128 39 25MB
English Pages [48] Year 2023
This Month’s Covers…. Front Cover: ©shutterstock.com/Frame Stock Footage Back Cover: ©shutterstock.com/Andrea Izzotti
IEEE AESS PUBLICATIONS BOARD Lance Kaplan, VP–Publications, Chair Daniel O’Hagan, Editor-in-Chief, Systems
Gokhan Inalhan, Editor-in-Chief, Transactions Amanda Osborn, Administrative Editor
IEEE AESS Society The IEEE Aerospace and Electronic Systems Society is a society, within the framework of the IEEE, of members with professional interests in the organization, design, development, integration and operation of complex systems for space, air, ocean, or ground environments. These systems include, but are not limited to, navigation, avionics, spacecraft, aerospace power, mobile electric power & electronics, military, law enforcement, radar, sonar, telemetry, defense, transportation, automatic test, simulators, and command & control. Many members are concerned with the practice of system engineering. All members of the IEEE are eligible for membership in the Society and receive the Society magazine Systems upon payment of the annual Society membership fee. The Transactions are unbundled, online only, and available at an additional fee. For information on joining, write to the IEEE at the address below. Member copies of publications are for personal use only.
THE INSTITUTE OF ELECTRICAL AND ELECTRONICS ENGINEERS, INC. Saifur Rahman, President & CEO Thomas M. Coughlin, President-Elect Forrest D. Wright, Director & Secretary Mary Ellen Randall, Director & Treasurer K. J. Ray Liu, Past President Rabab Ward, Director & Vice President, Educational Activities
Sergio Benedetto, Director & Vice President, Publication Services and Products Jill I. Gostin, Director & Vice President, Member and Geographic Activities Yu Yuan, Director & President, Standards Association John P. Verboncoeur, Director & Vice President, Technical Activities Eduardo F. Palacio, Director & President, IEEE-US
IEEE Publishing Operations Senior Director: DAWN MELLEY Director, Editorial Services: KEVIN LISANKIE Director, Production Services: PETER M. TUOHY Associate Director, Editorial Services: JEFFREY E. CICHOCKI Associate Director, Information Conversion and Editorial Support: NEELAM KHINVASARA Manager, Journals Production: PATRICK J. KEMPF Journals Production Manager: CATHERINE VAN SCIVER
IEEE Aerospace and Electronic Systems Magazine ® (ISSN 0885-8985; USPS 212-660) is published monthly by the Institute of Electrical and E lectronics Engineers, Inc. Responsibility for the contents rests upon the authors and not upon the IEEE, the Society/Council, or its members. IEEE Corporate Office: Three Park Avenue, 17th Floor, New York, NY 10016, USA. IEEE Operations Center: 445 Hoes Lane, P.O. Box 1331, Piscataway, NJ 08855, USA. NJ Telephone: 732-981-0060. Price/Publication Information: To order individual copies for members and nonmembers, please email the IEEE Contact Center at [email protected]. (Note: Postage and handling charges not included.) Member and nonmember subscription prices available upon request. Copyright and reprint permissions: Abstracting is permitted with credit to the source. Libraries are permitted to photocopy for private use of patrons, provided the per-copy fee indicated in the code at the bottom of the first page is paid through the Copyright Clearance Center, 222 Rosewood Drive, Danvers, MA 01923, USA. For all other copying, reprint, or republication permissions, write to the Copyrights and Permissions Department, IEEE Publications Administration, 445 Hoes Lane, P.O. Box 1331, Piscataway, NJ 08855. Copyright © 2023 by the Institute of Electrical and Electronics Engineers, Inc. All rights reserved. Periodicals postage paid at New York, NY, and at additional mailing offices. Postmaster: Send address changes to IEEE Aerospace and Electronic Systems Magazine, IEEE, 445 Hoes Lane, P.O. Box 1331, Piscataway, NJ 08855. GST Registration No. 125634188. CPC Sales Agreement #40013087. Return undeliverable Canada addresses to: Pitney Bowes IMEX, P.O. Box 4332, Stanton Road, Toronto, ON M5W 3J4, Canada. IEEE prohibits discrimination, harassment, and bullying. Printed in the U.S.A.
Editors Editor-in-Chief–Daniel W. O’Hagan, Fraunhofer FHR, Germany VP Publications–Lance Kaplan, U.S. Army Research Laboratory, USA AESS President–Mark Davis, Independent Consultant, USA Operations Manager, AESS–Amanda Osborn, Conference Catalysts, LLC, USA
Contributing Editors Awards–Fulvio Gini, University of Pisa, Italy Book Reviews Editor–Samuel Shapero, Georgia Tech Research Institute, USA Conferences–Braham Himed, Air Force Research Laboratory, USA Distinguished Lecturers & Tutorials–Alexander Charlish, Fraunhofer Institute for Communication, Information Processing and Ergonomics, Germany Education–Alexander Charlish, Fraunhofer Institute for Communication, Information Processing and Ergonomics, Germany History–Hugh Griffiths, University College London, UK Student Research–Federico Lombardi, University College London, UK Technical Panels–Michael Braasch, Ohio University, USA Tutorials–W. Dale Blair, Georgia Tech Research Institute, USA Website Updates–Amanda Osborn, Conference Catalysts, LLC, USA
Associate Editors and Areas of Specialty Scott Bawden–Energy Conversion Systems, Arctic Submarine Laboratory, USA Erik Blasch, US Air Force Research Lab (AFRL), USA Roberto Sabatini, RMIT University, Australia– Avionics Systems Stefan Brueggenwirth, Fraunhofer Institute for High Frequency Physics and Radar Techniques FHR, Germany–AI and ML in Aerospace Dietrich Fraenken, Hensoldt Sensors, Germany– Fusion and Signal Processing Lyudmila Mihaylova, The University of Sheffield, UK–Target Tracking Mauro De Sanctis, University of Rome “Tor Vergata,” Italy–Signal Processing and Communications Jason Gross, West Virginia University (WVU), USA–Navigation, Positioning Giancarmine Fasano, University of Naples Federico II, Italy–Unmanned Aircraft Systems Michael Brandfass, Hensoldt–Radar Systems Raktim Bhattacharya, Texas A&M, USA–Space Systems Haiying Liu, DRS Technologies, Inc., USA– Control and Robotic Systems Michael Cardinale, Retired, USA–Electro-Optic and Infrared Systems, Image Processing Ruhai Wang, Lamar University, USA–Systems Engineering Marco Frasca, MBDA, Italy–Quantum Technologies in Aerospace Systems
October 2023
ISSN 0885-8985
Volume 38 Number 10
COLUMNS In This Issue –Technically . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
3
