellites provide, smallsat SAR companies are aiming to launch constellations of such spacecraft to achieve higher

9 Remote Sensing Through Satellites and Sensor Networks

265

Fig. 9.14 Launch mass of currently operational spacecraft that have been launched since 1990 for Earth observation missions. Data sourced from SpaceTrak database (Seradata Database 2023) (Data from Seradata are used under an MIT license, and this figure is created by the authors) Table 9.1 Example commercial SAR smallsats working in the X-band

Provider Current size Mass [kg] Nr. satellites Revisit time Band Polarimetry Country

ICEYE (Iceye 2021) 10 85 18 Several hours X VV Finland

Capella (Capella Space 2021) 3 107 36 Hour X HH United States

Synspective (Synspective 2021) 1 150 30 2 hours X VV Japan

iQPS (iQPS 2021) 2 100 36 2 hours X

Umbra (Umbra 2021) 1 65 12 Several hours X

Japan

United States

temporal frequency and geographic coverage. Figure 9.15 compares the anticipated resolutions and maximum revisit times for these constellations. Form Table 9.1 and Fig. 9.15, it is clear how small satellites can achieve revisit time and spatial resolutions that are valuable for many new applications. It is also important to highlight the drawbacks of some of these new trends. An interesting study highlights this Filippazzo (2017), where specific indexes are introduced to critically analyze how the new systems perform with respect to existing and traditional designs. One challenge in use of small spacecraft, with optical payloads, is lack of onboard calibration systems. Such spacecraft rely on cross calibration with larger spacecraft, and with ground-based calibration targets. The quality of data collected can be variable, and its use for scientific applications for EM, and decision-making

266

S. L. Ullo and A. Siddiqi

Anticipated constellation revisit times and resolutions 1.2 iQPS Synspective

Range resolution [meters]

1.0

2.75 2.50

0.8

2.25 2.00

0.6 Iceye

Capella 1.75

0.4

1.50

Umbra 0.2

1.25

Maximum Revisit Time [Hours]

3.00

1.00

0.0 0.0

0.2

0.4

0.6

0.8

1.0

1.2

1.4

Azimuth resolution [meters] Fig. 9.15 Anticipated resolutions and maximum revisit times of smallsat SAR constellations. Figure courtesy of ESA Sentinel Online (2023)

can be limited (Siddiqi et al. 2021). Some recent studies have shown how calibration errors for optical payloads can be magnified in data products if radiometric data are used to construct indices that require differencing and thresholding (Siddiqi et al. 2020b; Baber et al. 2020). For instance, the widely used normalized difference vegetation index (NDVI) is one such example. New data processing techniques that correct for sensor drift and calibration issues, as well as development of small onboard calibration equipment, will help address these issues in the future. Another important issue, due to increase of small spacecraft constellations, is proliferation of orbital debris (Foreman et al. 2017). The “Index of Objects Launched into Outer Space” (Outer Space Objects Index 2023), maintained by the United Nations Office for Outer Space Affairs (2023), shows that there were 7389 individual satellites in Space at the end of April 2021. Of these, 3170 satellites are inactive, as reported on 1st January, 2021. The inactive systems contribute to the creation of “space junk” or orbital debris and are becoming a critical issue affecting viability and safety of active spacecraft operations. This problem has gained attention for technology development (where new spacecraft are being designed for cleaning orbital debris), as well as for policy (where new requirements are being proposed for safely de-orbiting and disposing spacecraft after the end of their operational life).

9 Remote Sensing Through Satellites and Sensor Networks

267

9.6 Summary This chapter presented an overview of the use of RS data from sensors onboard of satellites and distributed among cabled and wireless networks for Earth observation and monitoring. The new technologies in data acquisition and fusion, wherein information is combined from multiple sources, are opening up entirely new possibilities and applications that can improve safety, public health, and human well-being. However, with the advances in ubiquitous data collection and use, there are challenges that will need to be thoughtfully considered and carefully addressed. Some of these include issues of ensuring data quality and ensuring a sustainable space environment (through reducing orbital debris). In the interest of brevity, other (important) issues such as of ensuring data privacy, data security, and equity were not discussed here, but these issues also need to be recognized and duly addressed. RS will form a critical part of a future where large amounts of information are collected and used for building knowledge that will be utilized for informing and taking decisions. Engineers and researchers, while advancing these capabilities, must also ensure that the technologies contribute to a sustainable and peaceful future for people and for the planet. Acknowledgments The authors wish to thank Sheila Baber for assistance in compiling data for SAR constellations. Sheila participated in a joint program of MIT and University of Sannio through the MIT Science and Technology Initiative (MISTI) during her Independent Activity Period (IAP) in January 2021 jointly supervised by Afreen Siddiqi and Silvia L. Ullo.

References Addabbo P, di Bisceglie M, Focareta M, Galdi C, Maffei C, Ullo SL (2015) Combination of LANDSAT and EROS-B satellite images with GPS and LiDAR data for land monitoring. A case study: the Sant’Arcangelo Trimonte dump. In: 2015 IEEE international geoscience and remote sensing symposium (IGARSS), pp 882–885 Addabbo P, Focareta M, Marcuccio S, Votto C, Ullo S (2016) Land cover classification and monitoring through multisensor image and data combination. In 2016 IEEE international geoscience and remote sensing symposium (IGARSS) Alharbi N, Soh B (2019) Roles and challenges of network sensors in smart cities. In: IOP conference series: earth and environmental science, vol 322, no 1. IOP Publishing, Bristol, p 012002 Arco E, Boccardo P, Gandino F, Lingua A, Noardo F, Rebaudengo M (2016) An integrated approach for pollution monitoring: smart acquirement and smart information. In: ISPRS annals of photogrammetry, remote sensing & spatial information sciences, vol 3, no 1 Baber S, Siddiqi A, de Weck OL (2020) A quantitative assessment of radiometric calibration errors on crop cover classifications. In: AGU fall meeting 2020. AGU Barbosa M, Siddiqi A, de Weck O (2020) Error scaling with confusion matrices for global optical remote sensing of building and road detection. In: AGU fall meeting abstracts, vol 2020, pp GC058–0004 Born M, Wolf E (2013) Principles of optics: electromagnetic theory of propagation, interference and diffraction of light. Elsevier, Amsterdam

268

S. L. Ullo and A. Siddiqi

Campbell JB, Wynne RH (2011) Introduction to remote sensing. Guilford Press, New York City Capella Space (2021). Available Online: https://www.capellaspace.com/. Accessed 15 Feb 2021 Cavallaro G, Willsch D, Willsch M, Michielsen K, Riedel M (2020) Approaching remote sensing image classification with ensembles of support vector machines on the D-Wave quantum annealer. In: International geoscience and remote sensing symposium (IGARSS), pp 1973– 1976. https://doi.org/10.1109/IGARSS39084.2020.9323544 Chéour R, Jmal MW, Abid M (2018) New combined method for low energy consumption in wireless sensor network applications. Simulation 94(10):873–885 Chuvieco E (2016) Fundamentals of satellite remote sensing: an environmental approach. CRC Press, Boca Raton Cicala L, Angelino CV, Fiscante N, Ullo S (2018) Landsat-8 and Sentinel-2 for fire monitoring at a local scale: a case study on vesuvius. In: 2018 IEEE international conference on environmental engineering (EE), pp 1–6. [Online]. Available: https://ieeexplore.ieee.org/document/8385269/ Cumming I, Wong F (2005) Digital processing of SAR data. Artech House, Norwood Danielsen AS, Johansen TA, Garrett JL (2021) Self-organizing maps for clustering hyperspectral images on-board a cubesat. Remote Sensing 13(20). [Online]. Available: https://www.mdpi. com/2072-4292/13/20/4174 De Corso T, Mignone L, Sebastianelli A, Rosso MPD, Yost C, Ciampa E, Pecce M, Sica S, Ullo S (2020) Application of DInSAR technique to high coherence satellite images for strategic infrastructure monitoring. In: IGARSS 2020-2020 IEEE international geoscience and remote sensing symposium Del Rosso MP, Sebastianelli A, Spiller D, Mathieu PP, Ullo SL (2021a) On-board volcanic eruption detection through cnns and satellite multispectral imagery. Remote Sensing 13(17). [Online]. Available: https://www.mdpi.com/2072-4292/13/17/3479 Del Rosso MP, Sebastianelli A, Ullo SL (eds) (2021b) Artificial intelligence applied to satellitebased remote sensing data for earth observation. Institution of Engineering and Technology. https://doi.org/10.1049/PBTE098E Di Martire D, Confuorto P, Frezza A, Ramondini M, Lòpez AV, Pia Del Rosso M, Sebastianelli A, Ullo SL (2018) X- and c-band sar data to monitoring ground deformations and slow-moving landslides for the 2016 manta and portoviejo earthquake (manabi, ecuador). In: 2018 IEEE international conference on environmental engineering (EE), pp 1–6 Diana L, Xu J, Fanucci L (2021) Oil spill identification from SAR images for low power embedded systems using CNN. Remote Sensing 13(18). [Online]. Available: https://www.mdpi.com/ 2072-4292/13/18/3606 Di Napoli M, Marsiglia P, Di Martire D, Ramondini M, Ullo SL, Calcaterra D (2020) Landslide susceptibility assessment of wildfire burnt areas through earth-observation techniques and a machine learning-based approach. Remote Sensing 12(15). [Online]. Available: https://www. mdpi.com/2072-4292/12/15/2505 Elmustafa SAA, Mujtaba EY (2019) Internet of things in smart environment: concept, applications, challenges, and future directions. World Sci News 134(1):1–51 ESA Sentinel Online (2023). https://sentinels.copernicus.eu/web/sentinel/missions/sentinel-1/ satellite-description/orbit Europe’s Copernicus programme (2023). https://www.esa.int/Applications/Observing_the_Earth/ Copernicus/Europe_s_Copernicus_programme Filippazzo G (2017) The potential impact of small satellite radar constellations on traditional space systems. In: Proceedings of the 5th federated and fractionated satellite systems workshop, Ithaca, NY, USA Focareta M, Marcuccio S, Ullo S, Votto C Combination of landsat 8 and Sentinel 1 data for the characterization of a site of interest. A case study: the royal palace of caserta. In: 1st international conference on metrology for archaeology Foreman VL, Siddiqi A, De Weck O (2017) Large satellite constellation orbital debris impacts: case studies of oneweb and spacex proposals. In: AIAA SPACE and astronautics forum and exposition, p 5200

9 Remote Sensing Through Satellites and Sensor Networks

269

Foreman V, Siddiqi A, Weck OD (2018) Advantages and limitations of small satellites in low earth orbit constellations: a prospective review. In: Small Satellite Conference, Utah. Available: https://digitalcommons.usu.edu/smallsat/2018/all2018/358/ Gawron P, Lewinski S (2020) Multi-spectral image classification with quantum neural networks. In: Proceedings of IGARSS Ghazaryan G, Dubovyk O, Graw V, Schellberg J (2018) Vegetation monitoring with satellite time series: an integrated approach for user-oriented knowledge extraction. In: Neale CMU, Maltese A (eds) Remote sensing for agriculture, ecosystems, and hydrology XX, vol 10783, International society for optics and photonics. SPIE, Cergy-Pontoise, pp 201–209. [Online]. Available: https://doi.org/10.1117/12.2325762 Giuffrida G, Fanucci L, Meoni G, Batiˇc M, Buckley L, Dunne A, Van Dijk C, Esposito M, Hefele J, Vercruyssen N, Furano G, Pastena M, Aschbacher J (2021) The -sat-1 mission: the first on-board deep neural network demonstrator for satellite earth observation. IEEE Trans Geosci Remote Sens 60:1–1 GSMA (2018) Air quality monitoring using IoT and big data. Available online: https://www.gsma. com/iot/wp-content/uploads/2018/02/iot_clean_air_02_18.pdf Henderson M, Gallina J, Brett M (2020) Methods for accelerating geospatial data processing using quantum computers. arXiv:2004.03079 Iceye (2021). Available Online: https://www.iceye.com/ Accessed on 14 Feb 2021 iQPS, Inc. (2021) Available Online: https://i-qps.net/tech/. Accessed 19 Feb 2021 Jovanovska EM, Davcev D (2020) No pollution smart city sightseeing based on WSN monitoring system. In: 2020 sixth international conference on mobile and secure services (MobiSecServ). IEEE, Piscataway, pp 1–6 Kamal R (2017) Lesson 11 internet connected environment (weather, air pollution and forest fire) monitoring, pp 1–41. Available online: https://www.dauniv.ac.in/public/frontassets/ coursematerial/InternetofThings/IoTCh12L11EnvironmentMonitoring.pdf Keola S, Andersson M, Hall O (2015) Monitoring economic development from space: using nighttime light and land cover data to measure economic growth. World Dev. 66:322–334. [Online]. Available: https://www.sciencedirect.com/science/article/pii/S0305750X14002551 Kulkarni P, Kute P (2016) Internet of things based system for remote monitoring of weather parameters and applications. Int J Adv Electron Comput Sci 3(2):68–73 Landsat Science (2023). https://landsat.gsfc.nasa.gov/data/where-to-get-data/ LeMoigne J, Dabney P, Foreman V, Grogan P, Hache S, Holland MP, Hughes SP, Nag S, Siddiqi A (2017) End-to-end trade-space analysis for designing constellation missions. In: AGU fall meeting abstracts, vol 2017 , pp IN13B–0071 Liu CA, Chen ZX, Shao Y, Chen JS, Hasi T, Pan HZ (2019) Research advances of SAR remote sensing for agriculture applications: a review. J Integr Agricult 18(3):506–525. [Online]. Available: https://www.sciencedirect.com/science/article/pii/S2095311918620167 Luca C, Sara P, Cesario Vincenzo A, Nicomino F, Ullo S, Pia A (2018) Unsupervised postfire assessment of burned areas with free and open multispectral data using OBIA. In: GEOBIA 2018 - From pixels to ecosystems and global sustainability. [Online]. Available: www.geobia2018.com, https://hal.archives-ouvertes.fr/hal-01957184 Mack CA (2014) The art of wireless sensor networks, in series: signals and communication technology, vol 1. Springer, Berlin, pp XVII, 830 Maciuca, D., Chow, J., Siddiqi, A., de Weck, O., Alban, S., Dewell, L., Howell, A., Lieb, J., Mottinger, B., Pandya, J. and Ramirez, J., 2009, September. A modular, high-fidelity tool to model the utility of fractionated space systems. In AIAA SPACE 2009 Conference & Exposition (p. 6765). Magliarditi E, Siddiqi A, de Weck O (2019a) Remote sensing for assessing natural capital in inclusive wealth of nations: current capabilities and gaps. In: IGARSS 2019–2019 IEEE international geoscience and remote sensing symposium. IEEE, Yokohama, Japan, pp 4411– 4414

270

S. L. Ullo and A. Siddiqi

Marcuccio S, Ullo S, Carminati M, Kanoun O (2019) Smaller satellites, larger constellations: trends and design issues for earth observation systems. IEEE Aerospace Electron Syst Mag 34(10):50–59 Moraguez M, Trujillo A, de Weck O, Siddiqi A (2020) Convolutional neural network for detection of residential photovoltalc systems in satellite imagery. In: IGARSS 2020–2020 IEEE international geoscience and remote sensing symposium. IEEE, pp 1600–1603. Available: https://ieeexplore.ieee.org/document/9324245 Mukhopadhyay S, Gupta G (2008) Smart sensors and sensing technology, in series: lecture notes in electrical engineering, vol 20. Springer, Berlin, pp 27–38 NASA, EarthData Open Access for Open Data (2023). https://earthdata.nasa.gov/learn/ backgrounders/what-is-sar Outer Space Objects Index (2023) https://www.unoosa.org/oosa/osoindex/index.jspx?lf_id= Paek SW, Balasubramanian S, Kim S, de Weck O (2020) Small-satellite synthetic aperture radar for continuous global biospheric monitoring: a review. Remote Sensing 12(16):2546. [Online]. Available: https://www.mdpi.com/2072-4292/12/16/2546 Pathak A, AmazUddin M, Abedin MJ, Andersson K, Mustafa R, Hossain MS (2019) IoT based smart system to support agricultural parameters: a case study. Procedia Comput Sci 155:648– 653 Pavithra G (2018) Intelligent monitoring device for agricultural greenhouse using IOT. J Agricult Sci Food Res 9(2):2–5 Rajendran GB, Kumarasamy UM, Zarro C, Divakarachari PB, Ullo SL (2020) Land-use and land-cover classification using a human group-based particle swarm optimization algorithm with an LSTM classifier on hybrid pre-processing remote-sensing images. Remote Sensing 12(24):4135. [Online]. Available: https://www.mdpi.com/2072-4292/12/24/4135 Rapuano E, Meoni G, Pacini T, Dinelli G, Furano G, Giuffrida G, Fanucci L (2021) An FPGAbased hardware accelerator for CNNs inference on board satellites: benchmarking with myriad 2-based solution for the cloudscout case study. Remote Sensing 13(8). [Online]. Available: https://www.mdpi.com/2072-4292/13/8/1518 Reid J, Zeng C, Wood D (2019) Combining social, environmental and design models to support the sustainable development goals. In 2019 IEEE aerospace conference, pp 1–13 SciencePhotoLibrary (2023) An image from space of Sputnik 1. https://www.sciencephoto.com/ media/1157162/view/sputnik-1-in-earth-orbit-illustration Sebastianelli A, Mauro F, Di Cosmo G, Passarini F, Carminati M, Ullo SL (2021) Airsense-to-act: a concept paper for covid-19 countermeasures based on artificial intelligence algorithms and multi-source data processing. ISPRS Int J Geo-Inf 10(1):34. [Online]. Available: https://www. mdpi.com/2220-9964/10/1/34 Sebastianelli A, Zaidenberg DA, Spiller D, Saux BL, Ullo SL (2022) On circuit-based hybrid quantum neural networks for remote sensing imagery classification. IEEE J Selec Topics Appl Earth Observ Remote Sensing 15:565–580 Sentinel Online (2023) Sentinel-1. https://sentinel.esa.int/web/sentinel/missions/sentinel-1 Seradata Database (2023) https://www.seradata.com/products/spacetrak/ Shankar R (2020) Fundamentals of physics II: electromagnetism, optics, and quantum mechanics. Yale University Press, New Haven Siddiqi A, Magliarditi E, DeWeck O (2019a) Small spacecraft earth observing missions for natural capital assessment. Available: https://iafastro.directory/iac/archive/browse/IAC-19/B4/ 1/54877/. In: International astronautical federation-70th international astronautical congress (2019) Siddiqi A, Magliarditi E, de Week O (2019b). Valuing new earth observation missions for system architecture trade-studies. Available: https://ieeexplore.ieee.org/document/8899126. In: IGARSS 2019–2019 IEEE international geoscience and remote sensing symposium. IEEE, Yokohama, Japan, pp 5297–5300 Siddiqi A, Baber S, de Weck OL, Durell C, Russell B, Holt J (2020a) Integrating globally dispersed calibration in small satellites mission value. Small Satellites Conference 2020, Utah. Available: https://digitalcommons.usu.edu/smallsat/2020/all2020/25/

9 Remote Sensing Through Satellites and Sensor Networks

271

Siddiqi A, Baber S, de Weck O, Durell C (2020b) Error and uncertainty in earth observation value chains. Available: https://ieeexplore.ieee.org/document/9323463. In: IGARSS 2020– 2020 IEEE international geoscience and remote sensing symposium. IEEE, Yokohama, Japan, pp 3158–3161 Siddiqi A, Baber S, De Weck O (2021) Valuing radiometric quality of remote sensing data for decisions. Available: https://ieeexplore.ieee.org/document/9553916. In: 2021 IEEE international geoscience and remote sensing symposium IGARSS. IEEE, pp 5724–5727 Sivakannu G, Balaji S (2017) Implementation of smart farm monitoring using IoT. Int J Curr Eng Sci Res 4(6):21–27 Supporting the Sustainable Development Goals (2023) https://www.unoosa.org/res/oosadoc/data/ documents/2018/stspace/stspace67_0_html/SDGs_EGNSSCopernicus_eBook.pdf Synspective (2021). Available Online: https://synspective.com/. Accessed 16 Feb 2021 Ulaby T, Moore K, Fung K (1981) Microwave remote sensing. Volume I: microwave remote sensing fundamentals and radiometry. Artech House, Norwood Ullo SL, Sinha G (2020) Advances in smart environment monitoring systems using IoT and sensors. Sensors 20(11):3113 Ullo SL, Sinha G (2021a) Advances in IoT and smart sensors for remote sensing and agriculture applications. Remote Sensing 13(13):2585 Ullo SL, Angelino CV, Cicala L, Fiscante N, Addabbo P, Del Rosso MP, Sebastianelli A (2018a) Sar interferometry with open Sentinel-1 data for environmental measurements: the case of ischia earthquake. In: 2018 IEEE international conference on environmental engineering (EE), pp 1–8 Ullo S, Angelino CV, Cicala L, Fiscante N, Addabbo P (2018b) Use of differential interferometry on Sentinel-1 images fot the measurement of ground displacements. ischia earthquake and comparison with Ingv data. In: IGARSS 2018 - 2018 IEEE international geoscience and remote sensing symposium, pp 2216–2219. [Online]. Available: https://ieeexplore.ieee.org/document/ 8518715/ Ullo S, Addabbo P, Di Martire D, Sica S, Fiscante N, Cicala L, Angelino VC (2019a) Application of dinsar technique to high coherence Sentinel-1 images for dam monitoring and result validation through in situ measurements. IEEE J Select Topics Appl Earth Observat Remote Sensing 12:875–890 Ullo SL, Langenkamp MS, Oikarinen TP, DelRosso MP, Sebastianelli A, Piccirillo F, Sica S (2019b) Landslide geohazard assessment with convolutional neural networks using sentinel2 imagery data. In IGARSS 2019–2019 IEEE international geoscience and remote sensing symposium, pp 9646–9649 Ullo S, Zarro C, Wojtowicz K, Meoli G, Focareta M (2020) LiDAR-Based system and optical VHR data for building detection and mapping. Sensors 20:1285 Ullo SL, Mohan A, Sebastianelli A, Ahamed SE, Kumar B, Dwivedi R, Sinha GR (2021b) A new mask R-CNN-based method for improved landslide detection. IEEE J Selec Top Appl Earth Observ Remote Sensing 14:3799–3810 Umbra (2021). Available Online: https://umbra.space. Accessed 1 March 2021 UN-ARIES (2021) United Nations SEEA: artificial intelligence for ecosystem accounting. https:// seea.un.org/content/aries-for-seea UN-SEEA (2021) United Nations: system of environmental economic accounting. https://seea.un. org United Nations Office for Outer Space Affairs (2023). https://www.unoosa.org/oosa/index.html Weng Q (2009) Remote sensing and GIS integration: theories, methods, and applications: theory, methods, and applications. McGraw-Hill Education, New York Werner D (2021) SpaceNews. https://spacenews.com/spacety-releases-first-sar-images/ Wong MS, Wang T, Ho HC, Kwok CY, Lu K, Abbas S (2018) Towards a smart city: development and application of an improved integrated environmental monitoring system. Sustainability 10(3):623 Woodhouse IH (2005) Introduction to microwave remote sensing. CRC Press, Boca Raton

272

S. L. Ullo and A. Siddiqi

Yang L, Siddiqi A, de Weck OL (2019) Urban roads network detection from high resolution remote sensing. In: IGARSS 2019–2019 IEEE international geoscience and remote sensing symposium. Available: https://ieeexplore.ieee.org/document/8899328. IEEE, Yokohama, Japan, pp 7431–7434 Zaidenberg DA, Sebastianelli A, Spiller D, Saux BL, Ullo SL (2021) Advantages and bottlenecks of quantum machine learning for remote sensing. arXiv:2101.10657 Zaidenberg DA, Sebastianelli A, Spiller D, Le Saux B, Ullo SL (2021) Advantages and bottlenecks of quantum machine learning for remote sensing. In: IEEE international geoscience and remote sensing symposium (IGARSS), 07 (2021) Ziaja M, Bosowski P, Myller M, Gajoch G, Gumiela M, Protich J, Borda K, Jayaraman D, Dividino R, Nalepa J (2021) Benchmarking deep learning for on-board space applications. Remote Sensing 13(19). [Online]. Available: https://www.mdpi.com/2072-4292/13/19/3981

Silvia Liberata Ullo IEEE Senior Member, Industry Liaison for IEEE Joint ComSoc/VTS Italy Chapter. National Referent for FIDAPA BPW Italy Science and Technology Task Force from 2019 to 2021. Researcher since 2004 at the University of Sannio, Benevento (Italy), she has authored 80+ research papers, co-authored many book chapters and two books. Full list at: https://iris.unisannio.it/simplesearch?query=Silvia%20Liberata%20Ullo She has served as editor of many special issues and is Associate Editor of IEEE JSTARS IET Image Processing, IET Wireless Sensor Networks, Section Editor (Section: Machine learning/radar area) of the journal Recent Advances in Computer Science and Communications (RACSC), Topical Editor (Topic: Geo-Informatics and Remote Sensing) of the Arabian Journal of Geosciences Member of Academic Senate and Ph.D. Professors’ Board. Teaching the courses: Principles of Mathematics, Signal theory and elaboration, Telecommunication networks and Optical and radar remote sensing, this latter for Ph.D. students. She graduated cum laude in Electronic Engineering in 1989 at Federico II University, Naples. Awarded with an ITALTEL scholarship, accomplished a Microeconomics Program at Harvard University, in August 1990, and the Master of Science in Management at Massachusetts Institute of Technology (MIT), in 1992 (Cambridge, Massachusetts-USA). She worked with ITALTEL S.p.A. for 8 years as Research Engineer and Production Manager, and for 4 years with the Benevento Municipality, as Officer at the Data Elaboration Center (CED). Member of the Board of Directors for the

9 Remote Sensing Through Satellites and Sensor Networks

273

Municipal Transport Society (AMTU) of Benevento for 3 years (1994–1997). Since 2004 Silvia Ullo is a researcher with the University of Sannio, Benevento (Italy), with interests in Data analysis through satellite remote sensing (RS) for Earth observation, in communication networks with a particular reference to sensor networks and smart grids, in radar systems and radar detection in non-Gaussian environment, in non-Gaussian models for the backscatter signal from natural surfaces. In the last years she is focusing on Machine Learning (ML) and Quantum Machine Learning (QML) applied to RS She has been the promoter of the agreement, signed in November 2017, between the University of Sannio and the MIT MISTI-Italy to involve MIT students in research projects at the University of Sannio, and she is the Coordinator of the Program. Moreover, she has promoted the (Memorandum Of Understanding) MOUs with many Indian universities, like the East Point College of Engineering and Technology in Bangalore, the GSSS Institute of Engineering Technology for Women in Mysuru, and many others. Silvia Ullo has been co-organizer of many national and international conferences as the MetroAeroSpace Workshop, since 2014, and the first Workshop on Networks and Cyber Security, in Benevento, 2017. She has also organized several special sessions in international conferences as, for instance, "Wireless Sensor Networks and Remote Sensing for Environmental Applications" within the first IEEE International Environmental Engineering Conference held in Milan, March 2018. She has been a Committee Member for the organization of the MIT Sloan Global Women’s Conference, held in New York, in 2017 and in 2019. Some notes: she has received university and professional society recognitions. In 1990 she got the "Golden Apple" from the "Marisa Bellisario" Foundation, that every year recognizes this award to women who are excelling in typically male jobs. Moreover, she has devoted may years to spread her experience and support young people to pursue scientific studies, above all young girls, by promoting several initiatives in which women can tell themselves as role models.