FEATURE ARTICLES Monitoring of Critical Undersea Infrastructures: The Nord Stream and Other Recent Case Studies G. Soldi, D. Gaglione, S. Raponi, N. Forti, E. d’Afflisio, P. Kowalski, L.M. Millefiori, D. Zissis, P. Braca, P. Willett, A. Maguer, S. Carniel, G. Sembenini, C. Warner . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
4
Efficient Transmitter Selection Strategies for Improved Information Gathering of Aerial Vehicle Navigation in GNSS-Denied Environments A.A. Nguyen, Z.M. Kassas . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
26
NEWS AND INFORMATION Call for Papers: IEEE International Symposium on Phased Array Systems and Technology . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
2
2023 AESS Senior Members Elevated . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
25
2023 AESS Organization and Representatives . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
41
History Column: William Sealy Gosset–“Student” H. Griffiths . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
42
AESS Virtual Distinguished Lecturer Series . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
44
2023 Aerospace & Electronic Systems Society: Meetings and Conferences . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . inside back cover
How to Reach Us We welcome letters to the editor; but, we reserve the right to edit for space, style, and clarity. Include address and daytime phone number with your correspondence. E-mail: Daniel W. O’Hagan, [email protected] Catherine Van Sciver, [email protected]
Publishers: Send books for review to Samuel Shapero, 407 Angier Place NE, Atlanta, GA 30308. If you have questions, contact Samuel by e-mail at [email protected]. Advertisers If you are interested in advertising in the AESS SYSTEMS Magazine, please contact Anthony Land at Naylor Associates at [email protected] or Daniel W. O’Hagen at [email protected]
CALL FOR PAPERS IEEE International Symposium on Phased Array Systems and Technology 15 - 18 October 2024 Hynes Convention Center, Boston, Massachusetts, USA www.ieee-array.org
Gold Sponsors
Other Sponsors and Exhibitors Contact: sponsorships@ ieee-array.org
Technical Co-Sponsors
About the Symposium
Phased array systems continue to be a rapidly evolving technology with steady advances motivated by the challenges presented to modern military and commercial applications. This symposium will present the most recent advances in phased array technology and offer a unique opportunity for members of the international community to interact with colleagues in the field of phased array systems and technology. The committee is thrilled to announce two major changes to the symposium to better reflect the interest and pace of technology development: (1) moving to the larger Hynes Convention Center in the Back-Bay neighborhood of Boston; and (2) increasing the symposium frequency to a two-year cadence.
Be a Symposium Sponsor or Exhibitor
For sponsorship and exhibit opportunities please reach out to Mark McClure and Marc Angelucci at: [email protected].
Suggested Topics • 5G Arrays • Array Design • Array Measurements • Array Signal Processing • Automotive Arrays • Beamforming & Calibration • Digital Array Architectures • Dual Polarized Arrays • Low-Cost Commercial Arrays
Symposium Chairs
Sean Duffy (C), MIT LL Wajih Elsallal (VC), MITRE
Technical Program Chairs David Mooradd (C), MIT LL Glenn Hopkins (VC), GTRI
Special Sessions Chairs
Matt Facchine, NGC Kenneth E. Kolodziej, MIT LL
Plenary Session Chair
Will Moulder, MIT LL William Weedon, Applied Radar
Student Program
Matilda Livadaru, Raytheon Tech Justin Kasemodel, Raytheon Tech
Tutorials
Cara Kataria, MIT LL Frank Vliet, TNO
• • • • • • • • •
MIMO Arrays Medical Applications Metamaterial Phased Arrays mmWave and Terahertz T/R Modules Low Frequency Arrays SATCOM Arrays Weather Arrays Wideband Arrays
Paper Template and Submission Procedures
Template and submission procedures are available at: https://ieee-array.org/paper-submission
Media Sponsor
Committee Chairs
Important Dates • Full paper submission (2-8 pages including figures): 13 May 2024 • Author notification: 22 July 2024 • Author registration deadline: 01 Sept 2024 We are looking forward to seeing you at this next gathering.
Sponsorship and Exhibits
Marc Angelucci, LMC Mark McClure, STR
Digital Platforms Chairs
Pierre Dufilie, Raytheon Tech Jacob Houck, GTRI Mark Fosberry, MITRE Shireen Warnock, MIT LL
Publications/Publicity
Philip Zurek, MIT LL Jack Logan, NRL Elizabeth Kowalski, MIT LL
Poster Sessions Chair Honglei Chen, MathWorks
Advisors
Daniel Culkin, NGC Alan J. Fenn, MIT LL Jeffery S. Herd, MIT LL Bradley Perry, MIT LL
Arrangements/Admin.
Robert Alongi, IEEE Boston Kathleen Ballos, Ballos Assoc.
In This Issue –Technically MONITORING OF CRITICAL UNDERSEA INFRASTRUCTURES: THE NORD STREAM AND OTHER RECENT CASE STUDIES The explosions on 26 September 2022, which damaged the Nord Stream gas pipelines, have highlighted the need and urgency of improving the resilience of critical undersea infrastructures (CUIs). Comprising gas pipelines and power and communication cables, these connect countries worldwide and are critical for the global economy and stability. An attack targeting multiple of such infrastructures could potentially cause significant damage and greatly affect various aspects of daily life. Due to the increasing number of CUIs, existing underwater surveillance solutions, such as autonomous underwater vehicles or remotely operated vehicles, are not adequate enough to ensure thorough monitoring. We show that the combination of information from both underwater and above-water surveillance sensors enables achieving seabed-to-space situational awareness (S3A), mainly thanks to artificial intelligence and information fusion methodologies. These are designed to process immense volumes of information, fused from a variety of sources and generated from monitoring a very large number of assets. The learned knowledge can be used to anticipate future behaviors, identify threats, and determine critical situations concerning CUIs. To illustrate the capabilities and importance of S3A, we consider three events that occurred in the second half of 2022: the aforementioned Nord Stream explosions, the cutoff of the underwater communication cable SHEFA-2 connecting the Shetland Islands and the U.K. mainland, and the suspicious activity of a large vessel in the Adriatic Sea. Specifically, we provide analyses of the available data, from automatic identification system and satellite data, integrated with possible contextual information, e.g., bathymetry, patterns-of-life, weather conditions, and human intelligence.
EFFICIENT TRANSMITTER SELECTION STRATEGIES FOR IMPROVED INFORMATION GATHERING OF AERIAL VEHICLE NAVIGATION IN GNSS-DENIED ENVIRONMENTS Aerial vehicle navigation in global navigation satellite system (GNSS)-denied environments by utilizing pseudorange measurements from M terrestrial signals of opportunity (SOPs) is considered. To this end, the aerial vehicle is tasked with choosing K < M most informative terrestrial SOPs. Two computationally efficient, but suboptimal, transmitter selection strategies are proposed. These selection strategies, termed opportunistic greedy selection (OGS) and one-shot selection (OSS), exploit the additive, iterative properties of the Fisher information matrix (FIM), where OGS selects the most informative transmitters in finite iterations, while the OSS selects in one iteration. Monte Carlo simulation results are presented comparing the OGS and OSS strategies versus the optimal (exhaustive search) selection strategy, where it is concluded that OGS performs closely to the optimal selection, while executing in a fraction of the optimal selection’s time. Experimental results are presented of a U.S. Air Force highaltitude aircraft navigating without GNSS signals in: 1) a rural region and 2) a semiurban region. The performance of the aircraft’s navigation solution with the selected SOP transmitters via optimal, OGS, OSS are compared over a flight segment where the selection remained valid. The position root-mean-squared error with the optimal, OGS, and OSS were 4.53, 6.28, and 7.13 m in the rural region; and 5.83, 6.08, and 6.70 m in the semiurban region for an aircraft traversing a trajectory of 1.48 and 1.22 km, respectively.
OCTOBER 2023
IEEE A&E SYSTEMS MAGAZINE
3
Feature Article:
DOI. No. 10.1109/MAES.2023.3285075
Monitoring of Critical Undersea Infrastructures: The Nord Stream and Other Recent Case Studies Giovanni Soldi , Domenico Gaglione , Simone Raponi , Nicola Forti , Enrica d’Afflisio , Pawe» Kowalski , and Leonardo M. Millefiori , NATO STO Centre for Maritime Research and Experimentation (CMRE), 19126 La Spezia, Italy Dimitris Zissis , University of the Aegean, 84100 Syros, Greece Paolo Braca , NATO STO Centre for Maritime Research and Experimentation (CMRE), 19126 La Spezia, Italy Peter Willett , University of Connecticut, Storrs, CT 06269 USA Alain Maguer, Sandro Carniel , Giovanni Sembenini, and Catherine Warner, NATO STO Centre for Maritime Research and Experimentation (CMRE), 19126 La Spezia, Italy
INTRODUCTION On 26 September 2022, Danish and Swedish seismometers detected [1] a series of explosions on the Nord Stream 1 and Nord Stream 2 underwater gas pipelines in the Baltic Sea. These explosions, besides causing severe damage to the pipes, led to three underwater gas leaks, with the
Authors’ current address: Giovanni Soldi, Domenico Gaglione, Simone Raponi, Nicola Forti, Enrica d’Afflisio, Pawe» Kowalski, Leonardo M. Millefiori, Paolo Braca, Alain Maguer, Sandro Carniel, Giovanni Sembenini, and Catherine Warner are with the NATO STO Centre for Maritime Research and Experimentation (CMRE), 19126 La Spezia, Italy (e-mail: giovanni. [email protected]; [email protected]. int; [email protected]; [email protected]; enrica.d’[email protected]; Pawel.Kowalski@cmre. nato.int; [email protected]; [email protected]; [email protected]; [email protected]; giovanni.sembenini@cmre. nato.int; [email protected]). Dimitris Zissis is with the Department of Product and Systems Design Engineering, University of the Aegean, 84100 Syros, Greece (e-mail: [email protected]). Peter Willett is with the Electrical and Computer Engineering Department, University of Connecticut, Storrs, CT 06269 USA (e-mail: [email protected]). Manuscript received 6 February 2023, revised 5 May 2023; accepted 25 May 2023, and ready for publication 16 June 2023. Review handled by Dietrich Fraenken. 0885-8985/23/$26.00 ß 2023 IEEE 4
subsequent release of an enormous amount of methane into the atmosphere. Figure 1 shows an image of the methane leak acquired by the satellite constellation Pleiades Neo, made available by the European Space Agency (ESA). Despite suspicions of sabotage from certain authorities and organizations, particularly given the current political climate in Eastern Europe, there is currently no concrete evidence to indicate how and by whom the explosions were caused. At the same time, these events have also put the climate community on alert, due to the very powerful greenhouse effect of methane compared to carbon dioxide (approximately 30 times higher), especially in the short term. It has been estimated that more than 220,000 tons of methane—comparable to the annual anthropogenic methane emissions in Austria—had probably been released in the atmosphere during the Nord Stream leakage [2]. While the accident per se is not significantly changing the figures of greenhouse gas emissions leading to global warming and climate change, it nevertheless represents an unprecedented case of interlink between climate change and security aspects that needs to be properly accounted for by governments. The Nord Stream incident has brought attention to the vulnerability of critical undersea infrastructures (CUIs), such as gas pipelines and underwater cables. This has led to increased focus from both the public and policymakers on improving the resilience of these vital assets, as there is growing concern that similar malicious operations will occur in the future. Indeed, more recently, on 22 December 2022, Italian newspapers reported [3] the suspicious activity of a large vessel in close proximity to the Trans Adriatic Pipeline
IEEE A&E SYSTEMS MAGAZINE
OCTOBER 2023
Image provided by the authors and made available under the standard license of Shutterstock.
(TAP) in the Adriatic Sea. Therefore, the protection and surveillance of CUIs are crucial elements that will be included in any future maritime strategy. Sabotage of CUIs can occur through the use of surface assets such as warships or commercial vessels, or by underwater assets, as shown in Figure 2. In the former case, multiple above-water heterogeneous systems and sensors, e.g., automatic identification system (AIS), satellite sensors [4], and terrestrial radars, may have a crucial role to provide a seamless large-scale maritime surveillance (MS), even in
Figure 1. Nord Stream leak as captured by Pleiades Neo (ESA ß Pleiades Neo). The diameter of the leak is estimated in 0.5–0.7 km [2].
OCTOBER 2023
remote areas of the world. In the latter case, underwater sensors (such as active/passive sonars and cameras) installed on the CUIs [5], or equipped on unmanned underwater vehicles (UUVs) [6], would complement MS by providing undersea monitoring capabilities. Besides sensory data, the analysis of contextual information, such as bathymetry, weather data, human intelligence (HUMINT), and open source intelligence (OSINT), is of paramount importance. More in general, the joint use of underwater and above-water heterogeneous sensors, together with contextual and intelligence information is a key concept for the transition toward a seabed-to-space situational awareness (S3A). Given the scale of the problem and the large amount of data to be processed, achieving S3A related to the monitoring of CUIs can be done by using advanced artificial intelligence (AI) and information fusion (IF) techniques. These techniques allow for the integration of vast amounts of information from various sources and monitoring a large number of assets on a daily basis. Examples of such techniques include Bayesian multitarget tracking (MTT) techniques [7], [8], [9], multireasoning systems based on the Dempster–Shafer theory [10], and anomaly detection techniques. These enable the fusion of diverse information at multiple levels. The knowledge gained from these techniques, once extracted and made easily understandable, can provide end users, such as government authorities, defense forces, coast guards, and police, not only with a better understanding of the entities and actors involved in a specific event, such as the Nord Stream incident, their relations, and the potential consequences of these relations, but also with an effective tool to anticipate
IEEE A&E SYSTEMS MAGAZINE
5
Monitoring of Critical Undersea Infrastructures: The Nord Stream and Other Recent Case Studies
Figure 2. Conceptual illustration of the monitoring of a CUI. Depending on the location of the CUI and weather conditions, an above-water sabotage might be potentially conducted by a specialized diver supported by a surface vessel; an ROV may also be employed in some circumstances. Above-water systems and sensors, such as the AIS and SAR, may provide large-scale monitoring capabilities. These may be complemented by underwater assets, such as DAS, to ensure a comprehensive MS.