274

S. L. Ullo and A. Siddiqi

Afreen Siddiqi is a research scientist in the Department of Aeronautics and Astronautics at the Massachusetts Institute of Technology (MIT) and Adjunct Lecturer of Public Policy at Harvard Kennedy School in the USA. She has an SB in Mechanical Engineering, an SM in Aeronautics and Astronautics, and a PhD in Aerospace Systems, all from MIT. Her research and teaching are focused on problems at the intersection of engineering, development, and sustainability. Her work has led to over 110 publications, an edited book, and several opinion editorials and newspaper articles. Her research has been recognized by several national and international agencies. She is a contributing author to the sixth assessment report (working group II, chapter 4) of the Intergovernmental Panel on Climate Change (IPCC) on implications of water, energy, and food interconnections for climate change adaptation released in 2022. She was also commissioned by the United Nations Committee of Experts on Public Administration (UN-CEPA) for a background paper on Climate Change Action and Sustainability of Natural Resources for their 21st session in New York in April 2022. Her work on a new method for predicting long-term future water availability in arid regions was noted and invited for presentation at the US National Academies in Washington, DC, and was included in the academy’s decadal survey in social and behavioral sciences in 2019. Her study on the water-energy-food nexus of agricultural production was invited for presentation at the national Planning Commission in Pakistan, and she was invited by the International Institute for Applied Systems Analysis (IIASA) in multiple international knowledge forums. She has been a recipient of the Amelia Earhart Fellowship, Richard D. DuPont Fellowship, and the Rene H. Miller Prize in Systems Engineering. Prior to engaging in research, she worked as an Applications Engineer and later as Software Engineer in Research and Development (R&D) at National Instruments in Austin, Texas, and co-developed the Controls Toolkit for LabVIEW. More recently, she has collaborated with industrial partners on optimization of autonomous driving systems in future

9 Remote Sensing Through Satellites and Sensor Networks

275

autonomous vehicles (AVs), and quantifying value of radiometric calibration and data quality in remote sensing data products. She has conducted modeling and analysis of communication satellites, earth observation satellite constellations, planetary surface vehicles for exploration of Moon and Mars, hydropower systems, desalination plants, large-scale irrigation systems, and waste-toenergy systems. Overall, she is motivated by urgent problems of development and works toward enabling sufficient and smart infrastructure, equitable use of technology, and sustainable resource use for improving human well-being. She serves as a reviewer for several scientific and engineering journals, provides expert review for international fellowships, serves on scientific technical committees for several professional societies, teaches graduate-level and professional education courses, and engages internationally with policy makers on issues of development, technology, and sustainability.

Chapter 10

Uphill on Two Fronts Carole Perry

Back in the 1970s when I first entered the wireless world of amateur radio, the world was a very different place, or was it? In retrospect, it seems to me that some underlying patterns of human behavior have always been there and still are to this day. The experiences I have had both in the wireless industry and in the hobby and service of amateur radio are truly worth sharing. In the early 1970s, I was offered the position of Educational Consultant and Gal Friday at a Brooklyn, New York electronics manufacturing company. It was a startup company with some very talented young engineers and an innovative entrepreneur who had two patent – pending’s on wireless devices. The company had automatic insertion machines as well as a small plating facility and therefore had the capabilities of manufacturing prototypes for other companies as well as for in house production. Among other responsibilities, I became the person who took “tours” through the small facility. To say that visitors from the business community were surprised to see a woman who was knowledgeable about manufacturing circuit boards would be an understatement. Had I not been married at the time, I could have enjoyed a very active social life. It was not unusual for the tour to end with an invitation for lunch or dinner. This would be the beginning of my experiences with dealing with the reactions of people in a male-dominated field and their preconceived notions of gender roles. I was soon introduced to the world of HVAC (heating, ventilation, and air conditioning) trade shows. One of the devices our small company was promoting was our invention of the first digital clock wireless set back thermostat. I became the main demonstrator at trade shows all across the country. Really technical questions were referred to the engineers who were there as well. One very memorable

C. Perry () Radio Club of America, Island, NY, USA © The Author(s), under exclusive license to Springer Nature Switzerland AG 2023 M. S. Greco et al. (eds.), Women in Telecommunications, Women in Engineering and Science, https://doi.org/10.1007/978-3-031-21975-7_10

277

278

C. Perry

experience took place at the McCormick Convention Center in Chicago during one of these shows. I was in the ladies’ room and soon realized the lock was broken on my stall. I called out for about 10 minutes and then realized the unlikely possibility of another female walking in there anytime soon. I was right. These were all pre-cell phone days, of course. The memory of my slipping out from under the door is one of the reasons that to this day, I never let the cell phone out of my hand when I am off by myself at a male-dominated event. Because I was instrumental in closing the deal on a major purchase of the thermostats from a large utility company, I was selected to accompany the engineers to the manufacturing company in Hong Kong that we had hired to do the large runs for us. One day, while there, I went unaccompanied to speak with the head of the assembly line at their plant. I sat in the reception room, ignored, for 20 minutes. When I finally got up to inquire why I was being kept waiting when my arrival had been announced, here was the response I got from the female receptionist, “I just assumed you were waiting for your boss to come.” It was unthinkable in those days that a female could have been the decision maker or the person in authority. I was always treated respectfully while in Hong Kong, but it was clear at every single introduction that my position as the person in charge had to be clarified. Carrying business cards that declared me as the executive vice president (I was promoted through the years) helped enormously. It was a fascinating cultural enlightenment, being a female and being a tall woman as well. I continued to conduct business, always mindful that I needed to be a notch better than my male counterparts to be taken seriously. At some point I even got used to being stared at as I “towered over” my Hong Kong business associates when in close contact such as in an elevator. After 2 weeks, I could sense that the familiarity with who I was and why I was qualified to be there dominated the reactions to be more comfortable for all of us. In the end, I was sent there twice and thoroughly enjoyed the international experience. Back at our plant in Brooklyn, New York, there were several engineers who were amateur radio operators (hams). During our lunch breaks, they would invite me to get on the air at the station they had installed. I thoroughly enjoyed the experience of reaching out and contacting hams locally and in other countries. The local hams didn’t seem surprised to hear a female voice because I was being introduced by the licensed hams. It wasn’t unusual to have husbands and/or fathers pass the microphone to their wives and/or children. The “unusual” thing would have been for a licensed female operator on the air in those days. One day, after a really enjoyable QSO (conversation) with a ham in Florida, I said, “That was really fun. Maybe I’ll get my own FCC license.” The response, which I have never forgotten, all these years later from one of the engineers, was, “It is fun. But you’ll never get a license. This is something women just don’t do.” They might as well have dropped a gauntlet on the floor. I was then determined to get my FCC amateur radio license, even if I never picked up a microphone again. From that time on, I listened intently to the hams as they spoke on the air. I paid attention to the procedures and protocols and studied from the few radio manuals

10 Uphill on Two Fronts

279

that were available at that time. When I felt ready to take my Novice license exam, I went to a session with others where a volunteer examiner gave us the exam and tested our CW (Morse code) proficiency at five words per minute. I was really excited to have passed and took every opportunity to get on the radio after that. To say that getting involved with ham radio was a life-altering decision for me would be a major understatement. I was to learn the extraordinary value of ham radio operators providing emergency communications, as well as the fun of engaging with hams from other countries, thereby promoting international good will. Since one of the basic tenets of amateur radio is volunteerism, I even had the extraordinary experiences of being a radio volunteer at the New York City Marathon for several years, the Israeli Day Parade in Manhattan, and countless local community fund raisers and mall demonstrations. When the electronics manufacturing company decided to relocate to another state, I had some big decisions to make. I eventually decided to return to my first love of teaching. I phased back slowly doing substitute teaching work in the local middle schools. The principal at one middle school where I frequently worked offered me a position for one term in the Industrial Arts and Technology Department, commonly referred to as the Shop Department. I was allowed to pick my own choice of curriculum, since it was “only temporary” while the Shop teacher was out on sick leave for 5 months. I convinced a very enlightened principal to let me teach ham radio to the classes I would be seeing twice a week. I brought my portable 2-meter rig into the classroom but wasn’t able to get reliable communications due to the steel girders and structure of the building. Despite certain obstacles, the students seemed genuinely interested in learning about a hobby where they could get licensed and get their own radios. This was back in the early 1980s when CB radio was all the rage. When pointing out the differences between CB radio and ham radio, their enthusiasm grew. By all standards; it was a very successful experience in teaching something new. My background was in teaching science, and I suspected the principal would be offering me a full-time position in the Science Department for the next term. I ran the idea of suggesting a ham radio class past my friends and fellow hams. This was my introduction to the world of naysayers who are always there to assure us we shouldn’t be looking to expand our horizons, for fear of failure. No one believed that the NYC Board of Education would ever approve the teaching of such a course. I was also assured that no principal would want such a program full time in their school. Once again, I followed my own instincts and volunteered to be a guest speaker at the school’s PTA (Parent Teacher Association) meeting one evening. I planned a really fun demo with my radio and told the parents about the benefits, educationally, to their children. I also emphasized the possibilities that they, too, could get an FCC Amateur Radio license along with their children. So before even approaching the principal with my innovative idea, I knew I had the support of the parents behind me. Most principals in most schools are extremely eager to please and to support the parents, who are usually organized in a PT or PTA who are a terrific group that raises funds for school projects and who volunteers to help at most school events.

280

C. Perry

The principal was receptive to my proposal to teach an introductory course in Amateur Radio that was an enrichment program for motivating the students in all areas of the school’s curricula. It would be an experiment for one term. I ended up teaching the course for almost 30 years and brought tons of great publicity to the school with our numerous astronaut contacts that brought citywide media attention to my program, to our school, and to the New York City Board of Education. The biggest impact in my life, of getting involved with ham radio, was my fortuitous opportunity to write curriculum for the New York City Board of Education along with other selected hams and then to be able to teach “Introduction to Amateur Radio” for almost 30 years. I taught the radio program in a Staten Island Middle School to 6th, 7th, and 8th graders. My program was under the aegis of the Industrial Arts and Technology (Shop) Department. I taught 12 classes most terms, meeting each class twice a week, and one class three times a week. I often received up to 40 students in each of the heterogenous classes. The class was not an elective. The administrators loved the idea that it was taught as an enrichment class, not as a licensing class. That way, every single student could be programmed into my class during their tenure at the school. I created lessons that applied to the 6th, 7th and 8th grade curricula based on the radio contacts we had made. Therefore, a radio contact with a ham in Texas would prompt a social studies lesson about that state. It also would include pulling a string of yarn on our world map on the wall from Staten Island, New York, to Houston, Texas, so that we could visualize where that ham radio operator had contacted us from. Geography was taught on a need-to-know basis. I believe that was a highly effective way of teaching geography skills. Some of the students would locate the origin of the contact in atlases or on paper maps, thereby learning research skills. Sometimes we were lucky enough to have the hams we met on the radio visit us in person in our classroom. When we had an ongoing scheduled meeting on 20 meters with Father Mike from Sierra Leone, Africa, he promised to visit with us if he were ever in New York. He was a very exciting guest when he kept that promise and he got to meet with all the students he had spoken with on the air. Of all the interesting, creative, and unique contacts we made through the years, the ones that I never got jaded about or failed to get incredibly excited about were the many astronaut contacts we were fortunate to make. In order to involve young people, NASA and the Amateur Radio Relay League (ARRL) and the Amateur Satellite Corporation (AMSAT) created a program that would enable school kids around the world ask questions of astronauts while they were on board the space shuttles. The Shuttle Amateur Radio Experiment (SAREX) program facilitated communications between astronauts in orbit with students on the ground. With the help of local Staten Island amateur radio clubs, in the summer of 1985, we set up all the equipment needed to make a scheduled contact in our school auditorium. All the media was there to capture our scheduled contact. When the first pass over NY proved to be unsuccessful for a radio contact, a TV reporter asked me how I felt about our failure. After my 10-minute speech explaining how inaccurate the use of the word “failure” was, the TV crew agreed to stay for the next pass and the next attempt

10 Uphill on Two Fronts

281

at a contact. They didn’t realize the shuttle circles the globe every 90 minutes. They apologized for being pessimistic. Thanks to the determination and skill of all the hams who put so much time and effort into the project, and the fact that astronaut Tony England W0ORE was able to free up time, we were able to send my picture up to the shuttle via slow scan TV (SSTV) and the audience went wild when Tony’s picture was received via SSTV in our school auditorium. We brought great publicity to the school by being featured in the NYC newspapers and on the TV news that night. I wrote lesson plans for the ARRL which were shared with other teachers who had scheduled contacts with the shuttles. I created several lessons in our unit on Space Communications that were a terrific background for the numerous astronaut contacts we were lucky enough to have made during the years. My students and I were thrilled when I was invited to attend the launch at the Kennedy Space Station of the “All Ham Crew” on board the eighth flight of the Space Shuttle Atlantis, a six-day mission with the primary goal of launching the Compton Gamma Ray Observatory. Having been privileged to have met the crew in person at the Johnson Space Center where they were training for the mission, it was even a more incredible experience that I could have ever imagined. When I heard the announcement “We have liftoff of the Atlantis” I had an unparalleled surge of pride. I also thought to myself about all the people who tried to discourage me from pursuing my goals of teaching the program that led to this moment. This was one spring break when the teacher would have the best “Show and Tell.” The SAREX program was eventually superseded by the Amateur Radio on the International Space Station (ARRIS) program. Through the years, because the astronauts are groomed to be encouraging of young people to get involved with the space program, they would always volunteer to speak at my Youth Forums whenever they appeared at ham radio conventions I was at. It has been my honor to have met 16 of the United States astronauts in person; many of whom we had spoken to via the radio in my classroom. Without a doubt, these are some of the most fascinating people in the world. I was especially delighted to have met several female astronauts so that I could relay their struggles and stories back to my students. I found it fascinating to learn that many of the astronauts (both male and female) had to apply for the astronaut training program numerous times before they were accepted. Evidently the trait of “pushing back” when others around you are telling you to give up is a common theme among many successful people. One would expect, of course, that astronauts would be determined, persevering, and tenacious individuals. Whenever I speak in person at different STEM and STEAM classes across the country, I make sure to include the stories of the backgrounds of the female astronauts who are still breaking records and making history in space travel today. In my opinion, the astronauts and mission specialists who risk their lives while 220 miles above earth are the true pioneers of our times. In May of 1987, a great honor was bestowed on me. I was awarded the prestigious Dayton Hamvention 1987 Amateur Radio Operator or the Year title. While waiting for the car to arrive that would take me to the Awards Banquet at the Dayton

282

C. Perry

radio convention, a ham approached me to ask how I was enjoying the convention weekend. We chatted for a while, and then he asked if my husband whom he assumed was the ham was enjoying the weekend as well. I remember telling him he would have to ask my non-ham husband himself, since I didn’t want to be late receiving my award that evening. To this day, 33 years later, that same ham finds me at Dayton Hamvention every year, so we can both laugh together. That same year, the ARRL bestowed their Instructor of the Year award on me. The recognition of my work with young people in radio from organizations like those two, that I have so much respect for, has served to motivate me every time I began a new initiative with students below the age of 18. Invariably, there was someone giving me all the reasons why these new projects and approaches would never work and would never be accepted. The fact is that many national and international radio organizations invite me to speak at their gatherings because they now recognize the importance of recruiting and retaining young people in technology. My approach is to engage and motivate them through the fun of ham radio in the classroom, in scouting, and in youth groups. Lots of things have changed in this world since my early experiences in the 1970s, fielding the preconceived expectations of women’s roles, especially in technical fields and endeavors. So, let’s fast forward to 2019 as I was returning from the world’s biggest international ham radio convention in Dayton Ohio. I had just completed my 32nd year as moderator of the Hamvention Youth Forum and Instructors’ Forums. I was heading home, on the TSA pre-approval line, at the Dayton International airport, when I was detained by the agent at the podium. He kept reading and rereading my pre-approved boarding pass and looking at my ham radio jacket which had my name and call sign embroidered on it. He then announced that he could not let me through the pre-approved line that day. I would have to go through the regular check in process. When I respectfully inquired why, here was his response: “I see you are a ham radio person, and I know women don’t do that.” There it was! It was 2019 and once again I was being told that women don’t do technical things. Since almost everyone in the TSA line behind me knew of my work with young hams and of my 32 years of moderating two prime forums at one of the largest international ham radio conventions in the world, they laughed and spoke up. A gentleman behind me assured the TSA agent that I was a well-known person at the convention with a prominent role there. With the public verification that it was okay to allow me through into the airport, the agent stared at me and, with apparent reluctance, allowed me through. At that moment I didn’t know whether to feel grateful and to move along or to be totally humiliated that I needed to be vouched for due to the agent’s preconceived notions about the role of women. Today, I choose to think about all the incidents, like this one through the decades, as how I was forced to push through other people’s assumptions in order to pursue my own lifestyle choices. Happily, because I have done that, I have managed to carve out a career and to have a very productive and gratifying retirement with my volunteering efforts via the Radio Club of America. This non-for-profit organization, founded in 1909, is the world’s oldest communications society.

10 Uphill on Two Fronts

283

I was elected to the RCA Board of Directors in 2008 and have been elevated to Fellow and made Chairperson of the RCA Youth Activities program, an initiative I spearheaded. I have created the RCA Young Achiever program which showcases and supports talented young hams whom I select to give presentations at my national and international Youth Forums. Many of these young people stay in touch with me and keep me informed about their choice of technical schools as well as their career choices in the wireless industry, primarily. I often choose them to give presentations at the RCA Technical Symposium, where as adults, they can be examples of the importance of supporting and encouraging them in their technical pursuits as youngsters. One of the initiatives I really have enjoyed via my RCA Youth Activities program is traveling to different museums across the country to facilitate setting up ham radio licensing classes on the weekends. This has proven to be especially successful in children’s and science museums. There is usually a local radio club who invites me to assist, thereby providing a networking and support opportunity for those who are successful in getting their FCC licenses. One of the most successful programs for introducing young ladies to the wonderful world of science has been through STEM in classrooms and clubs. I am frequently invited to introduce ham radio into established programs already using the STEM approach to learning. I believe I was using this approach in my classes long before the word was even coined. I always encouraged my students to analyze and strategize about problem solving rather than just “spitting back” answers to my questions. I always felt gratified when I generated more questions than answers on their part. A STEM program should allow students (boys and girls) to contribute to something larger than themselves. Amateur (ham) radio was formed by experimenters who worked to push the boundaries of radio technology. Ham radio in the classroom is the perfect complement to the STEM (Science, Technology, Engineering, and Math) format. In my decades of teaching with this approach, I observed that the girls did just as well as the boys when it came to excelling at radio technology and attaining FCC licenses. STEM in schools, utilizing ham radio in the curriculum, invites young ladies to get exposed to a new way of approaching problems and to areas of science, like radio astronomy, aerospace, satellite technology, weather phenomena, building circuits, and radio technology they might not otherwise encounter. My volunteer work in RCA with Youth Activities has allowed me to forge inroads into male-dominated areas of study and into previously all patriarchal systems in education and in the wireless industry. In closing, I will share that when I am invited to give graduation speeches of encouragement every year, at some of our local schools, I always lead with this advice, “When you are passionate about something; surround yourself with YaySayers; NOT Nay-Sayers. Don’t just follow your dreams; aggressively RUN after them. You may just catch one.”

284

C. Perry

Carole Perry WB2MGP worked as an executive secretary in an electronics manufacturing company, Rapid Circuit Inc. for 16 years. In 1980, when the company relocated she returned to Intermediate School 72 in Staten Island, NY, where she worked until her retirement in 2004, teaching “Introduction to Amateur Radio” to 6th, 7th, and 8th graders for almost 30 years. Carole wrote the curriculum for “Introduction to Amateur Radio” a very successful program which had 950 students a year coming through it. Carole Perry is the recipient of the prestigious 1987 Dayton Ham of The Year Award, the 1987 ARRL Instructor of The Year Award, the 1991 Marconi Wireless Memorial Award, the 1993 QCWA President’s Award, the 1996 Radio Club of America (RCA) Barry Goldwater Amateur Radio Award, the 2009 RCA President’s Award, the 2012 RCA President’s Award, and the 2015 Vivian Carr Award for Women in Radio. She is the winner of the 2016 SOAR (Sisterhood of Amateur Radio) Legacy award for Pioneering Women in Amateur Radio and the 2016 recipient of the YASME Foundation Award for Excellence. In 2017 she was the winner of the Brooklyn College Milton Fisher Second Harvest Award for her volunteer work with young people and technology, around the world. In May 2018 Carole was inducted into the “CQ Amateur Radio Hall of Fame.” In July 2018 “QST” magazine, Carole was the featured member in “Member Spotlight.” In February 2019, at Hamcation, Carole became the first recipient of the newly created “Carole Perry Educator of the Year Award.” In 2020 Carole was awarded the position of Fellow in the AWA (Antique Wireless Association). Carole is an RCA Fellow, and in 2007 she was elected to the RCA Board of Directors, a position she still holds, and she created the Youth Activities Committee which she now chairs. She serves on the RCA Scholarship committee as well. She also created the RCA Young Achiever’s Award, given to students in grade 12 and below who have demonstrated excellence and creativity in wireless communications. One hundred and forty-one youngsters have received this award along with a stipend, so far. Carole is also presently a Director for QCWA (Quarter Century Wireless Association). The QCWA Youth Activities program was created and is chaired by Carole. Under Carole’s leadership, the RCA Youth Activities Committee goes into schools across the country to set up radio/technology programs. Equipment, cash grants, books, and supplies are donated to the chosen schools or youth groups. Carole has moderated the Dayton Hamvention Youth Forum and Instructors’ Forum for 32 years. She is a member/director of QCWA and RCA. She is also a member of ARRL, DARA (Dayton Amateur Radio Association), AWA (Antique Wireless Association), Portage County Amateur Radio Society, YL Harmonics, and Brandeis Women. Carole is a contributing columnist for CQ Magazine’s Youth Column.

Chapter 11

Access, Inclusion, and Accommodation Katherine Grace August

11.1 Background Many people become engineers to create solutions that will reduce suffering and improve the human experience. In early 2020, COVID-19, a transformative event, resulted in a rapidly developing worldwide humanitarian crisis requiring substantial technology solutions. Now is the time to investigate and understand the influences that will result in greater team engagement for solutions in computing and engineering that hold great potential to address healthcare and economic solutions, to benefit the wide diversity in our society addressing culture, language, gender, race, age, ethnic, ability, education, healthcare, and economic and technology gaps now and into the future (Azoulay and Jones 2020). From the earliest days of online services and business computing, there was anticipation and the great promise of a wide sweeping global community that would result from the most significant innovations since the Industrial Revolution. There was an optimism that everyone would benefit from these advances and we would experience a renaissance in our society. Still, many systems fail to meet needs and fail without sufficient alternative solutions revealing ongoing gaps and vulnerabilities. At the turn of the millennium, we presented a vision of the future services and access (August et al. 1999) – a world connected by a network of networks, providing mobile and fixed communication on a very large scale, information, and sophisticated features that incorporated the greatest advances in science and technology, that advantaged business development and consumer experience, and that promised to improve lives and opportunities, with communications services and access anytime, anywhere (August et al. 1999). Those networks and technology

K. G. August () Stevens Institute of Technology, Hoboken, NJ, USA © The Author(s), under exclusive license to Springer Nature Switzerland AG 2023 M. S. Greco et al. (eds.), Women in Telecommunications, Women in Engineering and Science, https://doi.org/10.1007/978-3-031-21975-7_11

285

286

K.G. August

platforms connected more individuals, governments, and businesses with more solutions than ever before. Twenty years later, many advances have been realized including the Internet of Things (IoT), web-scale IT, sophisticated implanted biomedical solutions, artificial intelligence and machine learning (AI/ML), smart homes, and robots. Yet, when the COVID-19 global pandemic emergency struck, it became obvious that so many of these advances benefited only segments of our community and businesses and only certain portions of our geography. Many shocking outcomes resulted from the disparities in our society including the digital divide. And upon reflection, public health, education, telecommuting, government, information, entertainment, and many other critical aspects of our lives were not entirely supported by the existing infrastructure, policies, and processes. Digital broadband services are now considered infrastructure, like roads, electricity, and water. It’s no longer optional; it’s essential. This reality has had a tremendous effect on underrepresented, minorities, those with disabling conditions, rural communities, economically disadvantaged, undocumented, their families, communities, and many others with no end in sight because of high costs and lack of infrastructure. It also seems likely that previously existing disparities will continue to cause harm after the COVID-19 emergency is resolved. The federal government has responded with programs and strategic plans (Federal Communications Commission 2021a). Disparity of technology and other influences including the needs represented in our society diversity such as Culturally and Linguistically Appropriate Systems – Health and Human Services (CLAS-HHS), effects of Social Determinants of Health (SDoH), and aims of the United Nations Sustainable Development Goals (UNSDGs) have been unmasked by the COVID-19 emergency (Centers for Disease Control 2021; United Nations Sustainable Development Goals UN SDG 2021). Whereas technology advances might improve opportunities for everyone, many people who experience disparity are instead at greater risk. Many were disproportionately impacted with the most horrific outcomes of morbidity and mortality. When some of us do not do well, the rest of us will not do well. Without broadband, people cannot telecommute, learn information about risks from coronavirus, obtain instructions from public health officials, remotely visit the doctor, or educate their children in the home. Many people do not have skills or resources to work with technology or help their children with education. These circumstances widen the gap. This has a lasting impact. Those who depend on others experience a widening gap, for example, when they are living in a group home, nursing home, or congregant setting or because they rely on a library or another source for connectivity, if they need hearing or vision assistance, or if they need human language interpreters, and the like. If these services are unavailable for basic needs, for COVID-19 and other information, for work, school, or to care for others, people experience widening gaps. They experience disparity and vulnerability across all human needs. Many people in the United States have no access to adequate broadband, smartphones, affordable service coverage, telemedicine, or adequate user interface. Underrepresented, minorities, elderly, and people with disabling conditions not only suffer great disparities in education, workplace, and the economy, but they

11 Access, Inclusion, and Accommodation

287

also face mortality and morbidity in significantly greater numbers, a tragedy. The consequences impact individuals, families, communities, and nations and will continue through generations. In our modern world, we face challenges that rely on technology solutions. Solutions are often designed to engage some version of average; however, we are a very diverse society. Even the field of engineering fails to represent the diversity in the community. Without representation, needs are not well understood and solutions fail to meet needs. It’s important to capture the lived experience. Although there are many programs designed to increase diversity, there are very few women and minority engineers. Lack of diversity in the engineering profession is a critical and complex problem to investigate and resolve. Future systems will require a new approach to more effectively reach people, to provide access, to engage more representative individuals through inclusion, and to engage the diverse abilities of humankind through accommodation. Given the substantial role technology plays in our success as human beings and society, our technology solutions should represent the community of people and also include them in its benefits. This chapter will investigate some of the influential mechanisms and issues at the intersection of digital disparity. Unmasked and escalated by COVID-19 and related problems, there is potential to shape a new vision of the future of the digital access ecosystem with inclusion and accommodation to meet the extraordinary needs of our diverse population.

11.2 The Future of Access, Inclusion, and Accommodation Future systems will require a new and modernized approach to more effectively reach people where they are through digital connectivity access, engage more representative individuals through inclusion, and incorporate the diverse abilities of humankind through accommodation. In order to achieve these goals, engineers must become aware of complex and multidimensional influences leading to excellent outcomes. Inclusion and diversity on the team and in the profession is critical for success envisioning and creating new technologies and solving complex problems to meet the needs of a diverse society. We must turn our attention toward human-centered design; employ science, innovation, and technology; and collaborate on solutions that incorporate diverse representations of all humans. We must develop our skills as engineers in order to be prepared to pivot and to face new problems and opportunities, addressing each technology with its own properties and characteristics, considering consequences and impact, alone and in combinations or an ecosystem, to achieve critical success. We must encourage others, too. Find opportunities to engage, mentor, and recognize achievements. Human-centered design can improve resilience, with increased transparency, to reduce and/or eliminate digital disparity, and improve outcomes and well-being. Human-centered design incorporates a variety of properties at the intersection of

288

K.G. August

problems and solutions, systems, technology, communication, human interactions, and human performance. Important considerations include but are not limited to language, cultural preferences, ease of use, cost of solutions, plug and play devices, and so forth. For some users, frequent updates to a system might render the system challenging and potentially useless. In some regions service providers distribute equipment without persistent representation of the user changing the way activity is monitored. Some people don’t read well and memorize menu selections; when the service provider makes a change, the system is not easy to use. If the person doesn’t use the system frequently, or if the person has challenges learning how to use new features, then the cost to the individual is high. For example, if an elderly person already knows how to use a flip phone, and the cost of the flip phone is more appropriate than a more complex and expensive smartphone, then telemedicine provided on a flip phone is more likely to be easy to use for that group. But if telehealth requires a smartphone, then that person cannot access the service. A person need not be old to have challenges; when a person is infirmed, they may not be performing at their best making it difficult to use more sophisticated technology. A doctor might be able to effectively monitor a patient at home if the patient can use the device or telephone and if the phone is connected to reliable power and service, digital access. A more important part of the picture involves inadequacies of access itself. Engineers must attend to innovations to extend affordable broadband connectivity to all. If a person lives in a community without access to affordable broadband, then telemedicine, education, work at home, social connections, and other services and systems are simply not available to that person. Tragically, that is the situation in many places in the United States and around the world. The federal government Federal Communication Commission (FCC) strategic plan 2018–2022 includes: • Closing the Digital Divide – High-speed Internet access, or broadband, is critical to economic opportunity. But there are too many parts of the country where broadband is unavailable or unaffordable. The FCC has tools it can use to help close this digital divide, bring down the cost of deploying broadband, and create incentives for providers to connect consumers in hard-to-serve areas. • Promoting Innovation – A key priority for the FCC is to foster a competitive, dynamic, and innovative market for communications services through policies that promote the introduction of new technologies and services. We will ensure that the FCC’s actions and regulations reflect the realities of the current marketplace, promote entrepreneurship, expand economic opportunity, and remove barriers to entry and investment. • Protecting Consumers & Public Safety – The FCC’s core mission has always been to serve the broader public interest, and that means protecting consumers and keeping the public safe. We will work to combat unwanted and unlawful robocalls, which intrude into consumers’ lives, and to make communications accessible for people with disabilities. We will also protect public safety, and in particular, take steps to assist and safeguard the communications of our nation’s law enforcement officers and first responders. • Reforming the FCC’s Processes – As Chairman, I have made it a priority to implement process reforms to make the work of the FCC more transparent, open, and accountable to the American people. We will modernize and streamline the FCC’s operations and programs to improve decision-making, build consensus, reduce regulatory burdens, and simplify the public’s interactions with the Commission” (Federal Communications Commission 2021b).

11 Access, Inclusion, and Accommodation

289

In addition, people should be able to use the solutions, and the solutions should represent the best fit for needs. New communication and technologies represent new challenges for research, innovation, and human factors. Considering humancentered design can improve how technology can meet the needs of individuals and society, for example, facilitating plug and play solutions, understanding the user’s lived experience, engaging in standards development, involving stakeholders and community, addressing a range of segments of society, and extending the offerings. There are opportunities to incorporate legacy technologies, reducing power consumption, recycling, and incorporating affordable devices. To improve the human interface experience future systems should facilitate adapting to languages and cultural experience, ensure transparency and explainability, employ adaptive interfaces and universal design, explore approaches to address human learning curve for technology and subject matter, and find solutions for individuals and communities locally and globally.

11.3 Innovation, Inclusion, and Diversity Innovation can emerge anywhere, even in non-traditional and unexpected places. Fostering innovation can improve options. Future innovations and solutions for access will benefit from inclusion and diversity. Yet while access relies on innovations and advances in engineering, the field of engineering fails to represent the community at large. Slightly more than half of the people in the United States are women, and more than half of college graduates in the United States are women, yet the field of engineering itself does not represent the general population. Statistics indicate women tend to follow other women into specific fields of science and engineering indicating the significance of recognizing women as role models. Industries are not equally distributed around the United States, and therefore, women’s patents are associated with residents of those states representative of specific industries. Women following women into specific fields result in patents that include women in those particular industries and fields as depicted in Fig. 11.1. Some fields enjoy greater visibility in society and are more accessible to enter and to experience early support and success. Engineering takes more preparation, time, and resources, and once in the field, advancing seems to favor the dominant culture. People can gain experience and education at any age; lifelong learning is important to individuals and to the field of engineering. People can learn to develop their ideas and invent at any age. Inventing is an important way to record a person’s ideas and contributions. For information about inventing, and to connect to free education about patenting, please see the United States Patent and Trademark Office (USPTO) website, USPTO.gov (United States Patent and Trademark Office (USPTO) 2021) (Fig. 11.2).

290

K.G. August

Women Inventors at Select Top Patent Assignees, 2007-2016

PROCTER & GAMBLE CO

PROCTER & GAMBLE CO

BRISTOL-MYERS SQUIBB CO

BRISTOL-MYERS SQUIBB CO

ABBOTT LABORATORIES

ABBOTT LABORATORIES

MIT

MIT

AT&T INC

AT&T INC

XEROX CORP

XEROX CORP

IBM CORP

IBM CORP

3M CO

3M CO

VERIZON INC

VERIZON INC

EXXON MOBIL CORP

EXXON MOBIL CORP

ORACLE CORP

ORACLE CORP

UNITED STATES NAVY

UNITED STATES NAVY

MICROSOFT CORP

MICROSOFT CORP

QUALCOMM INC

QUALCOMM INC

GOOGLE INC

GOOGLE INC

INTEL CORP

INTEL CORP

TEXAS INSTRUMENTS INC

TEXAS INSTRUMENTS INC

GENERAL ELECTRIC CO

GENERAL ELECTRIC CO

AMAZON INC

AMAZON INC

CISCO SYSTEMS INC

CISCO SYSTEMS INC

ADOBE SYSTEMS INC

ADOBE SYSTEMS INC

APPLE INC

APPLE INC

BOEING CO

BOEING CO

HONEYWFII INTERNATIONAL INC

HONEYWFII INTERNATIONAL INC

FORD MOTOR CO

FORD MOTOR CO

LOCKHEED MARTIN CORP

LOCKHEED MARTIN CORP

RAYTHEON CO

RAYTHEON CO

ANALOG DEVIES INC

ANALOG DEVIES INC

CATERPILLAR INC

CATERPILLAR INC

DEERE & CO

DEERE & CO

0

5

10

15

20

25

WOMEN INVENTOR RATE (PERCENT)

30

0

1000

2000

3000

4000

5000

WOMEN INVENTOR RATE (COUNT)

Fig. 11.1 Women inventor rate top assignees by percent/count by individuals

There is significant disparity in hiring, promotions, incomes, patenting, opportunities, and recognition. Role models are needed. To increase the number of role models and improve inclusion and increase diversity, it is important to document and share our history. History can be an important source of inspiration. History is not just recording the distant past, it’s also recording the way we do things, our lived experience. We can write our own history. Recognition by traditional engineering measures is needed. Recognition is a way to develop role models and is needed to inspire and engage women, minority, underrepresented, and disabled engineers to the field and to successes in the field. Recognition is often provided through professional societies. In addition, writing and publication not only provides an opportunity to collaborate and share ideas, it also provides an opportunity to increase impact represented by a number of measures. Additionally, through inventing, engaging in standards work, becoming involved with humanitarian activities, we can network, share ideas, solve new problems, encourage others, develop resources, and increase visibility of women, minorities, underrepresented, and disabled. By documenting our ideas, we introduce our point of view, our style of problem solving that represents inclusion and diversity. We can make decisions every day in our work and other activities that are more informed

11 Access, Inclusion, and Accommodation

291

National Center for Science and Engineering Statistics | NSF 21-321

Median annual salary of scientists and engineers employed full time in S&E occupations, by age, highest degree level, and sex: 2019 140,000 120,000

Dollars

100,000 80,000 60,000 40,000 20,000

Ba Ba ch ch el or el or ’s ,a ’s ,2 ll ag 9 an es d y ou Ba ng ch er el or ’s Ba ,3 ch 039 el or ’s Ba ,4 ch 049 el or ’s ,5 075 M as M as te r’s te r’s ,a ,2 ll ag 9 an es d yo un M as ge te r r’s ,3 M 0 as -3 9 te r’s ,4 M 0as 49 te r’s ,5 075 D oc D oc to to ra ra te te ,a ,2 ll ag 9 an es d yo D u oc ng to er ra te ,3 D oc 0to 39 ra te ,4 D oc 0to 49 ra te ,5 075

0

Degree and age Women

Men

S&E = science and engineering. Note(s): S&E occupations include S&E postsecondary teachers. Salaries are rounded to nearest $1,000. Here only S&E occupations are included; S&E-related and non-S&E occupations are excluded. Source(s): National Center for Science and Engineering Statistics, National Survey of College Graduates, 2019. Related detailed data: WMPD table 9-16.

Fig. 11.2 2019 median salary scientists and engineers by highest degree/age women/men

about the influences and outcomes, to more effectively learn about how people experience their lives and technology, and reach them where they are through improved innovations in digital connectivity access, engage more representative individuals through inclusion, and incorporate the diverse abilities of humankind through universal design and accommodation.

11.3.1 Improving Inclusion and Increasing Diversity Although there are many programs designed to improve inclusion and increase diversity, there are still very few women, minority, underrepresented, and disabled engineers. Patterns observed in the pipeline predict continued disparity. For example, advanced degrees in certain science and engineering professions have declined since 2016, and there remains disparity in employment, salary, and patenting. Disparities lead to additional and accumulated disadvantages over a lifetime, for individuals, families, communities, and also for its impact upon technologies and what technology can offer to society. Programs continue to evolve.

292

K.G. August

Professional membership organizations offer many opportunities for engineers. Professional membership organizations can contribute by providing a community and opportunities for networking outside the limitations of a specific job. They can provide continued education and experience and opportunities to publish and attend conferences, to recognize achievements and achievers in the field, to promote role models, to become involved in humanitarian activities, and to engage individuals, communities, and stakeholders locally and globally. An important area for achievement for engineers is inventing. US patents tell the story of the progress of science and technology and also give insight into the state of inclusion and diversity in the fields of computing and engineering. Patents hold potential for an individual or team to record their own innovative ideas and for those ideas to be preserved in a meaningful way. The USPTO has a mandate to be inclusive, with a number of new strategies; they also use technology, employing artificial intelligence and other methods, to uncover more knowledge and understanding of ideas and innovations and how they are interrelated. A number of reports are available to explain the state of inventing in the United States from a variety of perspectives. For example, Progress and Potential: 2020 update on US women inventor-patentees USPTO (Progress and Potential: 2020 update on U.S. women inventor-patentees 2021). The SUCCESS Act Report to Congress involves participation of women, minorities, and veterans in the patent system (USPTO releases SUCCESS Act report to Congress 2021). In her Letter to the Members of the National Council for Expanding American Innovation (NCEIA), Gina Raimondo, Secretary, US Department of Commerce recognizes that only 12.8% of inventors listed on patents are women and that little or no information is available about other underrepresented groups (Fig. 11.3). She agreed to Chair the NCEAI and renamed it Council for Inclusive Innovation, or CI2 to “build back better through the enhanced diversity, equity, inclusion and accessibility (DEIA)” (Letter to the Members of the National Council for Expanding American Innovation (NCEIA) 2021). Industries are not evenly distributed around the United States. For many years, New Jersey has been involved with telecommunications, pharmaceuticals, and related fields. Our local statistics might reflect not only the representation of women in engineering but also women in telecommunications. The Institute of Electrical and Electronic Engineers (IEEE) is the largest nonprofit professional engineering membership organization focusing on advancing technology for humanity (Institute of Electrical and Electronics Engineers 2021). In our local section with 1124 members, we find only 84 women, 14 undetermined. Of those members, 5 women have been recognized as Fellows, while 69 men have been recognized. In our section, we began to investigate history in order to inspire our Members and encourage others to explore and understand the contributions of engineering. We uncovered several opportunities and engaged in new projects and programs (Interview with Kit August on the ways New Jersey Coast Section is Bringing History Alive 2021). In addition to few women whose membership was elevated to Fellow, we noticed few if any IEEE Milestones that recognize the work including teams with women. We noticed very few Distinguished

11 Access, Inclusion, and Accommodation

293

Fig. 11.3 Letter from Secretary of Commerce, Gina Raimondo

Lecturers who were women. In particular, there were procedures requiring our local section to have a chapter in a society sponsoring the Distinguished Lecturer, which complicated inviting engineers from interesting fields that were not already present in our section. Even though there are growing fields such as Robotics and also

294

K.G. August

Biomedical Engineering, we did not have technical chapters to host Distinguished Lecturers from those fields. In our section, we noticed years had elapsed since a woman had been a keynote speaker at our annual awards meeting. Yet at the same time, there are many programs and individuals who strive to reduce disparity, increase participation, and improve recognition. There are fundamental issues, for example, recognition is based on percentage of the population of Members. Recognition by percentage will result in exactly the same percentage of women receiving recognition as Fellow. Restricting Milestones to an achievement recommended by peers will also limit recognition to the dominant population. A more inclusive method might consider the fact women might have different priorities and express their solutions in different innovations and implementations. Women might have different obligations to care for family members. Networking opportunities might not be as effective for women. Change might require steps outside our traditional processes. Ideas presented by women might not fit into existing categories; an evolving view of patent categories might result in more inclusive inventions by women, minorities, disabled, and other underrepresented people. To address some of these issues, we have initiated a few new activities in our section some of which are depicted in Fig. 11.4 and represented in our websites including (IEEE New Jersey Coast Section History Wiki 2021) https://ethw.org/ IEEE_New_Jersey_Coast_Section_History: • “Celebrating Our History: Inventing Our Future,” to connect history to our current interests • Interviews with Members, for example, “Visits with Victor,” video conversations with Victor B Lawrence, PhD IEEE Life Fellow, Professor Stevens Institute of Technology and Member of New Jersey Inventors Hall of Fame. • Women’s History Month Roundtable Hosted by IEEE History Center Lisa Nocks, PhD and Stevens Institute of Technology • Martin Luther King Day Event, with Victor B Lawrence, PhD IEEE • IEEE Day trip to the Edison Museum • Founding an IEEE Humanitarian Activities Committee Special Interest Group for Humanitarian Technologies (HAC/SIGHT) (Katherine Grace August, PhD and Margaret J Lyons, PE) (IEEE Humanitarian Activities Committee/Special Interest Group for Humanitarian Technology 2021). https://sight.ieee.org/ • Two new IEEE Milestones (1) DIANA Project that introduced Radar Astronomy, and (2) Neutrodyne Circuit by Professor Louis A Hazeltine that introduced radio as affordable mass media (IEEE History Milestones Program 2021). https://ethw. org/Milestones:IEEE_Milestones_Program • Became involved with IEEE Standards Association (SA) Healthcare and Life Sciences, including a new Standard, a workflow for Telehealth Lexicon, and a competition about Remote Patient Monitoring (RPM) Re-Think the Machine (IEEE Standards Association (SA) Healthcare and Life Sciences, and Remote Patient Monitoring (RPM) Re-Think the Machine 2021) https://standards.ieee. org/practices/healthcare-life-sciences/index.html

11 Access, Inclusion, and Accommodation

295

Fig. 11.4 Participants from a selection of activities with IEEE and IEEE NJ Coast Section: History, SIGHT group, and inventing

• Kicked off a new workflow in IEEE SA Dignity Inclusion Identity Trust and Agency (DIITA) to address Transparent Design for Wellbeing (IEEE SA Dignity Inclusion Identity Trust and Agency (DIITA) workflow Transparent Design for Wellbeing 2021) https://standards.ieee.org/industry-connections/diita/index. html • Undertook a humanitarian event “Hear, here! Justice for All” • Created humanitarian projects including IEEE HAC/SIGHT Group projects involving engineers, stakeholders and High School Students in two States: “Do Good Things Justice for All,” and “Do Good Things Health and Justice for All, Health and Medical Device Literacy” • Invited women speakers to the section and to participate in projects • Held inventors’ meetings to network with inventors and encourage new inventors • Developed web content, media, blogs, and publications Importantly, during COVID-19 many of our activities went online. Offering more activities online created far more opportunities for people to participate with fewer complications and permitted us to invite lecturers from a variety of fields without expense. In addition, we were able to engage more volunteers who might not have been able to participate if they had to travel, and more people were able to attend activities with the high schools and courts for our humanitarian activities who might not have been able to participate if they had to travel. Digital activities offered more flexibility and connected us to one another. We were able to invite and feature more women Distinguished Lecturers and Distinguished Visitors and attract a wider audience to our activities.

296

K.G. August

We were and continue to be inspired by many role models including those close to home. From the earliest days of electronic communications and computing, personal inspiration, emergencies, and societal events necessitated advances and shifting focus. For example, the early days of modern communication and humanitarian activities were marked by the great work of Alexander Graham Bell, an immigrant, a teacher of linguistics, and a teacher of the deaf and with the participation and lifelong collaboration of his student and wife, Mabel Gardiner Hubbard Bell and her family; Bell’s invention and commercialization of the telephone and associated business models changed the world. The influence of immigrants in the expansion of science, invention, technology, and industry in the United States is significant. Various studies find immigrants account for a higher rate per capita of United States Patents and Nobel Prizes, hold more STEM degrees, and start firms more than native born (Alexander Graham Bell IEEE Profile 2022; Jones et al. 2021). One can also consider the impressive record of Thomas Alva Edison; he was certainly an accomplished inventor with significant influence (Thomas Alva Edison IEEE History Wiki 2022). Edison was a person of great vision in basic sciences, transformative methods, and industry. The Bells, Edison, and others were devoted not only to a wide range of scientific and technology innovations, for example, the Bells regularly held salons and invested in technologies such as those to advance human flight, but many of them were also engaged extensively in humanitarian interests. Typically, one would believe the inventor could not fully envision the impact of the invention and the additional innovations and industries that may be enabled; but these role models may indeed have envisioned and held ideals vast and far reaching. The Bells, Edison, and peers founded many organizations including the organization that later became IEEE, an international engineering membership nonprofit dedicated to advancing technology for humanity. These early innovators are exceptional role models. Many of their greatest legacies, achievements, and contributions have been and continue to be informed by the lived experience, which can either be their own, through collaborations, or through the experiences of others.

11.3.2 A Closer Look at STEM Lack of inclusion and diversity in all aspects of the engineering professions is a critical and complex problem to investigate, understand, and resolve. Whereas many people choose Science, Technology, Engineering, and Mathematics (STEM) careers for professional and economic opportunities, to advance scientific achievements, to engage in innovation, and/or to implement sustainable solutions that will improve resilience, reduce suffering, and improve quality experience of life for individuals and across society, at all scales, everywhere, STEM programs have not resulted in a change in a variety of outcome measures. We must consider how significant inclusion and diversity is and what role we can play as engineers in advancing the

11 Access, Inclusion, and Accommodation

297

Fig. 11.5 US Census Bureau percentage of women in STEM jobs 1970–2019; STEM median earnings by sex 2019

profession and the associated impact on the design and development of technology that bridges the digital divide and improves lives everywhere. A closer look at statistics from the US Census Bureau in Fig. 11.4 reveals the changes between 1970 when women made up 38% of workers and 8% in STEM fields and 2019 when women made up 48% of workers and 27% in STEM fields. In computers and engineering, women did not make big gains. In 2019 women represented 25% of workers in computer fields. In engineering fields, women numbered 3% in 1970 and 15% in 2019. This is a very small voice in such an influential field. Also of note is the statistically significant disparity in median income between men and women in a number of STEM Occupations even in 2019 (United States Census Bureau census.gov 2021) (Fig. 11.5). The field of engineering and telecommunications engineering lags behind. The United States Department of Commerce United States Patent and Trademark Office (USPTO) SUCCESS report describes a grim level of Patents in target groups. In addition, preconceived notions that Science and Engineering innovations are driven by the young are found to be untrue. Instead domain skills and knowledge are important. Experienced people are able to pivot and direct their advanced skills and efforts to important problems over a lifetime, and peak contributions are typically at older ages. For example, research of United States ventures (2007–2014) by Azoulay et al. demonstrates that the highest growing firms are founded by those in middle age (Azoulay et al. 2020). Research by Jones 2020 illustrates that individuals are in middle age at the time of their most notable discovery for Nobel Winners and Inventors, and that over time (before 1935, 1935–1965, and after 1965), that age at time of discovery is increasing (Jones 2020) (Fig. 11.6). There is a strong connection between STEM subjects, inventions, economic value, and impact on humans. Recently, research connected the inventory of US

298

K.G. August

Fig. 11.6 TL overall US startups/top 1% growth startups; TR age of entrepreneur by coefficient estimate; LL Nobel Prize winners age, great inventors age; LR age at Nobel achievement before 1935, 1935–1965, after 1965

Patents since 1975, cited prior art, the extensive body of scientific articles, and articles that were connected in references and related ideas (Ahmadpoor and Jones 2017). Research shows that 79.7% of scientific articles are part of prior knowledge to a future patent. In economic terms, top value science patents such as biotechnology, artificial intelligence (AI), and novel chemical compounds have an average value of $17.9 million and are double the average market value of those not connected to science (Watzinger et al. 2021). Given the substantial role technology plays in our success as human beings and in all aspects of society, our technology solutions should represent the community of people and also include them in all phases. And all humans should have access to the economic benefits of the fields of science and technology. The COVID-19 global pandemic, recent extreme weather events, and other emergencies have elevated our awareness of the promise of engineering and technology to meet the extraordinary needs of a diverse geography and population. Importantly, we have also become

11 Access, Inclusion, and Accommodation

299

aware of some of the many shortcomings of the currently available infrastructure, policy, and structural influences that attenuate disparity and lead to poor outcomes, not just now but in the future (Katsikopoulos 2021). These problems are on a scope and scale beyond what we traditionally encounter in our everyday pursuits and require significant engagement, intervention, and communication from many stakeholders. To participate in the ecosystem of science and technology requires access to education, jobs, mentors, and more. Many people choose STEM careers to advance scientific achievements, to engage in innovation, and/or to implement sustainable solutions that will improve resilience, reduce suffering, and improve quality experience of life for individuals and across society, at all scales, everywhere. For decades, technology has been designed, developed for specific purposes, and marketed to a specific and selective group of industries and applications, companies, and their customers, government, and research, for example. Infrequently and with some striking examples, technology has been effectively focused or scaled to a benevolent purpose or to serve the wider public. Even in times of emergency, for example, the Tylenol poisoning, solutions are often focused on needs of business. Access to telephone network services provided for a high volume of calls to play recorded information from a network server since there were limits to the number of calls that could be handled by live agents. The Tylenol use case relates not only to a poisoning crisis, but it also represents a crisis communication effort leading to retained value for the brand and consumer confidence. Literature demonstrates many such business-related examples. Literature tends to focus on business-related examples. There is a need for more inclusive and diverse research, methods, and information about how to meet the needs of humans, to reduce digital disparity, and to provide resilience for all people in order to advance technology for humanity.

11.4 Future Innovations for Access: Human-Centered Frameworks Future innovations for access can benefit from human centered design frameworks. These frameworks are useful for interdisciplinary collaborations and can improve integrating stakeholders from the community, gathering stories, use cases, and requirements. Although language and methods might be somewhat different from the field of engineering, the frameworks provide a bridge to improve outcomes and well-being for individuals, stakeholders, and communities, locally and globally. Taking a closer look at the frameworks in Fig. 11.7 improves understanding about how vital access, inclusion, and accommodation are to human-centered design. A framework of Social Determinants of Health (SDoH) provides a useful way for stakeholders to direct efforts to reduce disparity and improve health outcomes for people. For further information, see the World Health Organization website (Centers for Disease Control 2021). Many organizations, companies, governments,

300

K.G. August

Fig. 11.7 World Health Organization Social Determinants of Health Framework; United Nations Sustainable Development Goals

and stakeholders embrace the United Nations Sustainable Development Goals (UNSDGs) framework to direct efforts to a wide range of goals for humanity (United Nations Sustainable Development Goals UN SDG 2021). COVID-19 has placed digital disparity under a new lens. There are a great many effects of the disparity that existed prior to COVID-19. We have witnessed the impact on so many people in a short time. For example, digital access, inclusion, and accommodation are important for humans to benefit from information. It has become apparent that information can serve humans better if it is affordable, understandable information, available in a timely manner, in a language or presentation style or media that is accessible; that matches the literacy, skills, and culture of the human; that reflects the subject matter expertise; that is not complex; and that is transparent in its purpose, sources, interpretation, and more. During COVID-19, timely access to information has been crucial. Even before considering telehealth or school at home, affordable access to information is a fundamental problem to humans dealing with COVID-19.

11.4.1 An Example: Access for Severe Weather Warning Systems Adequate access is important to provide warning and actionable information to individuals and groups, for example, severe weather warning systems. Each geographic region, the nature of the infrastructure and access available, cost of the infrastructure, affordability of end user devices, and ability for the people to use the information are crucial even in ordinary times. How people receive and use the information and timeliness necessary may differ. For example, measures involving air quality or weather require very rapid responses to reduce harm.

11 Access, Inclusion, and Accommodation

301

Without information and the ability to respond to information, people experience disparity. IoT provides a network of networks processed in the cloud to sense, transfer, and process data to information and provides actionable messages to humans. If a person is not considered when the sensors are created and deployed, or when the data is collected and processed, then that individual is experiencing digital disparity. Innovations are needed to increase the availability of IoT, access, reduce costs, and improve usability of information, in culturally and linguistically appropriate systems. Severe weather, hurricane warning systems, and air quality notifications are broadcast in geographic areas at risk. For a hurricane warning, many people will be impacted, and in the community even if only a few receive the message, information will likely be widely shared through traditional methods. Air quality messages typically broadcast on television represent a more complex situation. In 2017 air pollution ranked fifth among global risk factors for mortality, and in 2019, Particulate Matter (PM) is ranked sixth, contributing to four million deaths in 2019 (Health Effects Institute 2019). Air quality is vital to improve heart and pulmonary health, and measurements of environment in the community as well as in the person’s location are both important in understanding the risk. Then through appropriate analysis, messages can be formulated for vulnerable people either by groups such as diagnosis or age or by plan for an individual. Together with healthcare providers, and appropriate healthcare strategies, air quality messages are necessary to effectively warn and inform. One of the most promising aspects for the modern access infrastructure is that pervasive connected public and private sensors such as environmental sensors in the community, public spaces, shopping places, homes, workplaces, vehicles, schools, churches, and venues can serve to significantly improve measurement information. These sensors and information can be provided by individuals or organizations, governments or science, and health services. As long as they are interoperable, and analysis is transparent and explainable, the information can be shared with individuals, organizations, healthcare providers, and the community and can be helpful. Cautions to avoid pitfalls that exploit individuals, impair ethics, or fail to meet needs, such policies can result in more severe impact and cause harm to individuals, groups, and populations. Innovations can be improved through responsible human-centered design. When combined with more personalized sensor data and reports, for example, from an Electronic Health Record (EHR) about a person’s status and history, information, modern data analytics, interpretation, inferences, recommendations, and ability to make better choices can more precisely help each person. Environmental information, for example, can benefit young, elders, people with known heart or pulmonary conditions including heart failure, asthma, and Chronic Obstructive Pulmonary Disease (COPD), and vulnerable populations make better choices about when and where to exercise, work, play, and more (Mirabellia et al. 2020). WHO recommendations include integrating air quality, energy, and climate policies to improve health benefits. Upon closer examination, we find a digital divide that impacts human well-being even in the measurement of environmental weather

302

K.G. August

and Particulate Matter (PM). Providing access, inclusion, and accommodation to information and application solutions holds great potential to improve health; recent studies of the immediate effect of elevated levels of PM indicate increased risk of heart failure, whereas efforts to reduce PM can result in significant healthcare cost savings and reduced mortality.

11.4.2 An Example: Access to Measures of Environmental Temperature Adequate access is important to provide warning and actionable information to individuals and groups, for example, measures of environmental temperature. Indoor and outdoor temperatures and temperatures in the workplace have a significant impact on mortality. High temperatures impact mortality in a short term, sometimes leading to dehydration and a typical cascade of effects. While low temperatures, even moderate changes in temperatures, result in significantly increased mortality over a much longer interval, in some studies demonstrated for up to 21 days and observed in outdoor measures as is illustrated in Fig. 11.8 (Gasparrini et al. 2015). Measuring outdoor temperature, indoor temperature, and body temperature can improve health and well-being. When an individual or caregiver can become informed not only about the temperature but also the influence of temperature on the health of a person through complex means including impact of environment on core temperature, measures can be taken to reduce the risk and improve outcomes and well-being. Human-centered frameworks facilitate the storytelling, use cases,

Fig. 11.8 Temperature impact on mortality

11 Access, Inclusion, and Accommodation

303

discussions, and potential to improve information available for individuals and caregivers to access data, combine it with knowledge, interpret in the context for the individual, and plan for a solution that results in better outcomes for the humans. For example, if a person is accustomed to being able to turn down the thermostat to save money in the winter, they might be at risk as they age. Sensors detecting room temperature and body temperature can be combined with a profile of a frail person, someone with a pre-existing condition or anemia, an aging or person with a disability, an athlete, or a person in an extreme job wearing protective equipment and automatically or recommend increasing the heat to bring the room to a safer, healthier temperature and also provide transparency about measures and algorithms and explanations about how data and the function was employed. And transparency and explainability about how AI/ML and algorithms, inferences, and predictions can improve outcome and well-being.