future threats to CUIs and other critical assets. We anticipate that these techniques will provide the ability to prevent potential future attacks on CUIs and will become of increasing interest to national and international institutions and agencies, as well as the maritime industry. This article is organized as follows. The “Critical Undersea Infrastructure” section provides an overview of CUIs and describes important aspects related to their resilience. The “Sensors and Contextual Information” section provides a description of sensor technologies that could effectively be used for monitoring CUIs, and details useful contextual information. The “Seabed-to-Space Situational Awareness” section provides an overview of 6
S3A and, in particular, it describes state-of-the-art IF, anomaly detection, and automated reasoning techniques. The “Case Studies” section presents analyses on the Nord Stream explosions, the cutoff of the underwater cable SHEFA-2 connecting the Shetland Islands and the U.K. mainland [11], and the anomalous behavior of a large vessel in the Adriatic Sea. Final remarks are provided in the “Conclusion” section.
CRITICAL UNDERSEA INFRASTRUCTURE Nowadays, submarine pipelines and cables represent a vital infrastructure for global finance, economy, maritime
IEEE A&E SYSTEMS MAGAZINE
OCTOBER 2023
Soldi et al. security, and everyday life. Due to their undersea concealment, concern about CUIs is usually risen to the public and institutional attention only after a major accident occurs, a phenomenon referred to in [12] as “collective sea blindness.” Despite their importance and influence on many aspects of our societies, CUIs have only recently seen increasing political and scholarly attention, mostly sparked by military concerns and recent accidents involving underwater gas pipes and cables. Submarine pipelines are the backbone for energy transportation to the market, e.g., oil and gas, as they connect increasingly complex structures, such as offshore rigs, floating storage, and floating processing units, that directly feed ashore. The pipes themselves are of steel, and concrete is used to prevent impacts and damage from ships’ anchors. But these resources are nonetheless vulnerable and thence of concern, witness the recent explosion of the Nord Stream pipeline and its effect on public concern about the resilience of these important infrastructures. Similarly, the submarine cable network, with more than 400 active cables spanning at least 1.3 million kilometers globally, is a vital asset composed of optical fibers and energy cables laid on the ocean floor. They constitute the most efficient and cost-effective solution to sending digital information across the globe. This network digitally connects countries worldwide, with more than 10 trillion USD dollars in financial transactions exchanged every day, and represents the backbone for internet communications globally. Compared to underwater energy pipelines, underwater cables are more vulnerable, since they are more flexible and fragile. The major causes of damage are represented by human errors and negligence. To make an example, 40% of the incidents occur due to trawling activities by fishing boats, while another 15% is caused by anchoring accidents, such as improperly stored anchors, anchoring outside approved areas, anchoring mispositioning due to weather conditions, and emergency dropping of an anchor. Other human-driven benign causes of incidents include oil and gas development, offshore wind and energy constructions, hydro-energy projects, and deep-sea mining operations. Intentional sabotage operations to underwater cables classified as hybrid warfare operations have not yet been officially documented. However, different Russian submarine activities in the proximity of underwater cables have been publicly reported since 2015, raising concern among NATO officials, as the Russian navy has clearly demonstrated an unprecedented interest in undersea cables [13]. In February 2022, Russia has conducted a naval exercise just outside Ireland’s exclusive economic zone, very close to several submarine cables connecting France, the U.K., and the USA. According to an Irish military source, the scope of this operation was a demonstration of capability to sabotage underwater cables [14]. The highest risk associated with underwater cables is represented by high-density maritime bottlenecks. For OCTOBER 2023
instance, seven intercontinental cables pass through the Strait of Gibraltar, between Morocco and the Iberian Peninsula, connecting the Mediterranean Sea with the Atlantic Ocean. Another critical point is the passage through the Red Sea between the Mediterranean Sea and the Indian Ocean, with sixteen underwater cables passing through the Egyptian mainland [12]. Even though most countries could cope with a significant decrease in bandwidth in case of simultaneous damage to multiple underwater cables, island states and oversea territories, not boasting the same redundancy, are the most vulnerable ones. A recent example is the interruption of communications with the Shetland Islands and the Faroe Islands, after the communication cable SHEFA-2 was severely damaged in two distinct points, most likely by trawling fishing boats. The aforementioned critical aspects of the underwater cable networks are analyzed in terms of graph robustness and resilience in the next section.
ROBUSTNESS AND RESILIENCE CUIs are increasingly interconnected and interdependent, thus providing valuable benefits in terms of efficiency, quality of service, performance, and cost reduction. However, these interdependencies increase the vulnerability of CUIs to accidental and malicious threats, as well as the risk of a domino effect on the whole networked infrastructure. Consequently, the impact of infrastructure components’ failures can be aggravated and more difficult to predict, compared to failures confined to a single infrastructure. As an example, blackouts can be caused by the outage of a single transmission element not properly managed by automatic control actions or operator intervention, gradually leading to cascading outages and eventually to the collapse of the entire network. Examples of cascading effects due to infrastructure interdependencies leading to catastrophic events across multiple infrastructures spanning wide geographical areas are documented in [15]. CUIs, such as pipelines, internet lines, and power cables, are part of a complex network of critical infrastructure elements on the bottom of the seas. A complex system can be analyzed by understanding how its components interact with each other by using network science [16]. A network representation offers a common language to study different types of CUIs through graph theory, where a network is described by the system’s components called nodes or vertices, and the direct interactions between them, called links or edges. The structure and topology of such a complex network plays an essential role in a system’s ability to survive random failures or deliberate attacks. In a network context, the system’s ability to carry out its basic functions even when some of its nodes and links may be missing is referred to as robustness. In addition, a system is resilient if it can
IEEE A&E SYSTEMS MAGAZINE
7
Monitoring of Critical Undersea Infrastructures: The Nord Stream and Other Recent Case Studies adapt to component failures by changing its mode of operation, without losing its ability to function. In order to improve resilience, it is important to identify the most critical nodes of networked CUIs that are most likely to be failure points or vulnerable to attacks, and assess the consequences. Recent research on robustness and resilience of complex networks to failures and attacks include, respectively, [17] and [18]. Moreover, different models (e.g., [19], [20]) have been proposed to capture the dynamics of cascading failures in systems characterized by some flow (e.g., information, natural gas, electric current) over a network. This allows us to understand the fraction of nodes that can be removed before global connectivity of the network is lost, how to stop a cascading failure, and how to enhance a system’s dynamical robustness. As already anticipated in the “Critical Undersea Infrastructure” section, considering the underwater cable infrastructure, maritime choke points can be identified as critical points due to their high density of cables and maritime traffic. According to a recent analysis on security threats and consequences for the EU [12], two key maritime bottlenecks are the Strait of Gibraltar and the passage between the Indian Ocean and the Mediterranean Sea via the Red Sea, respectively. The former, connecting the Mediterranean Sea and the Atlantic Ocean, is a dense area used for various maritime activities including submarine activities, with seven intercontinental cables passing through the strait. The latter represents the core connectivity to Asia where intercontinental cables pass through the Egyptian mainland adjacent to the Suez Canal to enhance the system’s dynamical robustness. Understanding and analyzing the interaction and interdependencies among CUIs is of utmost importance. Interdependency is a bidirectional relationship between two infrastructures through which the state of each infrastructure influences or is correlated to the state of the other. Types of interdependencies include the following [21]. Physical interdependencies, which arise from physical links or connections among elements of the network. In this context, disruptions and perturbations in one component can propagate to other elements. Cyber interdependencies, which occur when the state of a component depends on data transmitted through the information infrastructure. Such interdependencies result from the increased use of computer-based information systems for monitoring and management activities (e.g., supervisory control and data acquisition). Geographic interdependencies, which exist between two infrastructures when a local environmental event can provoke changes in both of them. This generally occurs when the components are in close spatial proximity, e.g., infrastructures that 8
cross borders or that provide cross-border services, thus impacting the interests of different nations. Finally, logical interdependencies, which gather all interdependencies that are not physical, cyber, or geographic, caused by, e.g., regulatory, legal, or policy constraints.