11.4.3 An Example: Access to Secure Gateways to Manage Power and Communication Adequate access is important to provide warning and actionable information to individuals and groups, for example, to manage energy and communication in the home, residence, or community. If there will be times when power or communication is unavailable such as scheduled brownouts, blackouts, service outages or emergencies, secure AI/ML gateway energy planning and management systems can keep use confidential for the consumer, with assistance of a trusted person or oversight service provider, schedule batteries for automatic charge cycles, negotiate other sources of power or connectivity, or arrange for alternative resources. Energy availability and source impact not only air quality but also health and well-being of vulnerable populations. During COVID-19 people are managing their own health or health of their loved ones at home or at a distance, creating new complex problems in communication and energy management. For a person who is helping a loved one at a distance, as a navigator, an easy-to-use secure gateway communication service that can be operated remotely by a trusted person can be most helpful to reduce loneliness and improve health and resource management (Sociavi 2021). Human-centered innovations can improve options for power and communication management and/or to simplify or improve managing healthcare at home. All services and infrastructure have grown significantly more complex. There are multiple providers and consumers lack negotiating influence. However, policy makers can embrace new models that prioritize classes of users. In particular, a group might include people who manage health in their home. There are opportunities to advocate for policy changes to improve transparency for well-being. For example, if a person is dependent on an oxygen generator and there is a power brown out or power failure, then the person is severely impacted. Infrastructure for power and energy, and the ability to pay is critical. Many sophisticated medical

304

K.G. August

devices already have remote monitoring features. However, there is not yet a comprehensive management strategy to ensure these systems are fail proof. If a person has a cardiac implant with a remote monitor, power and connectivity are essential. Else, alternatives must be planned for the case of emergencies during outages and during service interruptions. Even household appliances play an important role in home healthcare. If power is compromised, being able to keep a refrigerator at an appropriate temperature can impact a person’s health. Many people rely on refrigeration to keep insulin, food, and drink, not to mention air conditioning, air filtration systems, heat, lights, and communication. Secure intelligent network gateways can improve, simplify, and offer new management methods for coordinating and negotiating connectivity, energy, and remote monitoring, through microgrids and services provided into the home, clinic, or community, novel ways to improve access for health and well-being (Yaqu and Ahmad 2020; Yaqub et al. 2021).

11.4.4 Public Health Issues Inspire Action to Engage in Innovation Public health issues inspire action by engineers to collaborate on interdisciplinary teams to consider a broader understanding of the impact of technology to solve such problems and to innovate to realize the promise of technology to improve health and well-being. These human-centered designs provide inspiration to improve the reach of access everywhere, for example, in rural regions, to simplify and reduce costs and improve human factors and usability of digital content for all in the United States and also in regions across the globe. It’s also essential to retain legacy methods and systems and to provide effective redundancy and plan for failures. Transparency and explainability is needed so people can trust information and make decisions. It is important to develop tools that simplify the process and reduce complexity of deploying advanced technologies, to make them development and user friendly. It is important to apply the solutions for the benefit of humans, including use cases and data representative of the population for inclusion and diversity, and to provide a range of methods to use the systems and data that accommodate the needs of the humans. There is a need for interdisciplinary teams to innovate, design, and build systems that effectively coordinate service providers and integrate data from sensors and analysis that is unbiased and inclusive, to develop standards that improve interoperability in the technology and regulatory ecosystems. There are so many challenges to providing access that is inclusive and accommodating. There are significant challenges for modernization. There have been many false starts and promising prototypes never deployed or that faded away after funding ended. There is a risk that following COVID-19, any improvements facilitated by temporary relief might go away.

11 Access, Inclusion, and Accommodation

305

Even in a developed economy, there are challenges for basic solutions. For example, in a recent study, 19 percent of participants representing those residing in certain rural counties in the United States were found to be unable to access air quality information because measures are not taken in all rural counties. In addition, for regions where air quality measures are made and reported, there is a great deal of variability. Individuals may experience higher levels of PM depending on whether they live near a road or factory. Indoor air quality also varies for a number of reasons including smoking, cooking, and/or heating with wood, condition of appliances, cleaning and other chemicals, and more. All of these topics are complex and require significant attention to detail. And modern solutions use AI/ML, data fusion, and other techniques requiring explainability and transparency. These processes are vulnerable to sampling, scale, scope, diversity, and inclusion. Globally, the situation is much more complex. In the example of PM, in certain urban cities with populations above ten million particles are in concentrations of 100–300 micrograms/meter cubed compared with median concentrations of 15 in the US studies. Policies and data available around the globe vary which leaves many more vulnerable. Innovations are needed to measure, analyze, and present information and for tools to develop solutions. The modern IoT network promises to have a great impact in this area of public health, to reduce the impact of disparity for better health and well-being outcomes at a variety of scales: for individuals and communities, locally and globally (Shah et al. 2013). Innovation can play a significant role in doing good things for individuals, groups, and populations, locally and globally. Inclusion and diversity show promise to improve potential for the future of access, for well-being, and for advancing technology for humanity.

11.5 Access Through Accommodation: A Humanitarian Point of View Digital connectivity or access and representation of the population or inclusion are a critical part of the new view of infrastructure for resilience and well-being. As with any infrastructure, for example, sidewalks or buildings, accommodation is needed for people to fully benefit from digital access or many will be left behind. Again, there is great diversity in abilities, lifestyles, and needs and a large population with abilities, preferences, and those with disabling conditions (World Health Organization 2021). Without change, people who need accommodation will continue to experience a high burden of costs in many aspects of life, for example, access to the economy, information, education, healthcare, social connections, and more. In addition to direct costs of accommodation systems, there are more subtle costs faced by those who require accommodation. There is often a need to self-identify; people have to identify as being in need of accommodation. Sometimes these needs are challenged. There are limited resources. Reservations for accommodations are complex. In many situations, for example, reserving hearing

306

K.G. August

assistance is required long before attending a public event, attending a school, theater, church, or before attending a court case. There is a tendency for the best accommodations to require visits to the doctor or specially licensed professionals, to obtain prescriptions and process forms and get approvals by insurance, Medicare or Medicaid, etc. There are additional accommodations for communication access that require applications to state programs or agencies for help. There are often limited resources and long delays. And systems provided are separate from mainstream. Although many are obliged by US Americans with Disabilities Act (ADA), it is clearly not sufficient for a sustainable solution (U.S. Department of Labor Americans with Disabilities Act (ADA) 2021). Although ADA requires accommodation, there is always a reason to avoid the expenditure. ADA allows people to determine if the cost of accommodation is too high. Many employers, businesses, public places, professionals, healthcare providers, courts, and more fail to recognize why accommodation is needed. They might not be inspired to provide accommodation. Individuals requiring accommodation must take action with agencies, courts, and the like, engaging in an adversarial situation. They must incur the burden of costs and delays. Then when those accommodation systems are provided, they are separate and not necessarily in sync with those used by the general population, with their peers, or with their family and friends. These solutions are not practical, scalable, they are not widely available. They add a layer of impediment in addition to an already difficult situation. There are many people with needs globally; populations are depicted in Fig. 11.9 (IASC Guildelines, Inclusion of persons with disabilities in humanitarian action 2019).

Fig. 11.9 Global population of persons with disabilities IASC guidelines, 2019, inclusion of persons with disabilities in humanitarian action

11 Access, Inclusion, and Accommodation

307

For those without accommodation, there is a significant downside. Although in the past, participation in the digital world was taken to be optional and or almost entirely voluntary, or closely bound with one’s identity in relationship with the economic world – a job, or a social status – that is truly no longer the case. There is no longer a reasonable way to survive without access, inclusion, and accommodation. It is timely for crisis communication theory to evolve. The COVID19 emergency represented a gradually evolving complex situation. Decisions are made without complete information. Many adjustments were made that might be temporary or longer term. For those without direct access to information, lack of accommodation is a matter of life or death.

11.5.1 COVID-19 Global Pandemic Emergency: A Pivot Point Illustrating Need for Access Through Accommodation From the very first days of the COVID-19 pandemic emergency, the mainstream communication infrastructure was called upon to keep people informed. In a matter of approximately 6 weeks at the beginning of the COVID-19 pandemic emergency, nearly all services, employment, healthcare, information, education, and more went online in the United States. People needing accommodation but remaining at home during lockdowns and social distancing were not able to gain telecommunications and digital access in public spaces, workplaces, education centers, etc. Many did not have a system of accommodation in their homes, group homes, and communities. People were isolated and cut off from traditional support systems and accommodations in many places that included being cut off from family, friends, and people who had previously provided casual accommodation. Yet existing mainstream systems fail to meet needs for a significant number of people; a recent study illustrates increased risk and death across two waves of COVID-19 among disabled men and women across age groups in England depicted in Fig. 11.10 (Bosworth et al. 2021). Elderly or disabled people in the United States on Social Security and Medicare were faced with turning to online services because no government and few medical offices were open; call centers were scheduling months into the future. It took time for service providers to establish ordinary businesses for providers. In addition, regulatory constraints had to be adjusted. Regulations for medical licenses typically do not allow providers to cross jurisdictions, and stakeholders were faced with complex situations. Rapidly, changes and waivers were authorized. For example, telemedicine appointments, voice without video visits, and home monitoring were authorized for reimbursement due to declaration of emergency. For the first time, many more people had access to healthcare including behavioral healthcare, right where they live. Great insight has been observed and shared by stakeholders including people who cannot afford to travel can receive care by telemedicine, people who are unable to use or to afford technology can receive care by plain

308

K.G. August

Fig. 11.10 Age-adjusted cumulative COVID-19 mortality Jan 24, 2020 to Feb 28, 2021 by disability status and sex in England

telephones, and students away at college were able to receive care from their traditional healthcare providers including for behavioral health across state lines. People can receive help for significant health problems in a more timely manner. People living in rural areas can receive substance counseling, which was not available before. These are only a few examples of how the use cases in health and well-being allow stakeholders and technologists to share opportunities to understand underlying needs not only from a technology point of view but also from system design, accommodation, human factors, regulatory, and various other aspects. These are part of the lived experience. When the crisis ends, people will be faced with loss of many of these “temporary” advantages. During the COVID-19 pandemic, people reliant on others, which includes those with disabling conditions, or pre-existing conditions, those living in group homes or congregant settings, were more significantly impacted. Individuals rely on others to obtain, interpret, and present information or accommodations. At the interface of communication and information systems, there are significant opportunities for inclusion and accommodation of the individuals and their caregivers, for technology improvements, to reduce suffering and death, and to improve well-being. Systems can be designed up front and iterated to accommodate abilities, needs, and preferences of any and each individual, class, and group, in any context, time, and place. There are barriers to including people with disabling conditions in science and engineering teams which reduces the ability for designers to adequately understand, represent, and fill needs of the users. Such systems can be designed to provide plug and play solutions, for example, for screen readers for the blind, to provide augmentative communication for speech assistance and hearing loss, to supplement information, or to provide programming to improve mobility experience. However, at present the burden remains upon the individual to afford

11 Access, Inclusion, and Accommodation

309

accommodation. Some solutions are very expensive and involve specialized technology, with long complicated administrative, medical, or insurance processes, with less flexibility than general purpose solutions. People have to identify as disabled or in need, apply for assistance, and wait for approval. During COVID-19, this approach to access through accommodation fell apart in many ways. Interpreters, both formal and informal, were not permitted due to lockdown, social distancing, and precautions including medical, education, workplace, public transportation, etc. Some employers accommodate people at the office but not at home leaving some unable to work remotely. Legislation such as the US Americans with Disabilities Act (ADA) is clearly not sufficient for a sustainable solution (U.S. Department of Labor Americans with Disabilities Act (ADA) 2021). Even though there has been significant progress in accessibility and usability, there are “continued gaps” as mentioned in a recent 2020 Biennial Report to Congress (Rotarou et al. 2021). Many people with disabling conditions rely on others for information, live in groups, and suffer economic inequities. Technology can theoretically enable access to all people in all contexts. However, often and especially during COVID-19, individuals had to deal with many situations without adequate accessibility support or solutions. Early in the pandemic, people with disabling conditions suffered, became sick, and died at a higher rate due to exacerbation of existing vulnerabilities. Many shocking news stories and statistics reflect this ongoing disparity (USPTO.gov 2021). Access through accommodation can be described from several aspects, such as (i) at the interface of the system and human to use or operate the capabilities, what could be referred to as User Interaction and Experience (UIX), ensuring the UIX is adequate for every individual independent of their abilities through accommodations; (ii) with access to infrastructure such as adequate connectivity to network communications, Internet, and devices; and, (iii) transparent interoperability among systems, for example, medical devices and electronic records such as medical, financial, and others. COVID-19 challenged all aspects and exposed shortcomings of current infrastructure, systems, policies, features, and functions of technology solutions and models. COVID-19 has also unmasked the fact that lack of digital access is in itself, a form of disability often described as digital disparity. Sources and reasons for lack of access are complex. Accommodation to access is a multidimensional humanitarian problem. Innovation is important to advance solutions; to advance understanding the interplay of access to technology, ability to understand, use and benefit from information, and needs of a complex society and resources available so that individuals are able to experience agency; and to take action for well-being.

11.5.1.1

Secure Easy to Use Communication for Protected Persons

Access through accommodation can be simplified with secure gateways, enabling trusted persons to improve experience of technology solutions for protected persons, to reduce complexity, to provide a means for social engagement, and to support their loved ones even from afar. An excellent engineering solution invented and

310

K.G. August

developed by Paula Muller, PhD of SociAvi, improves and simplifies secure communication between people with special needs, protected persons, with their family caregivers and healthcare providers. In the invention, “Communication System for Use with Protected Persons,” US 10,848,711, individuals are connected with their loved ones using a secure and trusted platform and appropriate interface features that improve success and experience for the user (Sociavi 2021; Rotarou et al. 2021; InSIGHT Biography of a SIGHT Volunteer – Paula Muller 2021). The communication system provides access through accommodation for blended and virtual spaces including telemedicine and social interaction, trusted remote operation of features, entertainment, schedules, and more. Privately, objective measures are gathered using wearable devices and the cloud connected system processes (Sociavi 2021). In this scenario, an adult is living at home, or at a senior living setting such as a skilled nursing facility, an assisted living facility, or a memory care facility, and the person depends on several caregivers and care providers. Many aging adults or people dealing with disabling conditions have multiple chronic conditions and need the attention of several care providers. The caregiving responsibilities may be shared by several siblings or family members or other trusted persons. Other individuals like financial advisors, elder law attorneys, money managers, government insurers like Medicare in the United States, private insurers, and others have access to personal and private information of this individual. Having immediate access to the relevant information to make adequate decisions for the care and well-being of this person is critical, but it is crucial that any access is transparent and trustworthy and ensures the privacy of the individual. Such a solution is widely applicable. Paula Muller, PhD, serves as a role model, bringing important perspective to solutions for access, inclusion, and accommodation.

11.5.1.2

Special Interfaces for the Blind: A New Standard

Although there are many employers, schools, and public venues that provide accommodation for the blind according to provisions of the ADA, solutions might be expensive. There might be no incentive for the same institutions to provide accommodation to the blind while they are in their home. This limits the freedom of those individuals needing accommodation. During COVID-19, some people who were accommodated in the workplace or school were unable to be accommodated remotely, to work at home, to learn from home, and to shop or interact with others. Some individuals and companies stepped up to the challenge; Amanda Deol of Addteq, for example, provided a screen reader for the blind, a platform typically only available in the workplace for use in the home. In this way, Amanda Deol enabled people to continue working from home during the pandemic. Now her efforts are in extending functionality. She has introduced an image-based reader to extend content that can be read to those with blindness and low vision demonstrated in the following video https://youtu.be/XIEmm0KnhGc (Screen reader images for the blind n.d.). For more information about the screen reader, see the website

11 Access, Inclusion, and Accommodation

311

https://devpost.com/software/unstoppable-auto-caption (Screen reader for the blind n.d.). Amanda Deol also introduced a new IEEE Standard P2998 Special Interfaces for the Blind, to facilitate interoperable and simplified development of systems with screen reader features for the blind. Standards can provide explicit direction, the “how to” so that many more companies can create products that are interoperable and accommodate special needs. Human-centered innovation can greatly improve the impact technology can bring to inclusion, diversity, and accommodation to digital access for communication and for many individuals, families, communities, and society. Once a technology solution is established, standards can improve the potential to share simplified solutions that can be scaled, shared, and deployed in an interoperable ecosystem for well-being. In this way, we can do good things to promote wide access to the economy, education, healthcare, social interaction, creativity, and more.

11.5.1.3

Do Good Things Justice for All to Improve Speech Understanding

As the network of networks and digital access becomes more ubiquitous, intelligent, and sophisticated, connected sensors, features, and services can be applied to solutions in a standard way to improve accommodation to many more people with low cost, mainstream technologies. Failure to accommodate results in significant disparity in all aspects of life. In our recent IEEE Humanitarian Activities Committee Special Interest Group for Humanitarian Technologies (HAC/SIGHT) Project “Do Good Things Justice for All,” we investigated the effect of hearing loss on people in the Justice system (IEEE Humanitarian Activities Committee/Special Interest Group for Humanitarian Technology 2021). Costs to accommodate can be reduced by using mainstream technologies, for example, smart phones and free closed captioning, to improve speech understanding in many situations. In our project, an interdisciplinary team including stakeholders in the justice system, judges, attorneys, county clerk, people with hearing loss, specialists, students, a mock trial team, and engineers collaborated and learned more about the Americans with Disabilities Act (ADA) processes in action. The team learned that many people are unaware and do not appreciate that they must provide expensive accommodation to students, customers, patients, employees, and the public. They sometimes reported the belief that deaf people are “faking it.” They believe the deaf people can hear because they have “hearing aids,” or that they hear 1 minute, and then they don’t hear. The nature of hearing loss and speech understanding is complex. Accommodating people can lead to greater independence for individuals. Engineers are trained and can lend insights to hearing loss and modern methods to improve speech understanding. One of the stakeholders who is an ADA Mediator suggested that if people could “hear” what it’s like to have a hearing loss while trying to understand important speech of others, they might be more able to understand the situation.

312

K.G. August

Fig. 11.11 IEEE HAC/SIGHT Project “Justice for All” experiential learning system online

Our project concentrated on creating a simulator so people can listen to speech filtered as some of our team members hear. Figure 11.11 describes methods to access the Experiential Learning System Simulator. Not every person with hearing loss experience is the same. Not all hearing aids fully correct a hearing loss. People frequently require additional accommodation in a variety of situations. For example, in a theater, classroom, conference room, at work, or when using a telephone, hearing aids will require additional expensive equipment to provide assistance to the listener. If the person has hearing in only one ear, then stereo hearing is lost. Detecting which direction from which a sound is coming becomes complicated. Many hearing aids can only reduce background sounds or focus a sound enhancement algorithm when the listener faces the source of the sound and when the listener has two hearing aids working together to engage the features. These are merely a few situations where engineers can serve to improve the human experience through mainstream accommodation. We also demonstrated interoperability of several mainstream technologies depicted in Fig. 11.12 to provide low-cost simple solutions, connecting small displays such as smartphones, laptops, tablets, and the like for captioning. Many people were faced with unexpected challenges during COVID-19. Social distancing, plastic barriers, and masks reduced speech understanding even for people who were unaware of hearing loss. Illustrated in Fig. 11.12, smartphone captioning can improve understanding. During COVID-19 more people were attending meetings online. With captioning on smart phones and in online meetings, people became aware of the general benefit and improved speech understanding through captioning. Even bilingual people

11 Access, Inclusion, and Accommodation

313

Fig. 11.12 Captioning is an easy way to improve speech understanding

benefit from seeing the text. People learn to adapt and speak more slowly to improve the performance of the captioning. Still, many providers charge significant fees for automatic captioning placing financial burdens on speech understanding. We endorse and promote new efforts to advance technologies for humanity that extend beyond requiring accommodation, but that facilitate low cost, easy to use, easy to deploy accommodation that can be made available any time, any place. In our project, we employed available free technologies illustrated in Fig. 11.13 including YouTube captioning used to generate speech to text files, those that are already available in PowerPoint, and for special circumstances, with additional editing. Since digital access is so essential, intelligent cloud-based systems can be employed to provide appropriate accommodation to a widening range of human needs that can be expanded over time. With a system fostering standards, methods and solutions can be interoperable. Special value development can then be applied widely, reducing costs for everyone. Efforts must be undertaken to investigate needs and lifestyles and ensure the blended environment, digital and physical, that works for the majority is also fit for duty and accommodates those with differing abilities. Enablers improve resilience, safety, and protection and can reduce barriers and risk Fig. 11.14 (IEEE SA Digital Inclusion Identity Trust Agency (DIITA) 2021). For example, digital systems should be designed to be fit for duty for those with hearing loss, vision loss, mobility limitations, dexterity, temporary or long-term health conditions, cultural and linguistic differences, preferred learning modalities, subject matter expertise, cognitive abilities, memory issues, and more. Without planning up front, and enabling accommodation for diverse populations of users, and without built in, for example, plug and play accommodation, systems

314

K.G. August

Fig. 11.13 How to create custom closed caption to improve speech understanding

Fig. 11.14 Barriers and enablers to inclusion of persons with disabilities in humanitarian action

fail to be sustainable. Aftermarket costs to provide accommodation are much more expensive. During the COVID-19, emergency lockdowns prevented special accommodation for people with disabling conditions such as professional workplace screen readers for the blind, informal teleconferencing applications, sign language interpreters, and even informal assistance from family or friends. Many people were unable to work or learn from home. Accommodation provided by employers in the workplace and schools at campus and even in medical settings was no longer available. Social distancing created even more problems while group living situations placed vulnerable people at risk. During the COVID-19 emergency, several significant and swift changes were made in infrastructure; it is important for engineers to encourage local and national

11 Access, Inclusion, and Accommodation

315

governments, regulators, organizations, and companies to move swiftly to improve infrastructure that will continue to improve reliable connectivity and sustain the new models to benefit all. Government policies and regulations must be changed to continue to make progress for improving access, inclusion, and accommodation, even following the emergency. Such policy changes will help to address the big issues, direct resources to the situation, reduce digital disparity, close the gap, and improve resilience and wellness. Regulatory methods are no longer effectively providing modern sustainable systems of accommodation. There is a widening gap between the systems for the general population and those provided to accommodate. This situation inspires fundamental changes to technology infrastructure to remove impediments to accommodation. Regulatory approaches attenuate dependence upon processes and others, for example, interpreters, or expensive dedicated technologies like dedicated caption phones, with limited features and services. For example, at times during COVID-19 and in other emergencies, ADA fails to provide caregivers access to accompany their loved ones during medical procedures. It is no longer practical to provide accommodation exclusively through other humans or through expensive silos of separate technologies. Many people are overlooked and fall between the cracks leading to severe consequences. Along with a history of delays, costs, and complicated processes associated with such accommodations, it is time for engineers to step in and collaborate on interdisciplinary teams, to develop standards, to facilitate innovative interoperable systems, and to develop a vision of the future with all the potential of advancing technology for humanity.

11.5.1.4

Transparent Design for Well-Being: An IEEE DIITA Workflow

The future holds great potential for innovations consistent with humanitarian frameworks, in an ecosystem that facilitates built in accommodation for digital access, for access through accommodation, and for well-being. The IEEE SA Dignity Inclusion Identity Trust and Agency (DIITA) program provided resources for the team to explore and incubate solutions that will improve more generalizable solutions to the growing complexity around the fragmented environment of wellbeing (IEEE SA Digital Inclusion Identity Trust Agency (DIITA) 2021). The team recognized an overarching imperative to reduce complexity, to simplify the experience for people when they rely on technology. Without Transparent Design for Well-being and Accessible technology, humans, their families, and communities can suffer the most extreme consequences, interfering with life and well-being. We noticed during COVID-19, people were called upon to take on new responsibilities, learn more about healthcare and medical devices, and function outside their experience. People are expected to care for themselves or their loved ones in their home, using telemedicine, or remotely posing new challenges to literacy for people who have all levels of education and experience, many languages, and a range of literacy skills. Even healthcare professionals were called upon to engage

316

K.G. August

in COVID-19 duty, care for people using telemedicine, or care for patients with complex healthcare in the context of a global pandemic. In some examples, a simple medical device might have several user guides, each one for a given stakeholder. In addition to the user guide, there is reference material, evidence-based medical practices, and more. To apply the medical device in a given context, information about the human is needed. Sometimes the information is available in the EHR, or from the patient. The situation is complex, even for professional healthcare providers. Errors using devices can cause harm or fail to recognize the need of a patient. An example is the pulse oximeter device used in the home to monitor people to determine when they might require a higher level of care and also used in hospitals to monitor patient condition during the pandemic. Adding to complications, the Food and Drug Administration (FDA) issues notices periodically that are meant to inform everyone about issues. Keeping up with notices is challenging, especially when the healthcare system is stressed by a pandemic. One can see in Fig. 11.15 from the article, Racial Bias in Pulse Oximetry Measurement, that the FDA describes in the notice that begins as follows (Sjoding et al. 2020; Food and Drug Administration Safety Communications Pulse Oximeter Accuracy Limitations 2022): “Date Issued: February 19, 2021 The Coronavirus Disease 2019 (COVID-19) pandemic has caused an increase in the use of pulse oximeters, and a recent report (Sjoding et al. External Link Disclaimer) suggests that the devices may be less accurate in people with dark skin pigmentation. The U.S. Food and Drug Administration (FDA) is informing patients and healthcare providers that although pulse oximetry is useful for estimating blood oxygen levels, pulse oximeters have limitations and a risk of inaccuracy under certain circumstances that should be considered. Patients with conditions such as COVID-19 who monitor their condition at home should pay attention to all signs and symptoms of their condition and communicate any concerns to their healthcare provider.” With new data illustrated in Fig. 11.15 from real patients monitored during the pandemic, the pulse oximetry device is demonstrated to be not best fit for the purpose of monitoring every patient; recommendations include considering preexisting conditions, and other symptoms but not using the pulse oximeter as the only information source to determine low circulating oxygen or motivate need for higher level of care. The team recognized that with modern digital access technologies, transparent information can be provided within the context. Systems can bring important information to the attention of stakeholders. Simplifying presentation of the information might improve well-being for people caring for themselves or others. Information can support learning for professionals and also for patients and their families. Furthermore, if information can be presented in a Culturally and Linguistically Appropriate System, then people will be able to participate in their own healthcare or in the care of their loved ones.

11 Access, Inclusion, and Accommodation

317

Fig. 11.15 Racial bias in pulse oximetry measurement White patients/Black patients, pulse oximetry measures and blood oxygen measures compared

Without Transparent Design for Well-being and Accommodation technology through interoperability and human-centered design, for example, plug and play

318

K.G. August

Fig. 11.16 Do Good Things Health and Medical Device Literacy project

solutions for vision, hearing, mobility, cognitive, and other challenges, humans, their families, and communities can suffer the most extreme consequences, interfering with life and well-being. We have embarked upon this new effort incorporating an IEEE HAC/SIGHT Project, Do Good Things “Health and Medical Device Literacy” that will serve to provide an opportunity to engage stakeholders, healthcare professionals, community members, students, teachers, and engineers. There is a high-level description of the project, with the Community of Interest (CoI) by Language Needs, by Levels, reading skill, health skills, medical subject matter fluency, medical device knowledge, etc.; by User Group, patient, caregiver, first responder, healthcare assistant, healthcare professional, trainee, researcher, device manufacturer, standards, policy makers, etc.; and by Profiles, standardized, user specific, artificial intelligence machine learning (AI/ML), etc. Some project methods, crowdsourcing, training, certifying, reconfiguring, standards, recommended methods, etc., and identified additional needs, reduce bias, improve fairness in technology, research, and publications in culture, language, skin color, race, age, pre-existing conditions, etc., are depicted in Fig. 11.16. The early project will focus on devices elevated in need since COVID-19, for example, thermometer, blood pressure cuff, pulse oximeter, smart scale, and glucose meter. Through Transparent Design for Well-being, it is our intention to facilitate a vision of the modernization and advancement of technology for humanity that considers diversity and inclusion for all human beings. These intentions are synergistic with our IEEE Humanitarian Activities Committee (HAC) /Special Interest Group for Humanitarian Technologies (SIGHT) Group efforts, which will provide one of the pipelines into the Workflow and align with a variety of local and global humanitarian policies and efforts including but not limited to United

11 Access, Inclusion, and Accommodation

319

Nations Sustainable Development Goals, Culturally and Linguistically Appropriate Systems – Health and Human Services (CLAS-HHS), Social Determinants of Health (SDoH), Americans with Disabilities Act (ADA), etc. (Centers for Disease Control 2021; United Nations Sustainable Development Goals UN SDG 2021; IEEE Humanitarian Activities Committee/Special Interest Group for Humanitarian Technology 2021; U.S. Department of Labor Americans with Disabilities Act (ADA) 2021; Think Cultural Health - Standards - Culturally and Linguistically Appropriate Systems - Health and Human Services 2022). The team initially identified four key drivers for the workflow which are: Health and medical device literacy; Culturally and Linguistically Appropriate System (CLAS-HHS) design for dignity and identity; increase diversity and representation/reduce bias and address variability in research literature and projects; and increase diversity and representation in technology and AI ML (e.g., gender, skin color, age, pre-existing conditions, etc.). The IEEE DIITA workflow Transparent Design for Well-being, leveraging the IEEE HAC/SIGHT Group project “Do Good Things Health and Medical Device Literacy,” promotes interdisciplinary collaboration to bring inclusion and diversity to people through technology solutions, to elevate awareness and engage stakeholders locally and globally. Transparent Design for Well-being aims to bring local humanitarian collaborations to the global conversation to extend inclusion beyond local to the national and global community. The team aims to create models, methods, and standards that are the foundation of practical solutions that reduce complexity, improve transparency, and elevate literacy for all people, in other words, to advance technology for humanity. At this time, it is important for all of us to do good things. To participate in lifelong learning. To engage an inclusive diverse team, find and promote role models, promote innovation and invention, to promote resilience for the diversity of people, and include and accommodate every person everywhere. To record and share our own ideas – to invent solutions for the future. To eliminate the digital divide.

11.6 Improving Access in the United States In many areas of the United States, digital access is insufficient for the needs of society. In both urban and rural areas, the communication and medical infrastructure is insufficient, costs of infrastructure and services are high, and funding models have not caught up with present needs of society. Alternatives involve driving distances, or going without. Delays in healthcare and prevention come with consequences. Many parts of the United States lack access to high-speed internet, or even mobile telephone connectivity (Lai and Widmar 2021b). Although the Federal Communications Commission (FCC) has programs to modernize connectivity or access, methods of tracking underserved regions remain controversial and lead to inadequate funding for improving infrastructure, with approximately 14.5 million

320

K.G. August

Fig. 11.17 FCC indicates ~14.5 M lack broadband; Microsoft indicates ~120.4 M do not use the Internet at broadband speeds

underserved people in America according to the FCC, whereas Microsoft reports broadband utilization indicating approximately 120.4 million people do not use broadband internet as depicted in Fig. 11.17 (Kahan and Lavista Ferres 2020). Disparity in quality of broadband and quality of internet access is seen by race in the US groups as illustrated in Fig. 11.18. A great many innovative systems and solutions are facilitated through modern communications digital access, while business models can fail and the promise of intelligent solutions are not always widely appreciated. In many examples, the goals have been limited or legislated to specific stakeholders, or designed for brand image purposes, and the like. For each US dollar invested in Research and Development (R&D) which is about 2.8% of Gross Domestic Product (GDP), economic impact is five or ten dollars, representing a significant return on public investment. A useful example is the COVID-19-related Operation Warp Speed undertaken to bring novel vaccination technology to battle the global pandemic at a cost of about 1% of the expenditure on the pandemic, or $25 billion. To clarify why this is an important measure of reference, if the public investment resulted in ending the pandemic 1 day sooner, the investment paid for itself (Azoulay and Jones 2020). It’s challenging to explain or uncover all the potential benefits for a given science or technology investment. Inclusion and diversity at the table hold potential for a different perspective, a wider discussion, and more humanitarian interests. For a helpful discussion on the untapped resources of public policy in science and innovation, see Jones (2020) (Jones and Summers 2020) for the National Strategy

11 Access, Inclusion, and Accommodation

321

Fig. 11.18 dbDIG quality of Internet access versus quality of broadband by race in the United States

for the COVID-19 Response and Pandemic http://www.whitehouse.gov (National Strategy for the COVID-19 Response and Pandemic n.d.).

11.6.1 Resilience During the recent COVID-19 global pandemic emergency, many stakeholders have been forced to examine access, inclusion, and accommodation from a new resilience perspective. Society turned to technology to solve many problems since the early twentieth century; during the pandemic, most solutions rely on technology. There is a need for technology to address the needs of a wider range of stakeholders and associated business cases not only corporations but also individuals, their caregivers, diverse members of the communities, and to consider many more use cases. Some of the pandemic-related solutions are temporary and/or inadequate. Digital disparity has been heightened. People became victims not only of the pandemic but of a lack of access to technology to work, for education, for healthcare, social connections, shopping, to develop new technology skills, and more. Disparities in skills further divided social influences. Providing affordable broadband and easy to use associated devices as infrastructure for the public good and to alleviate the widening disparity is essential and not optional.

322

K.G. August

The field of digital access is rapidly expanding with a need for creative new business models, relationships, and innovative products. Many solutions that would have taken 10 or more years to evolve were adopted practically overnight for example, remote work, distance learning, and telehealth. In addition, technology skills are needed for everybody to do just about any job, communicate, shop, or enjoy the company of family and friends. Technologies change, and therefore required skills will change over time. Upskilling and education should be more available to everyone over a lifetime. Education should be available on the job, at home, in various creative models. Information should be open access. In a society that is clearly reliant on technology for the most important problems, technology, information, and literacy are needed for everyone. Paywalls, high cost of advanced education, and barriers to employment impact resilience. Without ongoing affordable education available to everyone, large segments of the population will suffer disparity. More people will have less access to work and the economy resulting in significant loss of resilience for the entire community. It is as important to avoid pitfalls of technology solutions and technology dependence. For example, overestimating the potential for technologies, and poor technology stewardship. There are several examples including failure of the company Theranos, with claims to revolutionize blood tests by using a pin prick rather than a blood draw on a simple small device and demonstrating “power is knowledge” (Carreyrou 2018; Foucault 1995). Through the Internet of Things (IoT), numerous devices produce massive amounts of data of all types supporting novel conceptual and physical embodiments such as smart buildings, medical monitoring, novel devices, functions, and solutions. These access and data innovations represent a growing paradigm shift and associated ecosystem that requires new skills for everyone. The high technology ecosystem is likely to require lifelong learning, too. Data once contained within a given device or closely bounded set of devices is now more routinely shared. Data is used in multiple places facilitated through one or many service providers, or owners of data, with processes that transform the data, software as a service, employing analysis for various purposes and generating new knowledge and recommendations. In addition to merely generating reports, or being used for decisions known to the data owners at the time, consent is given; now these connections enable repurposing and combining data and new functions that include physical entities and control of devices. New paradigms extend to pervasive presence, repurposing data, predicting and inferential algorithms, forecasting, and more. A new conceptualization of complementary tools, education, devices, human interface solutions, standards, cybersecurity, safety, skills training, laws, regulations, public policies is needed. Some of these issues have been brought to public attention and recognized as essential during COVID-19 to build back better. There is an important role for inclusion and diversity in collaborative interdisciplinary teams to improve understanding about how people live, think about things, solve problems, to enable needed access, inclusion, and accommodation for resilience.

11 Access, Inclusion, and Accommodation

323

11.6.2 Modernization: An Example from Telehealth To emphasize how critical the need is for modernization of access infrastructure and associated policies even in the single topic of telehealth, testimony was presented to Congress in October 2021 (Commerce.senate.gov 2021). High-tech corporations have benefited significantly from advances and growth in technology and exponential growth of the digital economy. Analysis by Deutsche Bank suggests potential for the industry itself to fund more comprehensive access to broadband, devices, mentoring, education, and the like to increase income potential and to reduce disparities. Blacks and Hispanics lag 10 years behind Whites in broadband. Once the system is available, all the public service solutions are enabled. The warning further indicates if left unattended, growth of the digital economy will leave many minorities, for example, 76% of Blacks and 62% of Hispanics, with little or no access to 86% of jobs in the United States in one generation, by 2045, a harsh future in addition to the economic disparity of the COVID-19 environment (Walia and Ravindran 2020; Lai and Widmar 2021a). To invent promising new technologies for the future, we must advocate for inclusion and diversity. Embrace a shift to a new paradigm to create technology in a model that will meet the needs for all people. Modernization of digital access including extending availability of IoT sensors and connectivity solutions will likely improve data available to determine environmental and other influences on individuals, families, communities, and populations, directly and by inference. Use cases for digital access can be transformative for health and well-being. In our vision of Services and Access from the era of 1999, teams were limited to consider business cases with explicit benefits for companies. For example, shelf tags and inventory systems were directed to benefit inventory and supply chain management. Digital media encoding was restricted to business cases for advertisers and proof of performance, for example. Inventory management was limited to describing transactions, dealing with payments, demand of goods and services, and the like. Today we can see how important access is, and how vital all manner of real time data is to society. Twenty years ago, we might have been discussing wireless shelf tags to track cold medicine sales and to predict demand. With that information, the manufacturer could predict needs for more product and the manufacturer could plan to distribute more product to replenish shelves and improve share of sales, possibly ahead of competitors. New business to business information reveals details of the complex opioid crisis where supply chains are smoothly coordinated to optimize earnings. In some cases, however, accountability is not as transparent. During COVID-19 global emergency, people are more aware of how detailed data typically kept by diverse providers when shared in a timely manner can improve surveillance tracking and public health response. Prior to COVID-19, many details about morbidity and mortality were not reported regularly; but when certain information is made available to researchers and others, then more timely knowledge can be made available. It’s important to identify data that is more generally useful and applicable

324

K.G. August

Fig. 11.19 WHO rapid mortality surveillance system

and facilitate policies, tools, standards and ecosystems to share. To better address inclusion and diversity, timely transparency of how systems are functioning will be important to reduce negative effects of bias and variance. Timeliness of information enabled effective management of public health, and new models were employed to accommodate needs of many as illustrated in Fig. 11.19 (Revealing the toll of COVID-19 2021). We can envision a new ecosystem where a variety of sources of data, for example, from sensors, IoT, and health-related data can be more widely yet securely accessed to improve information for individuals, groups, communities, and countries. Digital access to data improves potential not only for the benefit of individuals but also for the benefit of others through inference. Digital access to data improves prediction of who else might be at risk, for example, in COVID-19 contact tracing. Similarly, tracking supply chains in an ongoing way, with analysis and warning detection, can improve surveillance for the opioid emergency. Technology advances including innovations in telecommunication hold great potential to transform and benefit sustainability and resilience and outcomes for individuals and society; there is an urgent need to incorporate a multidisciplinary approach, to include more technologies, to transform policies, and to address the

11 Access, Inclusion, and Accommodation

325

needs of the wider groups of stakeholders including the inclusive diverse needs of humans and the communities (Olsson et al. 2014). For transformative systems it’s important to incorporate global-minded goals consistent with humanitarian frameworks including United Nations Sustainable Development Goals (UN-SDGs), with standards from other fields, for example, World Health Organization – Social Determinants of Health (WHO SDoH); Culturally and Linguistically Appropriate Systems, from Health and Human Services (CLAS-HHS); IEEE Standards; and others (Centers for Disease Control 2021; United Nations Sustainable Development Goals UN SDG 2021; IEEE Standards Association (SA) Healthcare and Life Sciences, and Remote Patient Monitoring (RPM) Re-Think the Machine 2021; IEEE SA Dignity Inclusion Identity Trust and Agency (DIITA) workflow Transparent Design for Wellbeing 2021; Think Cultural Health - Standards Culturally and Linguistically Appropriate Systems - Health and Human Services 2022). Systems and requirements are often modeled through use cases and are often communicated through storytelling about the lived experiences. These may be unfamiliar to engineers and others. There are many materials being developed to improve understanding of these goals and how they apply to communicating with stakeholders and working together with others to advance technology for humanity. Some advocates in crisis communication theory call to bridge commercial interests and reputational interests of organizations and the more private interests of those directly affected. Needs of humans and availability of quality access to communications as infrastructure for the public good have long since passed a tipping point. Not everyone has a smart device; not everyone can afford new technology or expensive subscriptions. Not everyone can easily learn new technologies or understand the same language. People need a system that works. But understanding what works is complicated. It is important to encourage the focus of the exceptional skills of engineers to work with policy makers to improve technologies because of the reliance of human well-being upon technology. It is time to get creative; transform and modernize without leaving people behind.

11.7 Conclusion and Recommendations In early 2020, the COVID-19 pandemic, a transformative event, resulted in a rapidly developing and evolving worldwide humanitarian crisis requiring substantial science, innovation, technology, engineering, mathematics, government, policy, and operational and interdisciplinary solutions. In addition to expanding technology solutions for infrastructure, business, and market drivers, as a mature industry there have been many omissions in telecommunications that have resulted in opportunities lost – an impaired level of resilience – the digital divide. Never before has sophisticated and intelligent telecommunications technologies demonstrated such potential to resolve problems for humans. Yet telecommunications on the whole failed to achieve its most

326

K.G. August

important imperative: improving well-being for all humans. Instead, our advanced and sophisticated technologies and solutions enhanced harmful disparities. Unless we seize the opportunity for inclusion and diversity in building back better, previous disparities will continue into the future. Many of the problems of infrastructure must be addressed by large investments and significant projects of innovation and building that can only be organized by governments, nonprofit, industry, and academic collaboration. There are opportunities for innovations and inventions to solve the range of issues across the various telecommunications, technologies, and other domains. Engaging in inclusion and diversity is essential to improve resilience, meet needs, and avoid pitfalls for individuals, families, communities, and society locally and globally. For the next era of telecommunication and technology solutions, it is important to meet goals of humanitarian frameworks including social determinants of health, UN SDGs, and CLAS-HHS, for example. These humanitarian frameworks may employ methods different from engineering alone but enable interdisciplinary teams to collaborate for common goals. Many engineering and related organizations have adopted humanitarian frameworks. A significant opportunity – outstanding deficit – is inclusion and diversity. There is little evidence inclusion and diversity has been a widely effective outcome measure. Statistics and evidence illustrate shortcomings in education, industry, inventions, income, ongoing disparity, and bias in performance of some technology and applications. Tragically, these inequities have also emerged as bias and disparity among those most impacted by the COVID-19 global pandemic, including people with disabilities. Although in the past, participation in the digital world was taken to be optional and or almost entirely voluntary, or closely bound with one’s identity in relationship with the economic world – a job, school, presence or perception of disability, health or pre-existing condition, culture, race, age, language, religion, skill, profession, or a social status – that is truly no longer the case. There is no longer a reasonable way to survive without access, inclusion, and accommodation. It is timely for crisis communication theory to evolve. It’s time for solutions to be adaptive to wider more comprehensive reliable infrastructure access, more inclusion, and use cases and accommodation to user interface needs. For example, people with hearing loss must no longer be subjected to long expensive processes and pay more for poor quality communication than peers. People with blindness must no longer endure reliance on employers for accommodation. Nobody should be hobbled – limited to expensive separate technology solutions when standardized generalizable solutions can be easily configured on low-cost mainstream access through inclusion and accommodation. Failure to accommodate is in reality constructing barriers. Accommodations and many more solutions should be standard and built in when products and solutions are planned, instead of an afterthought paid for individually by each person proving a need for accommodation. Aftermarket modifications are significantly expensive, and such accommodations fail to serve the people who need them most.

11 Access, Inclusion, and Accommodation

327

Inventions and innovations inspired by the need to close the digital divide such as Zoom conference calls, closed captioning, screen readers, streaming video, and social media have raised awareness across society and globally. An example of accommodation, closed captioning using automatic speech recognition transcription provides access for everyone to read speech while listening to a teacher, friend, or a colleague. People have hearing needs across a range. When solutions that assist people with hearing loss are generalized, then they can serve a wider range of needs. They can adapt; sensors can learn across a network. Special solutions and user preferences can extend practical solutions. Such a feature can improve speech understanding with or without hearing loss, in complex meetings, education, bilingual situations, or even in telemedicine when understanding is so important because the message is about someone’s health and well-being. Captioning creates a text file and transcript that can be further processed for many purposes including record keeping, reminders, language translations, delayed viewing of a meeting or lecture, illustrating how simple technology can serve many purposes. Better yet, providing key features and information in an appropriate context can reduce the time, energy, and skills required to repeat solutions thereby providing more transparent design for well-being. For example, if a person has a vision loss, a system can automatically offer screen reading features. Extending to another situation, if a person is using a particular medical device, a user guide and support information can be automatically accessed. The information can be presented in a format that is helpful to the user. All of this is greater than the value of the individual feature and more important than the cost of developing a feature that is widely available. By facing challenges of inclusion, diversity and accommodation to access, we learn and grow. Enhancing the user interface with a text to speech or with a link to a healthcare provider follows humanitarian frameworks, improving available infrastructure, improving healthcare, meeting needs for those with differing vision or hearing loss. Bridges the digital divide. By facing challenges of inclusion, diversity, and accommodation to access, we learn and grow. Without accommodation, anybody with a hearing loss, for example, is left out. Without accommodation, anybody with blindness is left out. By the time we add up all the people left out in the various situations, or that must pay for solutions, who is left? At what cost? People are widely diverse. Adapting to a wider set of needs is an imperative to reduce the digital divide. People who are left behind, those who rely on others for care or information, and those who cannot work at technology jobs or educate themselves or their families or children suffer more extreme consequences. They experience less access to the economy and other important resources which has lasting and wide impact across generations. Yet there are people who complain about technology. People grumble about the self-pump gas or self-checkout supermarket. Preferences persist. In reality, jobs have shifted into a technology model. Some will not embrace any form of technology. However, for some, technology is inaccessible and so is training. Disparity itself grew because of the scope and span of reliance

328

K.G. August

on technology, and the effect will continue after the pandemic resolves. Without better affordable broadband internet access, people will be unable to participate in decisions about their own health and education of their children or find or adapt to a job that involves technology. Inclusion and diversity in engineering can improve understanding of lived experience. Engineering is not inclusive nor diverse. There is a long lead time to engage inclusion and diversity in engineering. The pipeline is also not looking good for some years into the future. In spite of so many STEM programs, nothing has changed. Nothing at all. The reality is, there are so few women inventors in the United States; the government has a program to address the problem. There are so few minorities and underrepresented inventors, it is impossible to develop statistics. Resulting technology appears to be a reflection of this fact. There are many ways technology innovations can be applied to reduce the digital divide. Awareness can lead to efforts for a more equitable, inclusive, accommodating environment for everyone. Many enabling technologies exist, but solutions remain fragmented, cumbersome, expensive and are unavailable, inaccessible or do not accommodate segments of the population. There may be many obstacles. Solutions, services, or information might be behind paywalls, infrastructure is inadequate or too expensive, or users may require a navigator or coach to direct a transaction or improve a learning experience. People might not be an expert when they really want to search for an answer; there are indeed special sources for expert information. With a medical condition, for example, even an expert might need new information. Information and technology is continually evolving. Lifelong learning is not limited to one topic or skill. New adaptive mass education models will improve opportunities more than education with a high burden of cost to each individual. Even inexpensive solutions such as standard labels on pharmacy test kits can be automatically linked to an online database for extra information, health navigators, and translations. In this way, technology can transform everyday objects into high tech, information savvy, and accessible devices even in a low-cost model. If this gap is not soon bridged, many more people will find themselves living outside the mainstream economy, unable to benefit. Technology and solutions can be more effectively introduced widely, interoperable, as an effective infrastructure, through development of consortiums, with plans for modernization, development of standards, and more. Collaborations in industry, government, nonprofit, engaging many stakeholders, including individuals, families, members of communities locally and globally can improve understanding how people live, include important use cases, engage diversity, and understand more about how to make future services and access more generalizable, ubiquitous, and how to facilitate affordable and effective yet diverse accommodation of abilities and disabilities, languages and cultures, and personal preferences, across a range of essential contexts for more effective solutions. The IEEE SA DIITA workflow for Transparent Design for Well-being provides opportunities for interdisciplinary international participants to incubate ideas, plan projects, publish, and develop standards. For example, some of the digital and identity-related bias and variability identified by the team includes skin color,

11 Access, Inclusion, and Accommodation

329

age, hearing, vision, mobility, language diversity. The team investigates methods of developing new standards for objective measures to improve performance of solutions. Recognizing that technology played a role in the reemergence of the Cherokee language, the team is inspired to engage many technologists and stakeholders to bring more language and literacy technologies into an ecosystem facilitating interaction, enabling elevation of the dignity of individuals and groups based on culture and language, instead of enforced to conform to the dominant culture, rigid language and technology solutions, a form of digital divide. Maybe we do not seek consensus; maybe we seek multiple solutions. An updated vision of the future of services and access as infrastructure following humanitarian frameworks can facilitate an ecosystem, intraoperative, standardized, plug and play, pervasive, combining physical, virtual and software devices, blended reality, new models of business, education, communication, problem solving, meeting people where they live. Innovations and solutions are a reflection of the lived experiences, thinking, learning style, and creativity of those involved. Inclusion and diversity on interdisciplinary teams can improve how effective technology and solutions can reflect and meet needs of the diversity of humans. Anybody at any age can engage influences that will result in greater inclusion, diversity, and accommodations, to promote resilience for individuals, families, communities, and society locally and globally for better outcomes. The team identified a range of accommodation strategies to increase transparent accessibility to technology, connectivity, power, and energy. Strategies to improve identity and inclusion are based on inputs such as those from communities of interest (CoI), generic or refined identities, economic factors, language, culture, communication abilities, learning styles, subject matter expertise, language skills, and language learning styles. Considerations include communication modalities, location, and engagement with smart environments including smart buildings, transportation, healthcare, schools, and government agencies. The team also identified the importance of incorporating transitional phases of life, health, ability, and experience. Pervasive identities might soon improve recognizing individuals, needs, and preferences in various contexts, for example, at school, during healthcare, at work, in social settings, and delivering more targeted information, functionality, and features that can be made available when needed. Improved sensor fusion and data analysis, and improvements over time might result in better recommendations, forecasting, predictions, warnings, etc. Strategies already used for shopping loyalty programs to recommend coupons for shoppers in supermarkets and pharmacies, or to promote vacations to those who travel, or to design features in high-tech hotels might find more wide applicability and greater impact when applied to diversity, equity, accessibility, and inclusion for people in their everyday lives, providing greater access to the world, society, and to the economy. Such strategies can be used to deliver multimodal extended reality features and information, to improve diet and health, or to adapt the environment to accommodate people in their home, for aging in place, for people with disabilities, and for people in their regular lives. Importantly, they can be used to create powerful accessibility for all designed and developed technologies.

330

K.G. August

Inexpensive devices including watches, smart TVs, network gateways, and IoT sensors increase the reach of access and provide a model to accommodate simplicity and economic disparity even without computers and smartphones. At present, computers and smartphones are more likely helpful, but they might be too expensive and also challenging for certain users; however, slim inexpensive devices remotely operated by trusted navigators can improve likelihood for those with special needs, or protected person, to do well. Lifelong education to adapt to the latest needs for individuals, healthcare, jobs, government and agency interactions, and more. Largescale ongoing education, integrated in systems, for on-the-job training, and more can serve to democratize information improving opportunities for everyone including individuals, families, communities, and also employers, professionals, stakeholders, and more. In a society that turns to technology for resilience and to solve challenges, every person can benefit throughout their lifetime from having access to learning. The COVID-19 emergency represented a gradually evolving complex situation. Decisions were made without complete information. Stakeholders required information from diverse sources in a timely manner, and new tools, enhanced analysis, knowledge, and capabilities. Many people from a wide number of backgrounds had to learn new skills. Many people had to learn about science, technology, information, and public policy. There are adjustments that might have been temporary or might continue longer term. Many enabling technologies have been available during the emergency. It would be very unfortunate to lose the advances or to lose the momentum for progress. In conclusion, at the turn of the millennium, we presented a vision of the future of services and access – a world connected by a network of networks, providing mobile and fixed communication on a very large scale, information, and sophisticated features that advantaged business development and consumer experience and that promised to improve lives and opportunities for everyone, anytime, anywhere. Those networks and technology platforms connected more individuals and businesses with more solutions than ever before. Twenty years later many advances have been realized including the Internet of Things, web-scale IT, sophisticated implanted biomedical solutions, artificial intelligence and machine learning, smart homes, and robots. Yet, when the COVID-19 global pandemic emergency struck, it became obvious that so many of these advances benefited only segments of our community and businesses and only certain portions of our geography. Many shocking outcomes resulted from the disparities in our society including the digital divide. And upon reflection, public health, education, telecommuting, government, information, entertainment, and many other critical aspects of our lives were not entirely supported by the existing infrastructure. The infrastructure doesn’t provide sufficient affordable reliable access to broadband or mobile services. This reality has had a tremendous effect on women, underrepresented, minorities, those with disabling conditions, rural communities, economically disadvantaged, undocumented, their families, communities, and many others with no end in sight because of inequities, complexities, high costs, and lack of infrastructure. Temporary interventions helped during the pandemic, but afterwards disparities will continue.

11 Access, Inclusion, and Accommodation

331

Technology and cultural influences of disparity have been unmasked by the COVID-19 emergency. We relied on technology to vaccinate, conduct surveillance, conduct contact tracing, improve supply chain, engage in remote work, education, communication, and more. Whereas technology advances hold promise to improve opportunities for everyone, many people who are casually omitted from technology advantage are at greater risk instead. When some of us do not do well, the rest of us will not do well. Without broadband, people cannot telecommute, learn information about risks from coronavirus, obtain instructions from public health officials, remotely visit the doctor, or educate their children in the home. Many people worry that high-tech jobs will displace them; yet technology has become an important survival differentiator, improving well-being for individuals, families, communities, locally and globally. Access to skills and education, pervasively, everywhere, throughout life, can improve everyone’s potential to participate. Accommodation through standard technology solutions can simplify affordable solutions that meet people’s diverse needs that increase inclusion to access for those with different preferences, literacy, specialties, cultures and languages, disabling conditions, aging, and differing abilities. Standards improve opportunities in the ecosystem for all to participate. In our modern local and global communities, we face challenges that call for resilience. While we are infinitely connected as humans, we are diverse. We are not one size fits all. Diversity is an advantage, and inclusion is necessary. There are many strategies for accommodation. For modern solutions, we rely on technology that is fit for duty, easy to use, accessible, and widely available. Modern solutions emphasize technology solutions that incorporate science, technology, and innovation. Each technology has distinct properties. Digital and biological systems are different. Success of our systems relies on performance at the interface of technology and humans. Some technology solutions are impacted by variability of data; this is an ongoing opportunity. Many can benefit from standards to reduce bias that can affect individuals, groups, that also benefit diversity. Innovations must include requirements, contributions, and participation representative of all members of society and accommodate the people using the systems, and the expansion of essentials affordably provided to individuals and society through connectivity. Our individual and mutual success depends on the engagement of the largest proportions of our society. The present chapter addressed raising our voices to promote a vision of advancing telecommunications technologies using interdisciplinary collaborative teams and humanity frameworks, through multiple iterative strategies to improve equitable access, inclusion, and accommodation. Importantly, we can document our innovations through inventions, record history, recognize achievements of others, elevate role models, and engage individuals of all diverse backgrounds, ages, specialties, and abilities. Engage to invent the future.