LEGAL ASPECTS The legal status of underwater pipes and cables varies based on the legal zone in which they are located, as determined by the United Nations Convention of the Law Of the Sea. Within a country’s territorial waters, which extend up to 12 nautical miles from the coastline, the country has full jurisdiction over the pipe/cable. In the contiguous zone, which extends from 12 to 24 nautical miles from the coastline, states have specific law enforcement duties and obligations. Outside of these zones, particularly in the high seas (areas outside of national jurisdiction) as well as in the exclusive economic zones of states (i.e., up to 200 nautical miles from a nation’s coastline), the legal status of pipes/cables and responsibility for their protection is currently defined as “unclear” and “ambiguous” [12], [22]. Furthermore, underwater pipes may also be subject to additional regulations and laws depending on the activities they are associated with and their specific use. To make an example, oil and gas underwater pipes may be subject to additional environmental protection and safety-related regulations, while underwater pipes crossing national borders (or used for international trade) may be subject to additional international laws and agreements.
SENSORS AND CONTEXTUAL INFORMATION In this section, we describe different types of sensors and technologies that are available and can be exploited for CUIs surveillance.
COASTAL RADARS AND AIS Among the surveillance sensors commonly used for MS, S-band, or X-band pulse radar sensors installed along the coastline represent a relevant and consistent source of information [23]. However, their coverage area and maximum range might be limited by line-of-sight propagation. This limitation can be overcome by the installation and employment of long-range sensors, such as high-frequency surface-wave (HFSW) radars, which have been considered for ship localization and tracking [24], [25], [26], [27], [28]. Initially introduced for ocean remote sensing, HFSW radars could dramatically increase MS coverage by their ability to detect targets at over-the-horizon
IEEE A&E SYSTEMS MAGAZINE
OCTOBER 2023
Soldi et al. (OTH) distances. In particular, multiple HFSW radars [26], [27], [28], combined with other data sources, such as AIS, satellite images, and contextual information, have the potential to provide continuous-time coverage of large sea areas at OTH distances [8], [9]. Besides conventional and OTH coastal radars, AIS— an anticollision broadcast system of transponders automatically exchanging ship traffic information for maritime safety—definitely represents the major source of information by volume and granularity on surface vessel traffic. According to the International Association of Marine Aids to Navigation and Lighthouse Authorities, the scope of AIS is “to improve the maritime safety and efficiency of navigation, safety of life at sea and the protection of the marine environment” [29]. Since 2002, the International Maritime Organization (IMO) Safety of Life at Sea (SOLAS) convention [30] has included a mandate that requests many commercial vessels to fit onboard AIS. Specifically, IMO requires ships over 300 gross tonnage (GT), cargo vessels over 500 GT, all passenger ships, and all fishing vessels over 45 meters (in EU countries over 15 meters [31]) to be equipped with an AIS transponder onboard. AIS messages can be exchanged through both satellite and terrestrial receivers [4] and convey information about ship identifier, i.e., the maritime mobile service identity (MMSI), route (position, speed, course, and true heading), and other ship and voyage information, including ship and cargo type, size, destination, and estimated time of arrival. The analysis of AIS trajectories is used, among others, to pinpoint potentially illegal or illicit activity performed by vessels in specific areas of interest (see, e.g., [32], [33]). A first filtering of AIS trajectories could be performed according to some selection criteria. In particular, considering a bounding area of Dmax kilometers around a CUI or a point of interest, e.g., Nord Stream explosion point, one could select all the AIS trajectories that have spent at least a given period of time Tmin inside the area. A further filtering to spot stationary or drifting ships could be achieved by selecting those trajectories whose average speed is lower than a maximum speed Smax , or whose average rate of maneuvers per minute is higher than a predefined threshold. Nonkinematic information may be useful as well for AIS-based filtering. For example, a nonkinematic-based selection criteria may consist of excluding vessels whose small size would not allow them to represent a possible threat to underwater pipes (e.g., fishing vessels in very deep waters). A combination of kinematic-based and nonkinematic-based selection criteria can also be taken into account. In reality, AIS messages can be counterfeited and AIS transponders can be easily switched off, or vessels could navigate outside the coverage of coastal/satellite AIS receivers. For these and other reasons, AIS data are often complemented with data from other sensors or sources; for instance, historical geo- and time-referenced images provided by OCTOBER 2023
space-based sensors, e.g., synthetic aperture radar (SAR), multispectral (MSP), and hyperspectral (HSP) sensors.