332

K.G. August

Acknowledgments Thank you to everyone who helped. Thank you, Michelle Calabro for all the help and discussions. In Memory of Lisa Nocks, PhD, Historian.

References Ahmadpoor M, Jones BF (2017) The dual frontier: patented inventions and prior scientific advance. Science 357:583–587 Alexander Graham Bell IEEE Profile. https://ethw.org/Alexander_Graham_Bell. Accessed 5 Jan 2022 August KG, Lawrence VB, Saltzberg BR (1999) An introduction to future communications services and access. Bell Labs Tech J 4(2):3–20 Azoulay P, Jones B (2020) Beat COVID-19 through innovation. Science 368(6491):553. https:// doi.org/10.1126/science.abc5792 Azoulay P, Jones BF, Kim JD, Miranda J (2020) Age and high-growth entrepreneurship. Am Econ Rev Insights 2:65–82 Bosworth ML, Ayoubkhani D, Nafilyan V, Foubert J et al (2021) Deaths involving COVID-19 by self-reported disability status during the first two waves of the COVID-19 pandemic in England: a retrospective, population-based cohort study. Lancet Public Health 6:e817–e825. Published Online October 6, 2021. https://doi.org/10.1016/2468-2667(21)00206-1. Accessed 29 Dec 2021 Carreyrou J (2018) Bad blood – secrets and lies in a Silicon Valley startup. Alfred A. Knopf, Inc., New York Centers for Disease Control., CDC.gov Social Determinants of Health, SDoH. https:// www.cdc.gov/socialdeterminants/index.htm. Accessed 19 Sept 2021 Commerce.senate.gov State of telehealth: removing barriers to access and improving patient outcomes . . . . Accessed 27 Dec 2021 Federal Communications Commission, FCC.gov Homework Gap and Connectivity Divide. https:/ /www.fcc.gov/about-fcc/fcc-initiatives/homework-gap-and-connectivity-divide. Accessed 31 Dec 2021a Federal Communications Commission. FCC.gov Strategic Plans Budget. https://www.fcc.gov/ about/strategic-plans-budget. Accessed 31 Dec 2021b Food and Drug Administration Safety Communications Pulse Oximeter Accuracy Limitations. https://www.fda.gov/medical-devices/safety-communications/pulse-oximeter-accuracyand-limitations-fda-safety-communication. Accessed 5 Jan 2022 Foucault M (1995) Discipline and punish: the Birth of the Prison. Second Vintage Books Ed Gasparrini A, Guo Y, Hashizume M, Lavigne E et al (2015) Mortality risk attributable to high and low ambient temperature: a multicountry observational study. Lancet 386:369–375. Published Online May 21, 2015 10.1016/S0140-6736(14)62114-0. Accessed 31 Dec 2021 Health Effects Institute (2019, 2021) https://www.healtheffects.org/publications. Accessed 27 Dec 2021 IASC Guildelines, Inclusion of persons with disabilities in humanitarian action, July 2019, Posted 12 Nov 2019 Originally published 12 Nov 2019. Accessed 2 Jan 2021 IEEE History Milestones Program. https://ethw.org/Milestones:IEEE_Milestones_Program. Accessed 29 Dec 2021 IEEE Humanitarian Activities Committee/Special Interest Group for Humanitarian Technology, IEEE HAC/SIGHT 14. https://sight.ieee.org/. Accessed 29 Dec 2021 IEEE New Jersey Coast Section History Wiki. https://ethw.org/ IEEE_New_Jersey_Coast_Section_History. Accessed 29 Dec 2021 IEEE SA Digital Inclusion Identity Trust Agency (DIITA). https://standards.ieee.org/industryconnections/diita/index.html. Accessed 27 Dec 2021

11 Access, Inclusion, and Accommodation

333

IEEE SA Dignity Inclusion Identity Trust and Agency (DIITA) workflow Transparent Design for Wellbeing. https://standards.ieee.org/industry-connections/diita/index.html. Accessed 29 Dec 2021 IEEE Standards Association (SA) Healthcare and Life Sciences, and Remote Patient Monitoring (RPM) Re-Think the Machine. https://standards.ieee.org/practices/healthcare-life-sciences/ index.html. Accessed 29 Dec 2021 InSIGHT Biography of a SIGHT Volunteer – Paula Muller, PhD IEEE SIGHT. Blog https:// sight.ieee.org/1969-2/. Accessed 27 Dec 2021 Institute of Electrical and Electronics Engineers, IEEE. https://www.ieee.org. Accessed 31 Dec 2021 Interview with Kit August on the ways New Jersey Coast Section is Bringing History Alive, IEEE, IEEE History Center Issue 108, November 2018, pp 3–5. Accessed 29 Dec 2021 Jones BF (2020) Science and innovation: the under-fueled engine of prosperity. Aspen Economic Strategy Group, Aspen Institute Jones BF, Summers LH (2020) A calculation of the social returns to innovation. In: Goolsbee A, Jones BF (eds) Innovation and public policy. University of Chicago Press Jones BF, Kim JD, Miranda J (2021) Immigration and entrepreneurship in United States, 2021/2/18. https://scholar.google.com/citations?view_op the =view_citation&hl=en&user=uzkrkCUAAAAJ&sortby=pubdate&alert_preview_top_rm=2& citation_for_view=uzkrkCUAAAAJ:Z5m8FVwuT1cC. Accessed 5 Jan 2021 Kahan J, Lavista Ferres J (2020) United States Broadband Usage Percentages Dataset. John Kahan Vice President, Chief Data Analytics Officer, Juan Lavista Ferres Chief Scientist, Microsoft AI for Good Research Lab. Please contact for questions. https://github.com/ microsoft/USBroadbandUsagePercentages. Accessed 10 Oct 2021 Katsikopoulos PV (2021) Individual and community resilience in natural disaster risks and pandemics (covid-19): risk and crisis communication. Mind Society 20:113–118. https://doi.org/ 10.1007/s11299-020-00254-0. 13 Received: 16 July 2020/Accepted: 25 August 2020/Published online: 10 September 2020. Accessed 5 Jan 2022 Lai J, Widmar NO (2021a) Revisiting the digital divide in the COVID-19 era. Appl Econ Perspect Policy 43(1):458–464. https://doi.org/10.1002/aepp.13104. Accessed 27 Dec 2021 Lai J, Widmar NO (2021b) Revisiting the digital divide in the COVID-19 era. Appl Econ Perspect Policy 43(1):458–464. https://doi.org/10.1002/aepp.13104. https://onlinelibrary.wiley.com/doi/ epdf/10.1002/aepp.13104. Accessed 5 Jan 2021 Letter to the Members of the National Council for Expanding American Innovation (NCEIA), Gina Raimondo, Secretary, U.S. Department of Commerce. https://www.google.com/ search?client=firefox-b-1-d&q=CI2-SGRCouncilLetter.pdf+. Accessed 27 Dec 2021 Mirabellia MC, Ebeltb S, Damona SA (2020) Air Quality Index and air quality awareness among adults in the United States. Environ Res 183:109185. https://doi.org/10.1016/ j.envres.2020.109185. Accessed on 27 Dec 2021 National Strategy for the COVID-19 Response and Pandemic. http://whitehouse.gov Olsson P, Galaz V, Boonstra WJ (2014) Sustainability transformations: a resilience perspective. Ecol Soc 19(4):1. https://doi.org/10.5751/ES-06799-190401. Accessed 27 Dec 2021 Progress and Potential: 2020 update on U.S. women inventor-patentees | USPTO. https:/ /www.uspto.gov/ip-policy/economic-research/publications/reports/progress-potential Progress and Potential: 2020 update on U.S. women inventor-patentees. Accessed 31 Dec 2021 Revealing the toll of COVID-19. https://www.who.int/publications/i/item/revealing-the-toll-ofcovid-19. Accessed 27 Dec 2021 Rotarou ES, Sakellariou D, Kakoullis EJ, Warren N (2021) Disabled people in the time of COVID-19: identifying needs, promoting inclusivity. J Glob Health 11:03007. https:// www.ncbi.nlm.nih.gov/pmc/articles/PMC7897447/. Accessed 19 Sept 2021 Screen reader for the blind. https://devpost.com/software/unstoppable-auto-caption Screen reader images for the blind. https://youtu.be/XIEmm0KnhGc Shah ASV, Langrish JP, Nair H, McAllister DA et al (2013) Global association of air pollution and heart failure: a systematic review and meta-analysis. Lancet 382. September 21,

334

K.G. August

2013 www.thelancet.com. Published Online July 10, 2013 https://doi.org/10.1016/S01406736(13)60898-3. Accessed on 27 Dec 2021 Sjoding MW, Dickson RP, Iwashyna TJ, Gay SE, Valley TS (2020) New Engl J Med 383:25. nejm.org December 17, 2020. Accessed 21 June 2021 Sociavi. https://sociavi.com/. Accessed 31 Dec 2021 Think Cultural Health - Standards - Culturally and Linguistically Appropriate Systems - Health and Human Services. https://thinkculturalhealth.hhs.gov/clas. Accessed 5 Jan 2022 Thomas Alva Edison IEEE History Wiki. https://ethw.org/Thomas_Alva_Edison. 5 Jan 2022 U.S. Department of Labor Americans with Disabilities Act (ADA). https://www.dol.gov/general/ topic/disability/ada. Accessed 19 Sept 2021 United Nations Sustainable Development Goals UN SDG. https://www.un.org/ sustainabledevelopment/health/. Accessed 19 Sept 2021 United States Census Bureau census.gov. https://www.census.gov/library/stories/2021/01/womenmaking-gains-in-stem-occupations-but-still-underrepresented.html. Accessed 27 Dec 2021 United States Patent and Trademark Office (USPTO) website, USPTO.gov. https://www.uspto.gov/ . Accessed 31 Dec 2021 USPTO releases SUCCESS Act report to Congress | USPTO. https://www.uspto.gov/about-us/ news-updates/uspto-releases-success-act-report-congress. Accessed 31 Dec 2021 USPTO.gov US 10,848,711 Communication System for Use with Persons, Paula Muller. https://patft.uspto.gov/netacgi/nphProtected Parser?Sect1=PTO2&Sect2=HITOFF&p=1&u=%2Fnetahtml%2FPTO%2Fsearchbool.html&r=1&f=G&l=50&co1=AND&d=PTXT&s1=10848711&OS=10848711&RS=10848711. Accessed 31 Dec 2021 Walia A, Ravindran S (2020) America’s Racial Gap & Big Tech’s Closing Window, Deutsche Bank Research, North America, Technology Strategy Thematic Research, September 2, 2020. Accessed 27 Dec 2021 Watzinger M, Krieger J, Schnitzer M (2021) Standing on the shoulders of science, Centre for economic policy research discussion paper #13766. Accessed on 27 Dec 2021 World Health Organization. www.who.int/disabilities/world_report. Accessed 27 Dec 2021 Yaqu R, Ahmad S (2020) Artificial intelligence assisted consumer privacy and electrical energy management. Global J Comp Sci Technol 20(1) Version 1.0 Year 2020 Online ISSN: 09754172 & Print ISSN: 09754350 Yaqub R, Ali M, Ali H (2021) DC microgrid utilizing artificial intelligence and phasor measurement unit assisted inverter. Energies 2021(14):6086. https://doi.org/10.3390/en14196086

Katherine Grace August [SM] received her PhD in biomedical engineering from Newark College of Engineering, New Jersey Institute of Technology, MSc in computer science-MIS from Marist College, Poughkeepsie, New York, and BFA in communications design at Parsons, The New School for Design, New York, New York. She is a research guest at the Stevens Institute of Technology. She is a biomedical and communications engineer. Formerly, she was with Bell Labs as an MTS in New Service Concepts Systems Engineering, 1991–2002. She is currently a volunteer with IEEE SA DIITA Transparent Design for Wellbeing, Chair of the NJ Coast PACE SIGHT Group, Vice Chair of Region 1 PACE, and Vice Chair of the Computer Chapter and IEEE ComSoc History Committee. LinkedIn profile:https://urldefense.com/v3/__https:// www.linkedin .com/in/kit-august-0000331/__;!!NLFGqXoFfo8 MMQ!vrudFUDuGuJe2i0DDNwg_cy9ZpOyTzS0ZH308r5rDfBdli BpykX42sagNU2XBaV4Nd-pkn776F7NBVyxD85K EU7MJgnHg$

Chapter 12

Private Land Mobile Radio Services In-building System Design Considerations Asuncion (Beng) Connell

Abbreviations and Acronyms AHJ BDA CFR CO DAQ DAS dB dBc dBm DL EMI F1 F2 FCC FO LMR NFPA PIM PLMRS RF RFI RX TDI TIA TSB UHF UL VHF VSWR

Authority Having Jurisdiction Bidirectional Amplifier Code of Federal Regulations Certificate of Occupancy Delivered Audio Quality Distributed Antenna System Decibel Decibel Relative to Carrier Power Decibel Referred to One Milliwatt Downlink Electromagnetic Interference Frequency 1 Frequency 2 Federal Communications Commission Fiber Optic Land Mobile Radio National Fire Protection Agency Passive Intermodulation Private Land Mobile Radio Service Radio Frequency Radio Frequency Interference Receive Time Delay Interference Telecommunications Industry Association Telecommunications Service Bulletin Ultra High Frequency Uplink Very High Frequency Voltage Standing Wave Ratio

A. (Beng) Connell () Jacobs, Walnut Ave., Clark, NJ, USA © The Author(s), under exclusive license to Springer Nature Switzerland AG 2023 M. S. Greco et al. (eds.), Women in Telecommunications, Women in Engineering and Science, https://doi.org/10.1007/978-3-031-21975-7_12

335

336

A. (Beng) Connell

12.1 Introduction PLMRS are used by companies, local governments, and other organizations to meet a wide range of communication requirements, including coordination of people and materials, important safety and security needs, and quick response in times of emergency.1 In the United States, the Federal Communications Commission (FCC) regulates all PLMRS through Code of Federal Regulations (CFR) Chapter 47, Part 90., Subpart 219 which is on the use of signal boosters. There are two mobile radio services that fall under this category: the industrial/business or regular/standard land mobile radio (LMR) for use by companies, industries, and private businesses such as food and restaurants, manufacturing, mining, transportation, hotels and resorts, sports and entertainment, education, and many more that do not fall under the category of commercial cell carrier or provider and public safety mobile radio systems for use by public safety entities and first responders in responding to different types of emergencies. Both require reliable communication in their place of business and operation. Public safety radio services have more stringent requirements than the standard LMR services for both indoor and outdoor coverage to deliver secure and reliable mission-critical communications in a variety of environments, scenarios, and emergencies. The next pages of this chapter will discuss the following topics: • History of in-building systems and why it failed • Designing an in-building system, the steps, the requirements, and the different types • Design considerations • Elements of an in-building system, equipment, devices, and component • In-building software tools that aid the designer

12.2 History In the early days, radio coverage was mainly provided through high-power radio systems such as repeaters and base stations (Fig. 12.1). Radio communication designers take into consideration the coverage inside the building by adding an assumed value of loss figure depending on the known composition of the outside wall of the structure such as 10 dB, 20 dB, or even as high as 40 dB. In underground areas such as tunnels (Fig. 12.2), radio coverage is provided through high-power radio systems or repeaters with a high gain directional antenna pointing toward the tunnel or by using coaxial cable mounted along the tunnel walls with slotted holes throughout its length (leaky cable) where RF energy slowly leaks as it travels along the length of the cable.

1 FCC

47 CFR, Part 90

12 Private Land Mobile Radio Services In-building System Design Considerations

337

Fig. 12.1 Portable inside multistory building without DAS

Fig. 12.2 Portable inside tunnel with leaky cable

However, this technique was proven unreliable. It does not consider the end user’s equipment capability to talk-back (uplink) to the far away repeater. While it is easy for the repeater with its high elevation and Effective Radiated Power (ERP) to talk-out (downlink) to the end user, the end user’s device has difficulty penetrating those high loss walls due to its low ERP compared to the repeater. Additionally, there are areas inside, especially within the core of the building that are shaded that cannot be reached by the outdoor repeater. For public safety, these areas are the priority including elevators, stairwells, and all building egresses. And with the growing need/requirement of providing good quality and reliable communication, designing radio communication systems using the high-powered repeater and outside walls alone for indoor use is not a good enough solution. A system must be designed such that the source of energy is as close as possible to the target area.

338

A. (Beng) Connell

12.3 Designing an In-building System An in-building system has two main elements, headend and distribution. A typical in-building system has a source of RF and a target area that requires radio coverage that cannot be reached directly by a remote repeater or a donor site. A good inbuilding system is one that can provide a uniform level of RF or within ±2 dB throughout the target/service area. An in-building system that is properly balanced will be more efficient, and the area that can be covered increases significantly. The first step in designing a Distributed Antenna System (DAS) is gathering information such as the frequencies that are involved, whether VHF, UHF, or 700/800 MHz, the required number of channels, and the intended user whether regular LMR, public safety, or others. Today’s in-building technology allows for multiple band DAS where LMR and public safety can be combined in one system, but not with commercial cellular systems (yet). The next step is determining the need for in-building treatment and knowing the level of RF energy available for distribution to the structure or building of concern. The RF source can originate remotely from an off the air repeater in which case an outdoor or roof antenna (called donor antenna) will be required to receive and transmit signal back to the repeater. It can also be a radio base station inside the target building where the RF can be tapped from its output ports. The performance of the RF source (called donor site) is very important as it establishes the quality of signal that will be fed and distributed into the DAS. Baseline measurement is necessary to identify the donor site as well as the available RF power to feed the in-building system. Signal measurements include RF level received from the donor site on the roof and at each building floor area, building wall to wall and floor to floor losses. Once signal measurement is performed and measured received levels are analyzed, a preliminary link budget can be developed.

12.3.1 Link Budget Calculation The link budget calculations for most radio systems are relatively straightforward. It is a list of system gains and losses and a target figure that defines satisfactory performance. Attainment of this figure is influenced by factors such as customer requirements, industry requirements, as well as local, state, and federal governing bodies. The link budget calculation has two components, the downlink path and uplink path. Downlink (talk-out) is the direction of communication from the repeater or donor site to the end user or subscriber. Uplink (talk-back) is from the end user back to the donor site. Both paths are equally important. The preliminary link budget will determine whether there is a need for inbuilding coverage enhancement and, if there is, the type of in-building system to be deployed whether passive, active, or a combination of both.

12 Private Land Mobile Radio Services In-building System Design Considerations

339

Fig. 12.3 Multistory building with passive DAS

12.3.2 Passive DAS A passive DAS uses a bidirectional amplifier (BDA) to compensate for path loss. It distributes RF signals by using coaxial cable, splitters, couplers, or taps terminated with no-gain indoor antenna or 50 ohm load (also called dummy antenna) at the end of the cable (Fig. 12.3). The farther away from the BDA, the more loss the RF energy will experience, the weaker the signal. It is important to carefully design and calculate link budget to ensure that uniform RF power will be distributed throughout the entire target area.

12.3.3 Active DAS An active DAS uses a bidirectional amplifier to compensate for path loss with the addition of active components that conditions, amplifies, and converts RF energy into optical energy (Fig. 12.4). It uses fiber-optic cable to transport the converted energy to a remote location. At the remote location, the optical energy is converted back to RF before finally distributing it to the target area. This technique allows for the RF energy to be transported over long distances.

340

A. (Beng) Connell

Fig. 12.4 High-rise building with active DAS

12.3.4 Comparison Between Different Types of DAS The Tables 12.1, 12.2 and 12.3 below summarize the benefits and weaknesses of the two main types of in-building distribution system.

Table 12.1 Passive and active DAS comparison DAS type Passive

Active

Advantages 1. Low infrastructure cost 2. Easy to install 3. Low maintenance cost 1. Excellent for long distances and complex wide area coverage due to low fiber-optic cable loss 2. System is easily expandable

Disadvantages 1. Small coverage area

1. High infrastructure cost 2. High maintenance cost due to active components 3. Require power and real estate at remote sites 4. Potential noise source due to the number of amplifiers in the system

12 Private Land Mobile Radio Services In-building System Design Considerations

341

Table 12.2 Coaxial cable and fiber-optic cable comparison RF transport means Coaxial cable

Fiber-optic cable

Advantages 1. Cheaper than FO cable to install 2. Cost-effective solution for simple and small area coverage requirement 1.Very low loss allowing for several kilometers without the need for amplification 2. Immunity to EMI or RFI due to fiber being non-conductive medium 3. Security against signal interception

Disadvantages 1.Distance limited 2. Need to introduce a line amplifier to boost signal back to acceptable level 1. More expensive compared to coaxial cable 2. Fiber-optic backbones can be very complicated in large systems 3. Home run fiber designs (linear) from the headend to the remotes are required

4. Flexible and lightweight for easy installation 5. Low maintenance cost 6. Simple installation cost effective compared to coaxial cable 7. Large bandwidth Table 12.3 Indoor antenna and radiating cable comparison Distribution means Indoor antenna

Radiating cable

Advantages 1. Ideal for large open areas 2. More effective for providing coverage to isolated areas 1. Signal is easy to control 2. Reduce isolation requirement

Disadvantages 1. Antenna to antenna isolation can be tricky but attainable

1. High coupling loss 2. Coverage is not targeted only to desired areas

3. Less susceptible to interference due to low power on the cable

12.3.5 Customer and Industry Requirements The job of the engineer is to design the system to meet specific customer requirements and within budget while meeting all codes, standards, rules, and regulations (Merle 1999). The design should meet certain criteria set by the industry as to system reliability, percent coverage of the target area. Customer requirements include coverage in priority areas, reliable communications especially in places where a commercial radio service cannot reach. For businesses, both private and public and even public safety agencies, the cost of building DAS is a major consideration in decision making as to the type of DAS to be deployed. Public safety DAS have more stringent requirements than the standard LMR DAS. For example, public safety DAS requires a minimum signal level that is

342

A. (Beng) Connell

between −95 dBm and −85 dBm to ensure full quieting and proper operation of fully digital encrypted systems which require these levels or better (Telecommunications Industry Association 2012). Any system should use these as a benchmark. The Telecommunications Industry Association (TIA) publication TSB-88.1-D (2012) provides guidelines and industry recommended values and requirements in designing land mobile radio systems and in calculating link budget. TSB88.1-D Table A.1 shows figures used in determining minimum RF level in a faded environment, TSB-88.1-D Table D5 shows typical portable loss values for various frequency bands. TSB-88.1-D Table D6 shows minimum required coverage reliability of 90% for LMR and 95% for Public Safety. TSB-88.1-D Table D7 shows minimum required delivered audio quality (DAQ) 3.0 for LMR and DAQ 3.4 for public safety.

12.4 Design Considerations 12.4.1 System Noise Floor The addition of BDAs to a radio system can increase the noise in the system and potentially impact performance. Noise floor at the donor repeater station will rise abruptly as more in-building antenna systems are hooked up to the system. Within the DAS, the primary noise contributors are the BDA plus all the FO amplifiers in an active DAS. An amplifier by itself has inherent noise characterized by its noise fig. (NF). It will amplify not only the desired signal but also the noise that came with the signal. The higher the NF, the noisier the amplifier. When the DAS transmits signal back to the repeater or donor site, it will transmit the desired signal along with the noise generated within the in-building system. Careful selection of BDA and amplifier equipment with low noise characteristics is the key in keeping the noise level at a minimum and maintaining a high signal-to-noise ratio (SNR). A high SNR figure is desirable; it determines the noise present in the system is tolerable. If noise calculation shows that the noise level received from the proposed DAS to be much lower than the existing noise floor at the donor site, the impact of adding new DAS will be negligible. Below (Fig. 12.5) is an illustration of the noise sources in an in-building system depicted in Fig. 12.4.

12.4.2 Antenna Isolation Another important design criterion is the isolation between donor antenna and antennas inside the building. To obtain maximum isolation, select a high gain, highly directional donor antenna, ensure no indoor antenna is pointing directly to the donor antenna, and that no indoor antenna is mounted next to a window or near the exterior wall of the building. The isolation is the attenuation between the donor antenna

12 Private Land Mobile Radio Services In-building System Design Considerations

343

Fig. 12.5 Typical worst case uplink noise in an active DAS with three remote units

Fig. 12.6 Service antenna signal feedback to BDA

cable port near the BDA and indoor antenna cable port near the BDA; see Fig. 12.6 (Overby 2007). It can be calculated using the isolation formula below (Overby 2007). AI = PL + DAG + IAG + LL

.

where: AI = Antenna Isolation PL = Path Loss between the donor and indoor antenna DAG = Donor Antenna Gain in the Direction of Indoor Antenna = Donor Antenna Back Lobe Gain

344

A. (Beng) Connell

IAG = Indoor Antenna Gain in the Direction of Donor Antenna = Indoor Antenna Back Lobe Gain LL = Line loss (donor antenna to BDA and indoor antenna to BDA) PL is given by the Free Space Loss (FSL) formula: F SL = 32.4 + 20LogD + 20LogF

.

where: D = Distance between donor and indoor antenna in kilometers F = Frequency in MHz The antenna back lobe gain can be calculated using the antenna gain and front to back ratio on the antenna manufacturer’s published specifications. The BDA has two antennas, one outdoor (donor antenna) and one indoor (service antenna). In Fig. 12.6, note that the receive (RX) frequency (F1) of the donor antenna is the same as the transmit (TX) frequency (FI) of the indoor antenna. When there is no sufficient isolation between the donor antenna and any of the indoor antennas, the signal from the indoor antenna is fed back to the donor antenna forming a feedback loop, causing the BDA to oscillate. As the BDA oscillates, the RF input into the BDA increases; there is a point where the amplifier will be overwhelmed and reach saturation. When this occurs, it will generate harmful harmonics and intermodulation products and unwanted noise, will interfere with nearby radio systems, and will degrade the performance of the donor site. The industry standard for minimum isolation is 20 dB above the gain setting of the BDA. This is found in 2019 NFPA 1221 Chapter 9, Sect. 6.9 Donor Antenna (NFPA 2019). To comply with this standard, this value should be added to the isolation calculation result from the formula above.

12.4.3 Propagation Delay Propagation delay is the length of time from when a signal leaves its origin up to the time it reaches its destination. It is a function of the distance and frequency and the type of media that the signal passes through. The signal travels faster in free space and slows down as it encounters barriers in free space, more so as it travels through a system of cables, connectors, filters, splitters, and decouplers in an in-building system. Manufacturers of current land mobile radio networks in operation today, P25 phase 1 and some upgraded to phase 2, specify their equipment maximum tolerable propagation delay to be typically 33 and 15 microseconds, respectively. The in-building system must be designed within this limit (Fig. 12.7). When the two or more copies of the same signals arrive at the portable within the radio systems specified maximum tolerable propagation delay, the portable will perceive the signals as one. When one signal is a lot stronger than the other, the portable will see the stronger signal and reject the weak signal. When the two

12 Private Land Mobile Radio Services In-building System Design Considerations

345

Fig. 12.7 Multiple RF signal sources into BDA and portable

signals are more than the radio system’s specified maximum allowable propagation delay, and are typically within ±3 dB, TDI occurs. The impact is negative on system performance, degraded voice quality, and worst loss of communication. To avoid harmful time delay interference (TDI), ensure sufficient isolation between the donor and the indoor antenna, and select BDA equipment that meets the propagation delay requirement. In some cases, it is necessary to widen the filter bandwidth in order to meet delay requirement. A wide-bandwidth filter will pass the signal faster (short delay). A narrow-bandwidth filter will pass signal slower (long delay). Caution should be taken in employing this technique making sure that desired channel bandwidth and adjacent channel rejection are not compromised and do not violate passband requirements per FCC 47 CFR Part 90.

12.4.4 Intermodulation Distortion The mixing of two or more frequencies will result in the generation of new signals called intermodulation products which are not wanted. Some intermodulation products are harmful, some are not. Some, usually third-order products, will fall within the passband of the desired signal. And if the amplitude of this intermodulation product is high enough, it will cause signal distortion. Since a bidirectional amplifier (BDA) is essentially a repeater of multiple frequencies and sometimes of multiple frequency bands, intermodulation products will be generated. There are three major types of intermodulation interference: transmitter generated, receiver generated, and intermodulation due to passive components. Transmitter-generated intermodulation occurs when two or more transmitter frequencies mix with another at the transmitter final stage. Receiver generated are two or more transmitter frequencies mixing at the receiver’s final

346

A. (Beng) Connell

output amplifier stage. Passive intermodulation (PIM) occurs when two or more transmitter frequency mixes in a non-linear junction or passive devices such as connectors, switches, jumpers, splitters, couplers, and isolators (Passive Intermodulation Techniques 1999). In an in-building distribution system, IM products can be generated in any device where the transmit (TX) and receive (RX) frequencies are together such as in BDAs, power amplifiers, remote units, and radiating cable. Furthermore, there are many locations where passive intermodulation can potentially occur due to the splitters, combiners, and tappers that are necessary in the design to distribute the signal throughout the target area. When the transmitter, receiver, and passive generated IM products are combined on fiber, coax, or radiating cable, the products will end up at the equipment receivers. And if there are separate TX and RX radiating cable runs, TX frequencies and IM products can be coupled through space between the cables if they are in close proximity. To keep intermodulation distortion at a minimum, do not operate an amplifier in its non-linear region, and select amplifiers and passive components with low intermodulation distortion (IMD) characteristics. Typical intermodulation values are expressed in dBc (decibel relative to carrier power). When comparing intermodulation characteristics, it is important to consider the carrier power that the intermodulation level is measured against. Theoretically, third-order intermodulation level increases at a rate of 3 dB per 1 dB increase in carrier power. If an amplifier third-order intercept point (IP3) specification is available, select a high IP3 value. It defines the maximum input to the amplifier before it goes to compression, the higher the IP3 the better the linearity of the amplifier.

12.4.5 Composite Power and Power per Channel The composite RF power is the sum of all the carriers/channels’ power combined and passed through the same or common transmission system. The output power in the BDA equipment manufacturer’s published specification is a composite power expressed either in watts or in dBm. This value should not be exceeded when calculating power per channel budget. Given the composite power, the power per channel is obviously the composite power divided by the number of channels. The more channels, the lower will be the available power per channel. P ch = P comp/n

.

where: Pch = Power per Channel Pcomp = Composite power n = Number of channels

12 Private Land Mobile Radio Services In-building System Design Considerations

347

When composite power is expressed in dBm (decibel referred to 1 milliwatt) and since decibel is a logarithmic scale, the equation becomes: P ch = P comp − 10 log n

.

where: Pch = Power per Channel in dBm Pcomp = Composite Power in dBm n = number of channels When composite power is expressed in watts, first convert watts into dBm using conversion formula: Power in dBm = 10 log (Power in watts × 1000)

.

To convert power in dBm back to watts, the conversion formula is: 

Power in watts = 10

.

Power in dBm−30 10



12.4.6 Balancing BDA Gain, Output Power, and Isolation The rationale of an in-building distribution system is to provide uniform, sufficient signal to an area where there is no coverage or where coverage is inadequate. It must be designed such that it will deliver the required output power per channel that will provide the desired receive signal level at the target area. The required output per channel is a balance between the required received signal level by the end user on one side and the maximum rated (composite) output power of the BDA on the other side. The BDA maximum output power can be attained by its proper gain setting. The BDA gain setting is a balance between the RF level received from the donor site and the required antenna isolation. The BDA will require more gain to boost weak signals than strong signals. The higher the BDA gain, the higher the isolation requirement. Today’s BDAs are equipped with Automatic Gain Control (AGC) circuitry that automatically adjusts the BDA gain setting to ensure it will not exceed its maximum rated output. The BDA will output the same level of RF for a given range of input as long as it is within its dynamic range. While this feature provides convenience, note that while the BDA is operating in its dynamic range, its gain is varying, and so is the isolation requirement, but the calculated antenna-to-antenna isolation remains the same. To avoid isolation issues, consider that the BDA is operating in its maximum gain in designing the system. Ensure that the RF signal radiated at the donor antenna and on every indoor antenna in the distribution system will meet the required received signal level by the end user without exceeding FCC requirement of maximum ERP for signal boosters. FCC 47 CFR, Part 90 mandates that the maximum Effective Radiated

348

A. (Beng) Connell

Power (ERP) of the DAS shall not exceed 5 watts in both uplink and downlink for each retransmitted channel.

12.5 Elements of an In-building System An in-building system has two main elements: headend and distribution (Fig. 12.8). The headend (Headend DAS) consists of the RF energy source and BDA location. It is concerned on how to obtain the RF energy, where the energy will come from and how it will be extended to locations that require coverage treatment. There are two classifications of BDAs, Class A and Class B. Class A BDA is designed to retransmit two or more channels within a specific passband, on a per channel basis (channelized). Each of its channel filter passbands do not exceed 75 KHz. Class B BDA is designed to retransmit signals within a wideband of frequency where the passband exceeds 75 KHz. It is desirable (if not a must) to use Class A BDA in public safety DAS installations. The distribution (building DAS) is a system of coaxial cable, radiating and nonradiating, and point source antenna that can be mounted on the ceiling or wall. An active DAS will have remote units at the point of distribution that provide amplification to the signal.

Fig. 12.8 In-building system elements

12.5.1 Headend DAS Equipment and Components 1. Directional antenna captures RF signals from the donor site ideally mounted on the roof and where there is a clear line of site path to the donor site. The antenna must have a high gain with a high front to back ratio. 2. Bidirectional amplifier (BDA) filters and amplifies captured RF signal. It can be Class A or Class B as long as it meets design criteria and FCC Rule Part § 90.219. The following are some BDA specifications to consider: • • • • • •

FCC type accepted Frequency range Channel bandwidth Channel capacity Channel filtering RF power output (uplink and downlink)

12 Private Land Mobile Radio Services In-building System Design Considerations

• • • • • • • • •

349

RF gain (uplink and downlink) Automatic gain control (uplink and downlink) Automatic squelch control (uplink and downlink) Intermodulation distortion Propagation delay Noise figure Input/output VSWR Power consumption Alarming, management, and control

3. Fiber/optical module is used in deployment of an active DAS. It converts downlink RF signal to light and uplink optical signal back to RF. This unit can be short range or long range depending on distance or the required optical budget or the number of fiber remote units required in the design. Some parameters to consider in choosing equipment are: • Optical parameters – – – – – – –

Frequency range Maximum optical input to receiver Wideband noise Fiber-optic cable type Output optical power Optical budget Alarming and control

• RF parameters – – – – – –

Port configuration/number of ports Maximum RF input level to TX side RF TX to RX port isolation RF link gain Input IP3 Uplink noise figure

• Environmental/mechanical – – – –

Temperature range (operational and storage) Humidity Dimension AC power consumption

12.5.2 Building DAS Equipment and Components 1. Single mode fiber-optic cable used to transport RF over long distances. Some FO cable characteristics to consider are:

350

• • • • • •

A. (Beng) Connell

Fiber-optic connector type Minimum bending radius Outer diameter Temperature range (operational and storage) Fire resistance Indoor or outdoor application

2. Fiber remote unit converts downlink optical signal back to RF before feeding it to the building DAS and converts uplink RF signal to light before sending it back to the optical module at the headend. This unit provides a set amount of gain to provide required power per channel for the number of channels in the system design. Some fiber remote unit specifications to consider are: • RF parameters – – – – – – –

Frequency range System gain Composite output power Max RF input level Uplink noise figure Wideband noise Spurious emission

• Fiber-optic parameters – – – – – –

Port configuration/number of fibers Separate DL and UL ports Wavelength: single fiber Separate DL and UL Temperature range (operational and storage) Max optical budget

• Environmental – – – – – 3. 4. 5. 6.

Operational temperature Humidity: 10–95% Electrical Power consumption Alarms, local, and remote

Coaxial cable – low loss, minimum bending radius, and fire rating Radiating cable – low loss, low coupling loss, and minimum bending radius Indoor antenna – unity gain, small form factor Passive RF devices should be low PIM • • • •

Signal splitters/combiners Couplers/decouplers Tappers Connectors

12 Private Land Mobile Radio Services In-building System Design Considerations

351

12.6 In-building System Design Tools Today there are a few in-building software modeling tools available in the market that aid the design engineer in accomplishing his task and making it easier and less difficult. These software tools have a built-in database of building properties and losses as well as a wide selection of devices, equipment, and components available with the manufacturer’s specification information. Such tools allow the designer to model a building in 3D space and be able to position the equipment and components, mount antennas, and run cable along walls, stairwells, and through floors exactly where the designer wants them. After running software analysis, the result can be viewed in 3D with color coded received signal levels and even print the bill of materials. (How easy was that!)

12.7 Conclusion The in-building industry is growing very fast and DAS installations are becoming bigger and complex. It is a race against the changing ordinances, building codes, rules and regulations driven by lessons learned in every calamity, disaster, emergency when communication is an absolute must but is failing. To ensure the welfare and safety of the general public, the 2021 edition of NFPA1, now requires that in all buildings, new and old, to allow or provide a means for the public safety first responder to operate their two-way mobile radio system inside buildings. This inbuilding system that is provided by the building owner is to be inspected by the Authority Having Jurisdiction (AHJ) to confirm that it meets requirements. And just like other fire codes, failure to comply means a Certificate of Occupancy (CO) will not be issued. With the growing requirements to building owners to provide not only their tenants but also the first responders with good quality communications, the development in manufacturing technology is also evolving. In-building design suites are emerging as well as all-in-one equipment solution, headend equipment, and remote amplifiers with improved specifications and performance. The development in computer software technology is also advancing; there will be more engineering tools that will be developed that will be available for the engineer/designer or anyone that has access to technology. But do not forget that computers and software are developed and programmed only as tools to aid the designer. Ultimately, it is the design engineer who is responsible and who will make the decision, formulate the link budget, carefully choose the appropriate equipment that meets the design criteria, and determine compliance with the customer and industry requirements as well as all applicable federal, state, and local laws, ordinances, regulations, and codes.

352

Appendix

A. (Beng) Connell

12 Private Land Mobile Radio Services In-building System Design Considerations

353

354

A. (Beng) Connell

References Federal Communications Commission, 47 CFR 90.219, use of signal boosters Fiber Optic Communications Tutorial Ford S (2005) The ARRL emergency communication handbook. WB81MY, ARRL, The National Association for Amateur Radio, Newington Gopala Krishan T, Beema Beevi S Antenna wave propagation, Noorul Islam College of Engineering, Kamaracoil Department of Electronics and Communications Engineering Imel KJ, Tolman T (2003) Understanding wireless communications in public safety, a guidebook to technology, issues, planning and management. The National Law Enforcement and Corrections Technology Center, Rocky Mountain Region. A Program of the National Institute of Justice Merle V (1999) Indoor network solutions. In: Nortel network presentation paper Mullett GJ (2013) Wireless telecommunications systems and networks. In: Introduction of wireless telecommunications systems and network NFPA 1221 (2019) Standard for the installation, maintenance, and use of emergency services communications system, 2019 Edition

12 Private Land Mobile Radio Services In-building System Design Considerations

355

Overby S (2007) Best practices for in-building communications by NPSTC in-building work group Passive Intermodulation Techniques (1999) Summitek instruments paper, Englewood, Colorado, USA Poor VH, Wang X, (2002) Wireless communications systems advance techniques for signal reception Rappaport T (1996) Indoor propagation model. Wirel Commun Princ Pract 3(11):123–138 Scholz P Basic antenna principles for mobile communications. Kathrein-Werke KG, Rosenheim Telecommunications Industry Association, TSB-88.1 (2012) Wireless communications systems performance in noise and interference-limited situations. Telecommunications Industry Association, Arlington Webster JG (1999) Antenna and propagation. In: Wiley encyclopedia of electrical and electronics engineering. Wiley, New York

Asuncion (Beng) Connell, E.C.E., P.M.P. Growing up in the Philippines, Beng attended college in 1980 with the intention of becoming a Chemical Engineer. However, after 2 years of navigating through Math, Physics, and Chemistry, she realized that Chemistry was not her strongest point and opted to switch her major to Electronics Engineering instead. She subsequently graduated with a Bachelor of Science degree in Electronics and Communications Engineering (BS ECE) from the University of Santo Tomas, Manila, Philippines, in 1985 and became a registered Professional ECE the same year. Upon the recommendation by one of her college professors after graduating, she began her career with Miltech Industries, Inc., Philippines, in land mobile radio communications as Radio Frequency (RF) Engineering assistant. While under their employment, she enjoyed designing two-way radio systems; doing site surveys and site testing; demonstrating equipment capability to customers; and attending public conferences and bidding. In 1990 after 4 years, she shifted from the private land mobile radio industry to join Express Telecommunications, Inc., a pioneer cellular carrier in the Philippines. She remained with them for a short period of 4 months, at which time she returned to Miltech Industries. Working on two-way radio systems again and furthering her education toward a Master’s Degree in Public Management, she attained that goal in 1996 from the University of the Philippines just prior to immigrating to the United States. In 1997, she joined RCC Consultants, Inc., headquartered in Woodbridge, New Jersey, as RF Engineer. This position encompassed several northeast markets in their nationwide 900MHz radio site acquisition and build-out project for BellSouth. She was with the Commercial Market Group until the project was completed and eventually joined the Public Safety Group of RCC, where her main assignments were various Port Authority of New York and New Jersey projects. She also assisted in other public safety projects for the City of Philadelphia, Pennsylvania, and the City of Norwalk, Connecticut, and in 2009 became a Project Management Professional. In 2015, RCC Consultants became a part of Black & Veatch. She remained with them in the same capacity until her most recent employment.

356

A. (Beng) Connell In 2019, Beng took a position with Jacobs Engineering Group and has been involved in several Private Land Mobile Radio Services (PLMRS) in-building design projects for Grand Coulee Dam, Columbia Boulevard Wastewater Treatment Plant, Delaware River Port Authority, Metropolitan Transit Authority, East and North River Tunnel Restoration for AMTRAK. She is currently on an on-call consultant assignment for the Port Authority of New York and New Jersey.

Chapter 13

On the Entanglement Role for the Quantum Internet Jessica Illiano and Angela Sara Cacciapuoti

13.1 Introduction Quantum Internet is envisioned to be the final stage of the quantum revolution, enabling quantum communications among remote nodes through the interconnection of heterogeneous quantum networks (Cacciapuoti et al. 2020). Quantum Internet thereby allows several astonishing applications (Kimble 2008; Pirandola and Braunstein 2016; Wehner et al. 2018; Cacciapuoti et al. 2019) ranging from cryptography (Ekert 1991; Lo et al. 2014) through secure communication (Gottesman and Lo 2003), to distributed quantum computing (Cuomo et al. 2020). Different from classical communications, quantum communications are based on the laws of quantum mechanics. Hence, the inherent different underlying physical mechanism poses the significant challenge to completely renew the design of protocols that allow the exchange of information, both classical and quantum . A new and different concept of connectivity (Illiano et al. 2022)—linked to entanglement, a phenomenon with no counterpart in the classical world (Cacciapuoti et al. 2020)— opens up to several issues that require such a paradigm shift. Indeed, there are many open questions related to the design of a quantum network, starting from the simple task of successfully transmitting quantum information between two nodes. In fact,

The results presented in this work were obtained in part using an IBM Q quantum computing system as part of the IBM Q Network. The views expressed are those of the authors and do not reflect the official policy or position of IBM or the IBM Q team. J. Illiano () · A. S. Cacciapuoti FLY: Future Communications Laboratory, Department of Electrical Engineering and Information Technology (DIETI), University of Naples Federico II, Naples, Italy e-mail: [email protected]; [email protected] https://www.quantuminternet.it © The Author(s), under exclusive license to Springer Nature Switzerland AG 2023 M. S. Greco et al. (eds.), Women in Telecommunications, Women in Engineering and Science, https://doi.org/10.1007/978-3-031-21975-7_13

357

358

J. Illiano and A. S. Cacciapuoti

although it is possible to directly exchange quantum information through a quantum information carrier propagating through a physical channel, the quantum noise, which is fundamentally different from the classical noise, can irreversibly damage the quantum information during transmission (Cacciapuoti and Caleffi 2019). In order to understand the nature and the rationale of new critical challenges related to the Quantum Internet design, in the following, we first provide some preliminaries to quantum information. Then, we introduce quantum teleportation, the key process exploiting entanglement for exchanging quantum information between two remote nodes, without the physical transfer of the particle encoding the information. In this context, it will also be underlined that quantum teleporting, together with quantum repeaters (Muralidharan et al. 2016), represent promising strategies for extending the communication range of a quantum network. But, different from classical communications, once the quantum teleportation process is completed, the entanglement—and, therefore, the corresponding virtual quantum communication channel enabled by it—is destroyed. As a consequence, a critical task of any quantum network is represented by the entanglement creation and distribution, and—being entanglement the fundamental resource for connectivity in quantum networks—it is worth to further discuss the different types of entanglement. Given the aim of providing the reader with a general but effective overview of quantum communications—and the role played by entanglement thereupon— noise effects cannot be ignored for an effective design of quantum communication protocols. Therefore, a preliminary analysis of the quality of the entanglement generated on a real platform is essential. For this, we conclude the chapter with an experimental verification of the generation of a fundamental entanglement resource, namely, the GHZ state (Dür et al. 2000). Specifically, we first generate this state through two different processors made available by the IBM Q-eXperience platform (IBM Quantum eXperience xxxx); then we evaluate the mismatch—namely, the generation noise—between the ideal state and the experimentally generated state through a key figure of merit, the so-called fidelity.

13.2 Quantum Information Preliminaries In this paragraph, we introduce some preliminaries needed to understand the peculiar features of the quantum communications field, whereas for a comprehensive analysis we refer the reader to Cacciapuoti et al. (2020). A quantum bit (qubit) is the simplest quantum mechanical system. Its state can be represented by a complex vector in a two-dimensional Hilbert space, spanned by two orthogonal states: zero (or ground) state and one (or excited) state (Nielsen and Chuang 2002). In fact, according to one of the quantum mechanics postulates, every closed or isolated quantum system is associated with a complex Hilbert space, which is equivalent to a vector space with an inner product for finite-dimensional systems (Nielsen and Chuang 2002). The system is fully described by its state vector, which is a unit

13 On the Entanglement Role for the Quantum Internet

359

vector belonging to this complex vector space, called the system state space.1 The orthonormal basis commonly used for one-qubit state space is the following:     1 0 . |0 ≡ , |1 ≡ , 0 1

(13.1)

which is also referred to as Z-basis or computational basis. Once a basis has been chosen, any qubit state .|ψ can be expressed as a linear combination of the two basis vectors, namely as a superposition2 of the basis vectors: .

|ψ = α |0 + β |1 .

(13.2)

The weights .α and .β of the linear combination in Eq. (13.2) are two complex numbers, called amplitude of .|ψ. These weights should satisfy the normalization condition: |α|2 + |β|2 = 1,

.

(13.3)

being .|ψ a unit vector, i.e., .ψ| |ψ = 1. Indeed, .α and .β are correlated to the probabilities that the measurement outcomes of the qubit are .|0 and .|1, respectively. However, according to the measurement postulate, measuring the qubit irreversibly affects the qubit and destroys quantum superposition. Specifically, by measuring the qubit .|ψ = α |0 + β |1 in the computational basis, then with probability .|α|2 the measurement outcome will be .|0, and with probability .|β|2 , the measurement outcome will be .|1. Furthermore, assuming as instance the measurement outcome being .|0, if a second measurement in the computational basis is performed on the same qubit, we will observe .|0 with probability 1. Therefore, the measurement of a quantum state irremediably alters the state of the system, and the measurement result is a probabilistic result. The above can be properly generalized to composite quantum systems, namely systems constituted by multi-qubits. In this case, the Hilbert space .H associated to a composite quantum system is the result of the tensor product of Hilbert spaces associated to the component subsystems: H = H1 ⊗ H2 ⊗ . . . ⊗ Hk ,

.

1 The

(13.4)

notation commonly used to describe a quantum state is the Dirac notation, also referred to as bra–ket notation. According to such a notation, the symbol .|.—called ket—denotes a column vector, while the symbol ..|—called bra—denotes the transposed conjugate vector of the corresponding ket. 2 The superposition depends on the chosen basis. Let us consider, as example, the basis, also called √ √ X-basis or Hadamard basis, constituted by the two vectors .|+ = |0+|1 and .|− = |0−|1 . 2 2 Clearly, the state .|+ is in a superposition with respect to the computational basis, but not with respect to the Hadamard basis.

360

J. Illiano and A. S. Cacciapuoti

where the dimensionality of .H is .dim(H) = dim(H1 )dim(H2 ) . . . dim(Hk ) (Nielsen and Chuang 2002). As instance, let us consider a two-qubit system. The computational basis for the composite system is given by the set: .{|00 , |01 , |10 , |11}. Hence, any state of the composite system can be represented by a linear combination of the basis as follows: ⎡ ⎤ α0 ⎢α1 ⎥ ⎥ . |ψ = α0 |00 + α1 |01 + α2 |10 + α3 |11 ≡ ⎢ ⎣α2 ⎦ . α3

(13.5)

Within the state space of two-qubit systems, a state .|ψ is called separable if it can be expressed as tensor product of single-qubit states. Hence, for a two-qubit system, a separable state can be written in the form: .

|ψ = |i ⊗ |j  , with |i ∈ H1 and |j  ∈ H2 .

(13.6)

Every state .|ψ that cannot be written in the form of Eq. (13.6) is referred to as entangled state. A remarkable example of two-qubit entangled states is given by the so-called Bell states or EPR pairs (Bohm and Aharonov 1957): .

|00 + |11 |00 − |11 |10 + |01 , |Φ −  = , |Ψ +  = , √ √ √ 2 2 2 |10 − |01 |Ψ −  = . √ 2 |Φ +  =

(13.7)

Physically, entanglement is a quantum property with no counterpart in the classical world. When two qubits are entangled, they exist in a shared state, such that any action on one qubit affects instantaneously the other qubit as well, regardless of the distance. To better understand the aforementioned statement, let us consider the state .|Φ + , and let us perform independent measurements on each qubit of this entangled pair. When we measure the first qubit according to the canonical basis, the observed measurement outcomes are .|0 with probability . 12 and .|1 with probability 1 . . Surprisingly, after the measurement of the first qubit, the state of the second qubit 2 is deterministically fixed. Specifically, the independent measurement of each qubit gives an uniform distribution of zero and one. Nevertheless, by comparing the two measurement results, the two corresponding measurement outcomes are correlated (Cacciapuoti et al. 2020). And, measuring one qubit instantaneously changes the state of the second qubit. This property holds even if the two entangled particles are far apart. Another remarkable difference between quantum information and classical information is given by the no cloning theorem, which states that it is impossible to copy an unknown quantum state. So any communication protocol based on

13 On the Entanglement Role for the Quantum Internet

361

the information retransmission is useless for quantum communications. Moreover, any quantum system inevitably interacts with the environment. These interactions irreversibly affect any quantum state by the process of decoherence (Cacciapuoti and Caleffi 2019). This kind of quantum noise has no direct counterpart in the classical world. In conclusion, superposition, entanglement, no cloning theorem, and decoherence are phenomena with no counterpart in classical networks that impose very challenging constraints for the network design. Indeed, although it is possible to directly transmit a qubit to a remote node—for example, through a photon on optical fiber—the transmitted quantum information can be irreversibly lost due to attenuation or corrupted by the decoherence phenomenon. Furthermore, this information cannot be copied and retransmitted nor recovered through measurement operations. Therefore, an effort to design new communication protocols accounting quantum mechanics peculiarities is needed. Surprisingly, quantum mechanics also gives us the possibility to face these challenges exploiting entanglement. In this regard, in the following, we will present the quantum teleportation process, an example of communication strategy that exploits entanglement for the “transmission” of unknown qubits in the absence of the physical transmission of the particle encoding the information.

13.3 Quantum Teleportation Quantum teleportation is a surprising strategy enabling the “transmission” of qubits without the physical transfer of the particle storing the qubit. Specifically, by sharing an EPR pair between the source and the destination, by performing local quantum operations, and by exchanging classical information, it is possible to reconstruct the desired quantum state at the destination (Cacciapuoti et al. 2020). As represented in Fig. 13.1, Alice is the node interested in transmitting the qubit and Bob is the receiving node. Suppose that these two nodes share an EPR pair—let say .|Φ + —and Alice wants to transmit the unknown qubit .|ψ = α |0 + β |1 to Fig. 13.1 Schematic representation of the quantum teleportation process. Figure taken from Cacciapuoti et al. (2020)

362

J. Illiano and A. S. Cacciapuoti

Bob. Hence, the initial state of the composite quantum system given by Alice and Bob is given by .

|ϕ1  = |ψA1 ⊗ |Φ + A2 B =

  1

= √ α|0A1 |00 + |11 A B + β|1A1 |00 + |11 A B = 2 2 2  1  = √ α |000 + α |011 + β |100 + β |111 , (13.8) A1 A2 B 2

where Alice holds, from left to right, the first qubit .A1 and the second qubit .A2 , while Bob holds the third qubit B.3 The first step of the process is given by a CNOT operation between .A1 and .A2 , with .A1 as controller qubit and .A2 as target qubit. The state of the system after the first step is .|ϕ2 : .

1 |ϕ2  = √ (α |000 + α |011 + β |110 + β |101). 2

(13.9)

Then Alice performs an Hadamard gate on the qubit .A1 , and the system evolves into state .|ϕ3 . By properly gathering the two qubits .A1 and .A2 belonging to Alice, .|ϕ3  is equivalent to .

|ϕ3  =

  1

|00A1 A2 ⊗ α |0 + β |1 B + |01A1 A2 ⊗ α |1 + β |0 B 2   + |10A1 A2 ⊗ α |0 − β |1 B + |11A1 A2 ⊗ α |1 − β |0 B . (13.10)

Then, Alice performs a joint measurement on .A1 and .A2 , by obtaining, with same probability equal to . 41 , one of the four possible states. The above described sequence of operations is also known as Bell State Measurement (BSM). The state of the qubit B depends on the result of the measurement carried out by Alice. Such a result can be communicated to Bob using two bits through a classic communication channel. Once the two classic bits have been received, Bob is able to reconstruct the information qubit by performing some post-processing operations. Table 13.1 shows the decoding operation to be performed at the receiver Bob, depending on the received bits. From the above description, it follows that quantum teleportation requires: (i) to generate and distribute an EPR pair between source and destination, (ii) to perform quantum pre-processing operations at the source, namely, a BSM, (iii) a classical

3 For the sake of simplicity, we will, from now on, assume that the assignment of the qubits is fixed, and we will omit the notation .A1 , A2 , B.

13 On the Entanglement Role for the Quantum Internet Table 13.1 Post-processing operations for the quantum teleportation process

Measurement results 00 01 10 11

363 State of qubit B .α |0 + β |1 .α |1 + β |0 .α |0 − β |1 .α |1 − β |0

Decoding operation I X Z Y

communication resource (for transmitting the two classical bits), and finally (iv) to perform quantum post-processing operations at the destination. Indeed, the correct generation of entangled pairs and their distribution to the involved nodes are crucial for transmitting the information, not limited to the quantum teleportation process. This in turn calls for a new concept of connectivity in the quantum communication field: the entanglement-based connectivity (Illiano et al. 2022). This new nature of connectivity leads to important considerations. First, it should be noted that, after the teleportation process,4 the entanglement between Alice’s particle and Bob’s is destroyed. As a consequence, if Alice wants to transmit a second information qubit, the entanglement-based quantum channel has to be reconstructed. In other words, the entanglement has to be re-generated and redistributed after the conclusion of each teleportation process. Hence, it follows that entanglement generation represents the preliminary functionality for enabling quantum communications. Indeed, without an effective entanglement generation, the entire process would be unreliable. Therefore, being able to verify the quality of the generated entangled states is crucial. A quantity that can be used to this aim is the fidelity, a measure of the mismatch between two quantum states. Formally, the expression of the fidelity is as follows:  √ √ 2 .F (ρ1 , ρ2 ) = T r ρ1 ρ2 ρ1 (13.11) with .ρ1 and .ρ2 denoting the density matrices (Nielsen and Chuang 2002) of two quantum states. Hence, with .F (ρ1 , ρ2 ), we evaluate how “close” the quantum states with density matrices given by .ρ1 and .ρ2 are, with .F (ρ1 , ρ2 ) ∈ [0, 1] achieving the maximum when the two quantum states are indistinguishable. It must be observed that the distribution of the entanglement has a deep impact on the concept of connectivity as well. In fact, the entanglement distribution is subject to decoherence effects, and the distribution rate decays exponentially with the distance between the source and the destination. Luckily, entanglement distribution over longer distances can be achieved through quantum repeaters, namely, devices

4 The quantum teleportation is an example of quantum communication protocol that exploits entanglement. However, the considerations made for the connectivity also hold for any other protocol that uses entanglement.