SATELLITE SENSORS: SAR, MSP, AND HSP SAR is a high-resolution imaging system typically employed on board satellites (or aircraft) that has become essential for wide-area monitoring to detect and track vessels at sea independently of their compliance with the SOLAS convention. It is an active remote sensing technology based on the transmission of an electromagnetic (EM) microwave signal toward the Earth, and the reception and processing of the signals scattered by any natural or artificial features on the surface. The use of EM microwave signals—that undergo a weak scattering and absorption from the atmosphere—makes the monitoring capability of SAR systems independent of sunlight illumination, thus allowing an effective sensing regardless of the weather conditions. SAR systems typically operate in a monostatic geometry, where the receiving antenna is co-located with the transmitting one. In this configuration, only the energy reflected in the backscattering direction is collected by the system. The received signals are then properly processed to form a two-dimensional (2D) image of the scene reflectivity in slant-range/azimuth coordinates. The features of the obtained image largely depend on both surface parameters, e.g., material composition, small- and large-scale roughness, and sensor parameters, e.g., viewing angle, frequency, polarization, and spatial resolution. While active microwave remote sensing systems, such as SAR, offer a key support for gathering rather coarse information at multiple frequencies, polarizations, and viewing angles, MSP and HSP optical sensors can be exploited fruitfully to infer more accurate details in the spatial and spectral domains. MSP and HSP sensors cover the entire optical region—between microwaves and X-rays— which refers to wavelengths ranging from 0.3 to 15 mm. Space-based MSP missions allow to collect imagery easy to interpret at a usually higher spatial resolution than SAR systems; however, MSP sensors are sensitive to cloud and sunlight conditions, and they usually cover limited areas during each acquisition. On the other side, HSP satellite images are characterized by very high-spectral resolution and are suited to accurate classification, but they have low spatial resolution, require a large computational burden, and are as well sensitive to cloud and sunlight conditions. For the interested reader, a comprehensive review of satellite sensors can be found in [4].
UUVS EQUIPPED WITH ACOUSTIC SENSORS UUVs equipped with acoustic sensors are rapidly becoming the predominant platforms for undersea observation and monitoring. UUVs can be generally classified into
IEEE A&E SYSTEMS MAGAZINE
9
Monitoring of Critical Undersea Infrastructures: The Nord Stream and Other Recent Case Studies remotely operated vehicles (ROVs) and autonomous underwater vehicles (AUVs) [34]. The former ones are tethered to ships or marine platforms and operated from above the water’s surface. The tether allows operators to receive sensor data, e.g., sea-bottom images from cameras, and, if needed, guide remotely the ROVs almost in real time; however, the operation range is limited by the tether’s length. On the other side, AUVs are untethered and computer controlled, with little or no operator interaction while performing their preprogrammed subsea mission; multiple AUVs can also cooperatively form an intelligent sensing network for the monitoring of large regions of interest [35], [36]. Even though AUVs can be programmed to survey larger areas than ROVs, their operating range is dependent on the duration of their batteries. Furthermore, underwater communications, commonly exploiting the sound channel, are unreliable and characterized by limited bandwidth and range, thus limiting the ability of AUVs to effectively share sensor information in real time. Therefore, the choice among ROVs and AUVs is dependent on the mission requirements as well as on maximum range, operating depth, time to cover a required distance, and type and size of sensors they could bring on board, e.g., acoustic, magnetic, optical, and oceanographic. Since optical and EM waves do not propagate well in seawater, acoustic sonars, which employ sound waves to detect and consecutively localize underwater objects, are nowadays the most common technology for undersea surveillance. Passive sonars rely on the reception and processing of acoustic information that is radiated by underwater noise sources e.g., the noise produced by ships or submarines propellers; active sonars, instead, send an acoustic waveform and process the signals reflected by underwater objects. Synthetic aperture sonars (SAS) [37] represent an established technology to collect high-resolution images of the seabed and underwater infrastructures. Similarly to SAR, an SAS continuously transmits acoustic signals and combines successive received pulses reflected by an object or a surface along a known track to create a 2D image of the illuminated area.
DISTRIBUTED ACOUSTIC SENSING (DAS) DAS is an emerging technology which is commonly employed for the detection and analysis of seismic waves on the ocean bottom and for submarine structural characterization [38], [39]. It is enabled by fiber optic installed along underwater infrastructures, that continuously allows the monitoring in real time of underwater assets. While traditional monitoring systems rely on discrete sensors measuring at prefixed points, the fiber optic cable enables a continuous monitoring along a very long portion of the underwater infrastructure. Even though current commercial DAS systems allow a thorough monitoring along a 10
maximum distance of 50 km, recent studies have shown that persistent monitoring could be enabled up to a hundred kilometers. The most common DAS technologies are based on phase sensitive optical time domain reflectometer (f-OTDR) and coherent OTDR [40]. A DAS interrogator unit generates a series of laser pulses, sends them through the optical fiber cable, and collects the backscattering of the light along the length of the fiber. The analysis of the backscattered signal by means of classification algorithms allows to detect and locate events, such as leaks, intrusion activities, cable faults, or other anomalous events.
CONTEXTUAL INFORMATION Contextual information is generally intended as information that does not directly refer to the assets under surveillance, but to their surroundings. Contextual information adds to the operational picture the clarity that is needed to drive the actions to be taken. As such, contextual information is seldom conveyed by a single piece of information alone, but is rather derived from a mixture of experience, domain knowledge, and data artifacts. In the MS setting, examples of contextual information include geographic databases, such as the bathymetry and the displacement of critical infrastructures; geospatial information, such as meteorology and oceanography; intelligence reports, comprising HUMINT and OSINT; reports on business ownership structures, sanctions, and criminal behavior of ship owners; and derived information from past/historical data, such as maritime patterns-of-life (POLs). To make a practical example, a vessel’s trajectory could raise major concern if the vessel was previously involved in criminal activities, if only opaque ownership-related information of the vessel is available, or if there is evidence of the vessel deploying specialized equipment in proximity of sensitive infrastructures. Moreover, taking into account bathymetry is crucial, since the difficulty and risk in performing sabotage operations are directly proportional to the sea depth. Furthermore, POLs can be considered as particular sets of behaviors and movements, e.g., waiting, navigating, or drifting, associated with specific entities, e.g., fishing vessels, cargo vessels, and oil tankers, over a defined period of time. In this context, density maps, built using historical AIS data in a given time interval and area, provide a preliminary insight on the most common POLs. This crucial information can be used for a preliminary classification of AIS trajectories. In fact, a selected AIS trajectory can be considered more or less suspect depending on whether its behavior seems compatible with the detected POLs in the given period of time and area. Figure 3(a) shows a density map of the entire maritime traffic in the Baltic Sea, built using AIS data collected from 1 September to 15 September, 2022. Purple patterns highlight the most common maritime routes in the considered region. The same data have been
IEEE A&E SYSTEMS MAGAZINE
OCTOBER 2023
Soldi et al.
Figure 3. Density maps of the maritime traffic in the Baltic Sea built using AIS data from 1 September to 15 September, 2022. A higher brightness corresponds to a higher traffic density. Green and red dotted lines show the Nord Stream 1 and 2 gas pipelines, respectively. (a) Density map of the entire maritime traffic of the Baltic Sea. (b) Density map of the stationary areas of the Baltic Sea.
used to derive a density map of the stationary areas, as shown in Figure 3(b).
SEABED-TO-SPACE SITUATIONAL AWARENESS The major challenge that operators and analysts face is identifying patterns emerging within very large datasets, e.g., AIS, SAR, optical, and MSP data, when the goal is to anticipate possible future behaviors of suspicious assets and the related threats. In this context, information undoubtedly plays a crucial role, and AI opens up unprecedented possibilities for surveillance systems to improve MS, and in particular the resilience of CUIs. AI and IF can easily process immense volumes of information, fused from a variety of sources and generated from a very large number of monitoring assets on a day-to-day basis, thus enabling a potential future transition to an holistic perspective of S3A. The learned knowledge therefrom can be used as a valuable support to the cognitive processes (perception, comprehension, and projection) of analysts and operators to anticipate future behaviors and/or identify threats and critical situations that might endanger CUIs. In the following, we will provide an overview of the state-ofthe-art Bayesian IF and MTT, anomaly detection, and automatic reasoning techniques, that might enable S3A and improve the monitoring of CUIs.
BAYESIAN IF AND MTT The main objective of a multisensor MTT method is to sequentially estimate the number of targets together with their states, e.g., position, velocity, course, and heading, in a particular maritime area, by fusing measurements from OCTOBER 2023
multiple heterogeneous sensors. Each measurement is either a noisy observation of a target’s kinematics, shapes, or other features, or a false alarm. In a Bayesian formulation, the MTT method amounts to estimating at each time (approximations of) the marginal posterior distributions of the detected targets’ states, using all the measurements available up to the current time. MTT methods have to deal with various challenges, for example, the heterogeneity of the different information sources [8], [9], asynchronicity, out-of-sequence measurements [41], latency, and the measurement-origin uncertainty (MOU) [42], i.e., the fact that it is unknown, which target (if any) generated which measurement. Existing MTT algorithms can be broadly classified as vector-type algorithms, such as the joint probabilistic data association filter [43] and the multiple hypothesis tracker [44], [45], and set-type algorithms, such as the (cardinalized) probability hypothesis density filter [46], [47] and multi-Bernoulli filters [48]. Vectortype algorithms represent the multitarget states and measurements by random vectors, whereas set-type algorithms represent them by random finite sets. Algorithms of both types have been developed and evaluated, and several limitations have been noted [9]. First, the fusion of heterogeneous information sources is not straightforward. Second, they do not adapt to time-varying model parameters. And third, their complexity usually does not scale well in relevant system parameters, e.g., the number of sensors. An emerging approach to MTT and IF—one with flexibility, low complexity, and useful scalability—is based on a factor graph and the sum-product algorithm (SPA) [42]. First, a factor graph representing the statistical model of the MTT problem is derived; then, the SPA is used to solve efficiently the MOU problem and obtain a principled and intuitive approximation of the Bayesian
IEEE A&E SYSTEMS MAGAZINE
11
Monitoring of Critical Undersea Infrastructures: The Nord Stream and Other Recent Case Studies inference needed for target detection and estimation. A major advantage of the SPA is its ability to exploit conditional independence properties of random variables for a drastic reduction of complexity; thereby, SPA-based MTT algorithms can achieve an attractive performance–complexity compromise, making them suitable for large-scale tracking scenarios involving a large number of targets, sensors, and measurements, and allowing their use on resource-limited devices. Generally, the complexity of an SPA-based MTT algorithm scales quadratically in the number of targets, linearly in the number of sensors, and linearly in the number of measurements per sensor, and outperforms previously proposed methods in terms of accuracy (see [42] and references therein). SPA-based MTT algorithms can be easily extended to automatically estimate unknown and time-varying parameters [49], such as detection probabilities of sensors, incorporate multiple dynamic models, and in particular fuse heterogeneous data [8], e.g., from terrestrial radars, SAR, optical sensors, and AIS. In the context of S3A and in particular CUIs monitoring, the fusion of satellite data, e.g., SAR, with AIS potentially allows the identification of ships, which switch off the AIS during anomalous activities or sabotage operations. The fusion of this information is often difficult due to the asynchronicity and sparsity of AIS messages, and the nontrivial association between messages and targets. Indeed, although each AIS message usually includes a unique identifier—the MMSI—this may be absent, incorrectly received, or observed for the first time, in which case no prior information is available on the target–message association. The SPA-based MTT method can be efficiently extended to fuse AIS messages and measurements obtained from SAR, optical images, or other sensors, and to identify (or label) each detected ship by means of the MMSI.
ANOMALY DETECTION Increasing automation through a large number of advanced data-driven methods and techniques for maritime anomaly detection has enabled the system and the operator to spot complex situations by correlating various events from all surveillance sensors and classify them into important incidents [50], [51], [52]. An anomaly in the maritime domain can be described as a behavior that is not “normal” or, more specifically, not expected to occur during regular operations [50], and it can refer to a sudden change in vessel kinematic behavior (such as unusual speed or location), deviation from standard sea lanes, unexpected AIS activity, unexpected port arrivals, close approach, and zone entry [53]. In particular, zone entry anomalies involve ships entering in protect environmental areas, or approaching military installations or CUIs. 12
Most anomaly detectors require learning an underlying model representing the normal behavior by using the available historical data. Based on the learned normalcy model, detectors can decide whether new data can be classified as normal or anomalous behavior. According to the available proposals and studies [51], [54], the data-driven methodologies for maritime anomaly detection can be divided into machine learning and stochastic approaches. Machine learning techniques are able to identify patterns emerging within huge amounts of maritime data, fused from various uncertain sources and generated from monitoring thousands of vessels a day, so as to act proactively and minimize the impact of possible threats. The general aim of such an approach includes frequent pattern discovery, trajectory pattern clustering in a multidimensional feature space, trajectory classification, forecasting, and anomalies/ outliers detection. The machine learning framework mainly comprises the following unsupervised and supervised methodologies. i) Distance-based clustering methods are mainly based on the nearest neighbor algorithm and implement a well-defined distance metric; the greater the distance of the object to its neighbor, the more likely it is to be an outlier. ii) Density-based clustering methods identify distinctive groups/clusters in the data based on the idea that a cluster in a data space is a contiguous region of high point density, separated from other such clusters by contiguous regions of low point density. In particular, the densitybased spatial clustering of applications with noise algorithm and its variations have become very popular for their convenient properties: these methods do not require specifying the number of clusters and have the ability to derive arbitrarily shaped clusters and identify outliers [55]. iii) Classification methods require the construction of a classifier, that is, a function that assigns a class label to instances described by a set of attributes. Using supervised learning approaches, trajectories or segments of a trajectory can be classified into some categories, which can be motions, human activities, or transportation modes. The main classification methods include neural networks, support vector machines, and decision trees. Stochastic approaches to maritime anomaly detection fit a statistical model representing normal vessel behavior to the given historical vessel movement data, and then apply a statistical inference test to determine whether a new vessel observation belongs to this model or not. Observations that have a low probability of being generated from the learned model, based on the applied test statistic, are declared as anomalous behaviors. Bayesian networks (BNs) and their temporal extensions dynamic BNs (DBNs), Gaussian processes, and the Gaussian mixture model represent the major methods within the stochastic approach. In particular, DBNs allow the modeling of causal relationships among latent variables, observable quantities, and also different hypothesis, through acyclic directed graphs. DBNs have been applied in various contexts, such