364

J. Illiano and A. S. Cacciapuoti

Fig. 13.2 Schematic representation of the quantum repeater functioning. The quantum repeater is an intermediate node between the source (Alice) and the destination (Bob). At first, the end nodes share each an entangled pair with the quantum repeater. Within the figure, the entangled qubits shared between Alice and the quantum repeater are represented as nodes with diagonal striped frame interconnected by a dotted line, whereas the entangled qubits shared between Bob and the quantum repeater are represented as nodes with anti-diagonal striped frame interconnected by a dotted line as well. By performing a BSM operation on the qubits at the quantum repeater, the entanglement is swapped, and an end-to-end entanglement is generated between Alice and Bob

implementing the physical process called entanglement swapping.5 As represented in Fig. 13.2, a quantum repeater is used as intermediary node between source and destination. It is based on the strategy of dividing the distance between source and destination into smaller links and distributing the entanglement over the individual sub-links. Through BSM operations at intermediate nodes, the entanglement is eventually distributed between source and destination. As already mentioned, the BSM operation destroys the original entanglement. Hence, once the entanglement swapping process is completed, the entanglement over the individual sub-links is consumed for generating end-to-end entanglement between the two end nodes.

5 The example reported in Fig. 13.2 refers to the so-called first generation of quantum repeaters. We

refer the reader to Muralidharan et al. (2016) for an in-depth introduction to the different repeaters generation.

13 On the Entanglement Role for the Quantum Internet

365

13.4 Beyond Bipartite Entanglement So far, we considered only bipartite entangled states, namely, states of a two-qubit system. However, also higher-dimensional systems, composed of more than two qubits, admit entangled states. Generally, given the n-qubit state .|ψ ∈ H, the state .|ψ is separable with respect to the .H-decomposition .H = H1 ⊗ H2 ⊗ . . . ⊗ Hk if it can be written as follows: .

|ψ = |ψ1  ⊗ . . . ⊗ |ψk  , with |ψi  ∈ Hi .

(13.12)

Clearly, whether it should not be possible to express .|ψ as a tensor product of states belonging to the subsystems .{Hi }, then .|ψ is an entangled state with respect to the considered decomposition. The study of multipartite entanglement is challenging, and it requires a specific mathematical framework that becomes more complex as the number of qubits increases. Indeed, for two-qubit systems, we can distinguish between two different classes of states, namely, unentangled states and entangled states. Differently, when considering quantum systems with more than two qubits, there exist several classes of entangled states with different properties. The simplest multipartite system is a tripartite system. Given three qubits, we can individuate three different classes of states. Specifically, these three qubits can exist in an unentangled state (separable state), in a biseparable state or in a genuinely entangled state (Dür et al. 2000; Rieffel and Polak 2011). From this, it results that even for the simplest multipartite system the classification is not trivial. When it comes to n-partite systems, there could be as many classes of multipartite entangled states as the number of partitions of a set of n elements, called the Bell’s number (Comtet 2012). Hence, if a different class of entangled states were associated with each partition, the number of classes for an n-qubit system would be n    n .

k=1

k

− 1,

(13.13)

where we subtract one to account for the partition corresponding to unentangled states. Indeed, it is not assured that each partition is associated with a class of entangled states. Some partitions may not allow entangled states, or conversely, more classes may be generated from the same partition. In this context, several classification criteria have been proposed (Dür et al. 1999; Rigolin et al. 2006; Lamata et al. 2006), but the research is still ongoing. In the following, we will focus on classification criteria and entanglement measures for tripartite entanglement, for which the knowledge is well-established. To this aim, the set of local operations and classical communications (LOCC) is used for defining equivalent entangled states. Specifically, two quantum states .|ϕ and .|φ are LOCC-equivalent if .|ϕ can be converted into .|φ via LOCC and vice

366

J. Illiano and A. S. Cacciapuoti

Fig. 13.3 Hierarchy of tripartite entangled class. The arrows represent allowed transformations between different classes through SLOCC. Classes in the same row are nonequivalent SLOCC classes

versa. This criterion leads to the definition of an infinite number of classes, as it is parameterized by a continuous variable.6 A finite number of equivalent classes for three-qubit states are obtained by relaxing the LOCC condition. More into details, a state .|ϕ can be converted into .|ψ by means of stochastic local operations and classical communications (SLOCC) if there exists a sequence of local operations with classical communications that maps .|ϕ into .|ψ with non-zero probability. In such a case, .|ϕ and .|ψ are defined SLOCC-equivalent. Accordingly, six classes for tripartite systems arise: separable states, three classes of biseparable states, and two classes of genuinely tripartite states. These classes can be organized into a hierarchy, as shown in Fig. 13.3, originally proposed in Rieffel and Polak (2011). At the top of the hierarchy, one can find the two classes of genuinely maximally entangled tripartite states, namely GH Z SLOCC-equivalent states, states SLOCC-equivalent to .

1  |GH Z = √ |000 + |111 , 2

and W SLOCC-equivalent states, states SLOCC-equivalent to 6 For

a comprehensive analysis, we refer the reader to Rieffel and Polak (2011).

(13.14)

13 On the Entanglement Role for the Quantum Internet

.

1 |W  = √ (|001 + |010 + |100 . 3

367

(13.15)

Such states belong to the same level of the hierarchy, but they do not belong to the same class, since there does not exist a sequence of local operations and classical communications (either stochastic or deterministic) transforming a GH Z state into a W state (Dür et al. 2000), or vice versa. Differently, a biseparable state can be obtained from a tripartite entangled state. Hence, within the hierarchy, the biseparable states are presented below GHZ and W states. Then, a separable state can be obtained through SLOCC from biseparable states. Furthermore, measuring a qubit of a genuine tripartite state leads to different states according to the SLOCC-equivalent class. Indeed, if we measure one of the qubits of a GHZ state in the Z-basis, the resulting state is unentangled, regardless of the measurement output. Conversely, if we measure one of the qubits of a W state, one obtains a different result with respect to the GHZ state. Specifically, if the measurement result is 1, the remaining qubits collapse into the separable state .|00, whereas when the measurement result is 0, the remaining qubits collapse into the state . √1 (|01 + |10) = |Ψ + , which is one of the four Bell states. This feature is 2 expressed by saying that .|W  states are more persistent than .|GH Z states. The persistency propriety—i.e., the minimum number of qubits that, when measured, reduces the state to an unentangled state regardless of the measurement result—indirectly quantifies, in terms of the number of qubits to be measured, the operational effort needed to fully destroy the entanglement (Briegel et al. 2001). Interestingly, if we measure a qubit—say the third qubit—of a GHZ state in the X-basis, by observing that can be rewritten as .

|GH Z =

1 1 |00 + |11 ⊗ |+ + |00 − |11 ⊗ |− , 2 2

(13.16)

it results that, regardless of the measurement output, the remaining qubits end up in a maximally entangled bipartite state. Conversely, for a W state, we obtain a maximally entangled bipartite state with probability . 13 . For this reason, GHZ states are said to be more entangled than W states in terms of maximally connectedness property. Specifically, a multipartite state .|φ is said to be maximally connected if, for any two qubits, there exists a sequence of single-qubit measurements on the other qubits assuring the two qubits end up in a maximally entangled state. Given the diversity of the two genuinely entangled tripartite classes, it is not possible to uniquely establish which classes of states have the major amount of entanglement; however, the study of these states shows that both can be useful for quantum communications. Indeed, multipartite entanglement has a crucial impact on the design of an entanglement-based quantum network. For the sake of exemplification, let us consider three nodes—say Alice, Bob, and Charlie—of a quantum network with only one qubit available for communication at each node. In an EPR-based network, only after having selected the pair source–destination, an EPR pair can be generated

368

J. Illiano and A. S. Cacciapuoti

and distributed for teleporting the desired qubit. Conversely, with multipartite entanglement, Alice, Bob, and Charlie can pre-share a GHZ state as entanglement resource, and then Alice can dynamically select its destination (Ghosh et al. 2002). This is a trivial yet illustrative example of the possibilities arising from multipartite entanglement.

13.5 Generating Entangled States on Real Devices When it comes to realistic scenarios for the design of entanglement-based quantum communication protocol, it is not possible to ignore the effects of quantum noise arising with the real devices. The decoherence corrupts the quantum state and degrades its quality (Cacciapuoti and Caleffi 2019; Cacciapuoti et al. 2020). Hence, it is crucial to evaluate the fidelity of the generated entangled states, since it deeply affects the performance of any communication protocol. To this aim, we experimentally generate GHZ states through IBM Q-eXperience (IBM Quantum eXperience xxxx), the IBM quantum platform that allows remote programming of quantum computers via cloud. The purpose of the experimentation is to generate GHZ states on different quantum processors and to evaluate the corresponding fidelity. This is achieved through the state tomography process.

13.5.1 State Tomography The goal is to evaluate the output state of the circuit that generates the entangled state .|GH Z. According to the measurement postulate, measuring a quantum state means projecting the state into one of the measurement basis. However, we want to evaluate the amplitudes of the state, meaning .α and .β in Eq. (13.2). The quantum state tomography process aims at reconstructing a description of the density matrix .ρ of a quantum state. Specifically, by performing repeated measurement of a tomographically complete set over multiple copies of the output of the same circuit, we obtain different measurements of a state described by the same density matrix. Given a circuit that prepares the system in a certain state, the state tomography process allows to reconstruct a description of the density matrix .ρ of the current state of the system, where with “current” we mean the quantum state immediately before the measurement. Clearly, it is not possible to characterize a given state, represented with the density matrix .ρ, having only a single copy of it available. Instead, it is necessary to prepare several copies of the same state in order to obtain an estimation of .ρ. Suppose we have several copies of the density matrix .ρ of a qubit. Any density matrix .ρ, representative of the state of a single qubit, can be represented by four parameters .S0 , Sx , Sy , Sz as follows (Nielsen and Chuang 2002; Altepeter et al. 2004):

13 On the Entanglement Role for the Quantum Internet

ρ=

.

1 Si σi , 2

369

(13.17)

i

with .{σi }0,x,y,z denoting the Pauli matrices, and .Si ≡ T r(σi ρ). Each .Si physically corresponds to a measurement result with respect to three different bases7 (Altepeter et al. 2004). Through the combination of these sets of measurements, we can estimate the probability distribution of the quantum state with respect to each measurement basis and hence the density matrix .ρ. It is important to note that the computational complexity of state tomography grows exponentially. In fact, three measurement circuits are required to perform state tomography for a single qubit, whereas .3n measurements circuits are required for state tomography of an n-qubit state.

13.5.2 GHZ State Starting from the three-qubit quantum state .|000, the tripartite maximally entangled state GHZ given in Eq. (13.14) is obtained by performing an Hadamard gate on the first qubit and then two CNOT gates, one with the first qubit as controller and the second qubit as target, and the other with the second qubit as controller and the third qubit as target, as illustrated in Fig. 13.4. Through the experiments, we generated the GHZ state with two different quantum processors, namely, ibmqx._manila shown in Fig. 13.5 and ibmq._quito shown in Fig. 13.6. Multiple state tomography procedures were performed for each processor. As already mentioned, each state tomography process is based on the repetition of a set of .3n = 27 measuring circuits. The number of experiments for each measuring circuit, also called experiment shots, is set to 8192, the maximum shots value allowed. The output of the state tomography has been processed through the least squares method (Axelsson 1987) to obtain an estimation of the density matrix of the GHZ state. Then we evaluate the fidelity by using the estimation of the density matrix obtained from the state tomography and the ideal density matrix of the three-qubit GH Z state: Fig. 13.4 Quantum circuit for generating a GHZ state, when the qubits are all initialized into the .|0 state

7 The

measurement required to compute .S0 is the same needed to compute .Sz .

370

J. Illiano and A. S. Cacciapuoti

5 Qubits

32

0

1

2

3

4

QV

2.8K

Processor Type:

Falcon r5.11L

Version:

1.0.21

Basis gates:

CX, ID, RZ, SX, X

Avg. CNOT Error:

1.217e-2

Avg. Readout Error:

3.234e-2

Avg. T1:

164.08 us

Avg. T2:

60.07 us

CLOPS

Fig. 13.5 Reproduction of specification values and topology diagram of the IBM chip named ibmq_manila chip. The topology diagram indicates the pairs of qubits that support two-qubit gate operations between them. Qubits are represented as circles, and the supported two-qubit gate operations are displayed as lines connecting the qubits

0

5

1

Qubits

16

3

QV

2.5K CLOPS

2

Processor Type:

Falcon r4T

Version:

1.1.19

Basis gates:

CX, ID, RZ, SX, X

Avg. CNOT Error:

1.117e-2

Avg. Readout Error:

3.068e-2

Avg. T1:

74.32 us

Avg. T2:

108.33 us

4

Fig. 13.6 Reproduction of specification values and topology diagram of the IBM chip named ibmq_quito chip. The topology diagram indicates the pairs of qubits that support two-qubit gate operations between them. Qubits are represented as circles, and the supported two-qubit gate operations are displayed as lines connecting the qubits

ρGH Z

.

⎡ 1  1 ⎢0 1 |000 + |111 000| + 111| = ⎢ = 2 2 ⎣0 1

0 0 0 0

0 0 0 0

⎤ 1 0⎥ ⎥. 0⎦

(13.18)

1

This matrix has 8 rows and 8 columns with real elements. In fact, the imaginary part, represented on the right in Fig. 13.7, has all null elements. The real part has only 4 non-null elements equal to . 12 . In the table shown in Fig. 13.8, each column corresponds to a different quantum processor, whereas each row corresponds to a value of fidelity obtained by the above procedure. As expected, the density matrix of the actual state of the system does not coincide perfectly with the theoretical result, as can be seen from the levels of the values of the real and imaginary parts represented in Figs. 13.7 and 13.9. This happens due to the noisy phenomenon of decoherence that affects open quantum systems such as real quantum devices. The state interacts with the surrounding

13 On the Entanglement Role for the Quantum Internet

371

Fig. 13.7 Representation of the density matrix of the GHZ state given in Eq. (13.18). The real part (left) of the .8 × 8 matrix has only 4 non-null elements equal to . 12 , whereas the imaginary part (right) has all null elements

Fig. 13.8 Table of fidelity values. The first column shows the fidelity values relating to the ibmqx_quito device, and the second those relating to ibmq_manila. Each row corresponds to the least square estimation resulting from the execution of a state tomography of 8192 iterations

environment and undergoes degradation. As it can be noted, the average values for the two devices are roughly 78% and 77%, respectively. The use of two different devices does not seem to have any effect on fidelity. In fact, the two devices are based on the same technology; moreover, the properties of the systems, such as the error gate and the implemented gates, do not change significantly between the two devices. More importantly, they have the same topology.

372

J. Illiano and A. S. Cacciapuoti

Fig. 13.9 Example of real part (left) and imaginary part (right) of the estimation through quantum state tomography of the density matrix of a .|GH Z state generated by a circuit on a real device

The experiment involved qubits 0, 1, and 2 in both the chips. And it can be observed from the specifications in Figs. 13.5 and 13.6, the links between the processed qubits are the same for the two devices, i.e., .q0 is directly linked to .q1 and .q1 is directly linked to .q2 . This is crucial because if they are directly linked through the topology diagram, also called coupling map or connectivity, then they support two-qubit gates. Differently, if they were not directly connected, it would not be possible to directly execute CNOT operations (Ferrari et al. 2021). As a consequence, we would observe an higher gate error and hence a lower fidelity. In conclusion, given experimental values of fidelity, we can state that the generation itself is not completely unreliable for any protocol. Formally, it is required that for any real quantum state the target fidelity value is above 50% (Horodecki and Horodecki 1999). However, the experimentation carried out is a crucial starting point for the design of communication protocols. Indeed, starting from the dataset provided, several elaborations can be carried out, for example, evaluating confidence regions or more specific statistics for a particular protocol.

13.6 Conclusion In this chapter, we explored some aspects of entanglement as a key resource for quantum communications. In Sect. 13.3, we explained how to build quantum communication channels by exploiting a pair of maximally entangled particles. In order to broaden the point of view on the entanglement resource, which is key for the network connectivity, the study continued with the analysis of entanglement for tripartite systems. Specifically, the genuinely entangled GHZ state was considered. This was experimentally reproduced via the IBM Q-eXperience platform on two different real devices. The purpose of the experiments is to evaluate the deviation between the ideally expected state and that actually generated one on real quantum

13 On the Entanglement Role for the Quantum Internet

373

processors. The evaluation is carried out by applying the state tomography procedure and then calculating the fidelity parameter. From the experimental results, it is possible to state that the effect of quantum noise does not completely degrade the state.

References Altepeter JB, James DFV, Kwiat PG (2004) 4 qubit quantum state tomography. Quantum state estimation. Springer, Berlin, pp 113–145 Axelsson O (1987) A generalized conjugate gradient, least square method. Numer Math 51(2):209– 227 Bohm D, Aharonov Y (1957) Discussion of experimental proof for the paradox of Einstein, Rosen, and Podolsky. Phys. Rev. 108(4):1070 Briegel, Hans J, Raussendorf R (2001) Persistent entanglement in arrays of interacting particles. Phys Rev Lett 86(5):910 Cacciapuoti AS, Caleffi M (2019) Toward the Quantum Internet: A directional-dependent noise model for quantum signal processing. In: ICASSP 2019-2019 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP). IEEE, New York Cacciapuoti AS et al (2019) Quantum Internet: networking challenges in distributed quantum computing. IEEE Netw 34(1):137–143 Cacciapuoti AS et al (2020) When entanglement meets classical communications: quantum teleportation for the Quantum Internet. IEEE Trans Commun 68(6):3808–3833 Comtet L (2012) Advanced combinatorics: the art of finite and infinite expansions. Springer, Berlin Cuomo D, Caleffi M, Cacciapuoti AS (2020) Towards a distributed quantum computing ecosystem. IET Quantum Communication 1(1):3–8 Dür W, Cirac JI, Tarrach R (1999) Separability and distillability of multiparticle quantum systems. Phys Rev Lett 83(17):3562 Dür W, Vidal G, Ignacio Cirac J (2000) Three qubits can be entangled in two inequivalent ways. Phys Rev A 62(6):062314 Ekert AK (1991) Quantum cryptography based on Bell’s theorem. Phys Rev Lett 67(6):661 Ferrari D, Cacciapuoti AS, Amoretti M, Caleffi M (2021) Compiler Design for Distributed Quantum Computing. In: IEEE Transactions on Quantum Engineering, vol 2, pp 1–20, Art no. 4100720. https://doi.org/10.1109/TQE.2021.3053921 Ghosh S et al (2002) Entanglement teleportation through GHZ-class states. New J Phys 4(1):48 Gottesman D, Lo H-K (2003) Proof of security of quantum key distribution with two-way classical communications. IEEE Trans Inf Theory 49(2):457–475 Horodecki M, Horodecki P (1999) Reduction criterion of separability and limits for a class of distillation protocols. Phys Rev A 59(6):4206 IBM Quantum eXperience. https://quantum-computing.ibm.com Illiano, Jessica, et al (2022) Quantum internet protocol stack: A comprehensive survey. Computer Networks, 109092 Kimble HJ (2008) The Quantum Internet. Nature 453(7198):1023–1030 Lamata L et al (2006) Inductive classification of multipartite entanglement under stochastic local operations and classical communication. Phys Rev A 74(5):052336 Lo H-K, Curty M, Tamaki K (2014) Secure quantum key distribution. Nat Photonics 8(8):595–604 Muralidharan S et al (2016) Optimal architectures for long distance quantum communication. Sci Rep 6(1):1–10 Nielsen MA, Chuang I (2002) Quantum computation and quantum information Pirandola S, Braunstein SL (2016) Physics: unite to build a Quantum Internet. Nature News 532(7598):169

374

J. Illiano and A. S. Cacciapuoti

Rieffel EG, Polak WH (2011) Quantum computing: a gentle introduction. MIT Press, New York Rigolin G, de Oliveira TR, de Oliveira MC (2006) Operational classification and quantification of multipartite entangled states. Phys Rev A 74(2):022314 Wehner S, Elkouss D, Hanson R (2018) Quantum Internet: A vision for the road ahead. Science 362:6412

Jessica Illiano is a Ph.D. student in Information Technologies and Electrical Engineering at University of Naples Federico II. In 2020 she was winner of the scholarship “Quantum Communication Protocols for Quantum Security and Quantum Internet” fully funded by TIM S.p.A. Since 2018, she is a member of the https://www.quantuminternet.it/ Research Group, FLY: Future Communications Laboratory. In 2018 she earned a bachelor’s degree and then in 2020 a master’s degree in Telecommunications Engineering, both summa cum laude from University of Naples Federico II. She has always been curious about understanding the true functioning of the complex world of scientific phenomena and the new technologies that surrounded her. Her first approach with quantum technologies occurred in 2018 by attending a seminar organized by Prof. Angela Sara Cacciapuoti and Prof. Marcello Caleffi. Starting with this seminar she understood that when it comes to quantum communications a strong effort is needed to design a network that accounts for different underlying physical mechanisms. Jessica was deeply involved with the innovative perspective and the interdisciplinary open problems arising from this subject. Hence, she decided to continue her studies in that direction through her thesis at first, and then through her Ph.D. program by joining the quantum internet research group.

Angela Sara Cacciapuoti is Associate Professor at the University of Naples Federico II (Italy). Since July 2018 she held the national habilitation as “Full Professor” in Telecommunications Engineering. Her work has appeared in first tier IEEE journals and she has received different awards and recognitions, including the “2021 N2Women: Stars in Networking and Communications”. She is a IEEE ComSoc Distinguished Lecturer for the class of 2022-2023! Angela Sara currently serves as Area Editor for IEEE Communications Letters, and as Editor/Associate Editor for the journals: IEEE Trans. on Communications, IEEE Trans. on Wireless Communications, IEEE Trans. on Quantum Engineering, IEEE Network. She was also the recipient of the 2017 Exemplary Editor Award of the IEEE Communications Letters. From 2020 to 2021, Angela Sara was the Vice-Chair of the IEEE ComSoc Women in Communications Engineering (WICE). Previously, she has been appointed as Publicity Chair of WICE. In 2016 she has been an appointed member of the IEEE ComSoc Young Professionals Standing Committee. From 2017 to 2020, she has been the Treasurer of the IEEE Women in Engineering (WIE) Affinity Group of the IEEE Italy Section. Her current research interests are mainly in quantum communications, quantum networks and quantum information processing.

Chapter 14

Space Sustainability: Toward the Future of Connectivity Ernestina Cianca and Marina Ruggieri

14.1 Introduction In the past years, humanity has faced an unprecedented situation of the globalization era due to the COVID19 pandemia. We have lost both stability and perspectives. We have been forced to slow down and even to stop locally and globally. Hopefully, many of us have taken the opportunity to rethink the apparent balance of our everyday life and of the whole society as well as the mistakes we are playing in the last chance to save our planet and, thus, our future. In the long days of lock down, we have also experienced the importance of connectivity to maintain the capability of working, socializing, and managing both common and emergency needs. Therefore, connectivity and sensitivity to the environment are pillars of the lesson learned in this pandemic. However, the special relationship between connectivity and sustainability goes much further. In fact, an intelligent approach to the connectivity infrastructures and vertical applications can be the best ally to the monitoring and management of sustainability matter. On the other hand, sustainability can be the major guideline to a conceptual change in the design approach of connectivity matter, with amazing advantages not only for the environment but also for the flexibility, pervasiveness, and endurance of connectivity systems. The above profound relationship between connectivity and sustainability is key not only on the terrestrial domain but in every domain where connectivity can be

E. Cianca () · M. Ruggieri University of Rome “Tor Vergata”, Rome, Italy e-mail: [email protected]; [email protected] © The Author(s), under exclusive license to Springer Nature Switzerland AG 2023 M. S. Greco et al. (eds.), Women in Telecommunications, Women in Engineering and Science, https://doi.org/10.1007/978-3-031-21975-7_14

375

376

E. Cianca and M. Ruggieri

deployed (space, air, underwater), and, at the same time, there is an environment to be respected and protected. Space is the domain this chapter focuses on. In particular, in Sect. 14.2, the holistic nature of sustainability and the related consequences on connectivity will be described. In Sect. 14.3, some current trends on space connectivity are highlighted also relating them to pros and cons for the sustainability of the space domain. In Sect. 14.4, the guidelines for a sustainable-prone approach to connectivity in space are provided along with the major technologies which are allies of space sustainability. In Sect. 14.5 conclusions are drawn, identifying some commitments for space sustainability that can pave the way to a proper future of space.

14.2 Holistic Approach to Sustainability The good or bad news about sensitivity to the environment is that a systems-ofsystems-like approach is needed on one hand to achieve a broad awareness about the issues and on the other hand to identify valuable and effective solutions to fix the issues themselves. Engineering the approach to sustainability would be a very interesting starting point to deal with the matter, but obviously humanity cannot be composed only of engineers! Nonetheless, a systems-of-systems-like approach can be embraced and applied by a much larger population than just engineers. What is necessary is the awareness that any element is the piece of a complex mosaic and the interdependence between the elements and the mosaic is key to success for both the elements and the mosaic. Curiosity, sensitivity, flexibility, and genuine desire to improve are the pillars to develop a systems-of-systems-like vision that, in turns, is key to be or to become an active enabler of sustainability. If we deal with the sustainability of our planet, the systems-of-systems approach implies that all continents are linked through an invisible chain and only if all of them behave properly with the environment the sustainability balance will be met (Ruggieri 2020). The effects of global warming and of the pandemic have shown that globalization is a tremendous accelerator of issues happening wherever in the planet. Obviously, this acceleration applies also to the virtuous behaviors. In this planet-centric vision of sustainability, Earth is the mosaic, and every citizen, every family, and every country is the element of the mosaic. The above frame would be enough to absorb the holistic nature of sustainability. The increasing awareness on the consequences of global warming and the many initiatives that are flourishing around the world should make us not too pessimistic about the capability of paving a decent future for humanity.

14 Space Sustainability: Toward the Future of Connectivity

377

However, our planet is not an isolated system, and its sustainability is not enough to guarantee the holistic sustainability of the systems-of-systems it belongs to. Earth is the element of a larger mosaic. This mosaic is space. The attention we paid so far to the sustainability of space is very unsatisfactory with respect to the number of years occurred since the first launch. Space has to be approached with respect, but this is not happening, and the plans are not very encouraging from this viewpoint. The suitable approach to the space mosaic needs a dramatic change in the whole paradigm of both space infrastructures and (manned/unmanned) missions. Are the current trends in space aligned or misaligned with the holistic sustainability? Are there some technologies that would encourage and support better than others a sustainable-prone design and implementation? How can we commit to space sustainability in practice? The above are just a few of the very long list of questions we could pose about the topic. In what follows, we highlight some answers, leaving to the reader the time to identify her/his own set of additional questions and – most important – the wish of drafting possible answers.

14.3 Major Trends in Space Connectivity has become a right of humanity, like water, food, and energy. Tremendous efforts are ongoing on the terrestrial side to render the connectivity infrastructure pervasive, performing and prone to all kind of vertical applications. A similar trend is now happening also in space. The deployment of very large constellations of satellites, the so-called mega-constellations is aimed at implementing the global connectivity paradigm in space and in an integrated mode with the terrestrial components. Table 14.1 shows some information on these new Low Earth Orbit (LEO) constellations made of hundreds or thousands of space vehicles to provide global connectivity and complementing existing terrestrial network infrastructures.

Table 14.1 Main planned mega-constellations Constellation OneWeb Starlink Amazon Kuiper TeleSat AST Spacemobile

Number of satellites (launched/planned) 212/648 1625/4425 0/3236 0/298 1/240

Altitude (km) 1200 340,550,1150 590 1015,1325 720

Frequency Ku band V-band, Ku band, 6 GHz Ka-band Ka-band