IEEE A&E SYSTEMS MAGAZINE
OCTOBER 2023
Soldi et al. as the detection of illegal diving and other anomalous activities [56], and the modeling of maritime piracy situations [57] and anomalous behaviors [58], i.e., deviation from nominal route, unexpected AIS activities, unexpected port arrival, close approach, and zone entry. Current research on maritime anomaly detection considers the aforementioned types of anomalies. The deviation from a standard route is the most prominent anomaly type that research addresses by extracting frequently traveled sea routes from historical AIS data, e.g., via clustering. Unknown AIS tracks can then be compared in order to investigate whether they are similar enough to the extracted routes, or in the case of clustering, belong to one of the identified route clusters. These approaches work very well in areas where many ships take similar routes [59], [60], [61]. A different approach [32], [62] faces the same anomaly within a stochastic framework by combining the available context data with a parametric model of the vessel’s kinematic behavior and running a hypothesis testing procedure to make decisions on the existence of anomalous deviations relying upon the available measurements (e.g., AIS, radar, SAR). This statistical framework for anomaly detection has also been applied in the context of real major events, such as the grounding of the container vessel Ever Given in the Suez Canal on 23 March 2021 [62]. Further stochastic strategies address the joint problem of sequential anomaly detection and tracking of a target subject to switching anomalous deviations in a Bayesian framework [63], [64], [65]. Other approaches particularly consider deviations from standard routes in the presence of unexpected AIS activity. Indeed, AIS tracks of ships are often characterized by large blind spots, and this may be due to unintentional technical problems, radio interference, attenuation, or actual equipment manipulation, such as intentionally turning off AIS transceivers [32], [66], [67]. Moreover, AIS signals can be easily spoofed by external attackers or the crew itself willing to obscure their locations [33], [68]. The AIS intentional interruption or spoofing could indicate a will to hide some illegal activities, such as smuggling on coast or with other ships, or entry in unauthorized areas. Zone entry as an anomaly type is considered only marginally [53], [69]. Restricted zones with entry ban are learned implicitly as part of the general shipping routes and trajectories [69], whereas more elaborate methods, such as predicting whether a zone entry is likely to occur soon are proposed in [53]. Anomalous port arrival is taken into account in [70] looking at ferries that run regular routes according to a fixed schedule, while close approach anomalies are investigated in [53], [71], [72].
REASONING FOR SITUATION AWARENESS Besides the data-driven approaches described earlier, there exist situations where higher level reasoning needs to be OCTOBER 2023
considered. This can be useful for aligning highly heterogeneous information sources, which range from HUMINT and OSINT descriptions of the vessels’ current and projected behavior, through contextual information to MTT tracks and anomaly reports. This is in line with the activitybased intelligence [73], [74] paradigm that has been in use since the war in Afghanistan and has brought a new vision of intelligence pushing forward the development of multiintelligence (multi-INT) capabilities, which aim at considering in an exhaustive way all sources of information. In order to make sense of this kind of data, each source needs to be corrected to account for its reliability and possibly contextual information. In cases where different sources provide reports on the same target property, these reports can be aligned in accordance to an appropriate correction model; if the reports refer to different properties, instead, they are used as inputs to a reasoning system that verifies their consistency. For example, consider source A that reports a particular vessel as a tanker, and a source B that reports it to be a cargo vessel. Given that the reports are concerned with the same property with different degrees of semantic granularity, this information can be readily aligned. On the other hand, a source may report on the vessel type, and the other on its speed. In this case, the properties can be aligned using a reasoning system verifying the consistency of the speed given the vessel type. This can highlight conflict between the sources, which can further lead to conclusions about spoofing or another anomaly, depending on the exact scenario. Tackling hybrid threats is particularly challenging as it is necessary to predict an event, which, above all, is rare and can be essentially considered a black-swan event. As such, it is unlikely to provide an analysis based on data and machine learning only; in these cases it is possible to leverage expert information using rule-based systems. One such approach, which allows a degree of semicausal reasoning, involves using valuation-based systems with the Dempster–Shafer theory [75] (also known as the theory of belief functions). Such systems are also known as evidential networks. The information provided by the myriad of sources described earlier is corrected and aligned with a common vocabulary using a mechanism, such as contextual belief correction [76], behavior-based correction [77], or its context-aware extension [10]. A set of rules elicited from the experts is used to construct a valuation network, which is defined by a set of variables (some of which may be directly mapped to the observations provided by the sources whereas the others are inferred) and the relations between them [78]. Possible uses of such rule-based approaches range from trivial to significantly more complex ones. A trivial illustration could be reasoning about a vessel’s inconsistent AIS status. For example, a trajectory classifier can be considered one source, and the vessel’s AIS navigational status the other. If the navigational status is inconsistent with the type of trajectory provided by the classifier, the AIS information
IEEE A&E SYSTEMS MAGAZINE
13
Monitoring of Critical Undersea Infrastructures: The Nord Stream and Other Recent Case Studies reported by the vessel can be considered inconsistent. This by itself is unlikely to mean that the vessel is involved in illegal activity, but it may be an indication of anomaly which should lead to further investigation. More complex evidential networks have also been proposed in this context to assess the vessel’s intent, likelihood of criminal activity, or the overall threat level posed. One particular strength of the belief-theoretic approach provided by the evidential networks is that the rules are considered to be partially uncertain, thus the reasoning can still be performed with missing data. This can be leveraged to identify which sources of information should be queried in order to improve the quality of the inference results. Finally, the transparent-bydesign property of expert systems facilitates the implementation of explanation abilities; these describe the relative contribution of each source to the decision made, the overall conflict and uncertainty in the answer obtained, as well as the impact of various uncertainties in data [79]. Furthermore, for a better understanding of overall situation awareness which includes multiple entities interconnected by a set of relations, we can leverage conceptual graphs or knowledge graphs. Recent work on these has explored embedding uncertainty in such graphs and establishing links with belief functions theory and evidential networks in order to allow uncertain reasoning about multiple entities [80]. In the context of hybrid threats, this is particularly relevant to synchronized threats originating simultaneously from several vessels coordinating their activity.
CASE STUDIES This section reports analyses on the Nord Stream and Shetland Islands accidents, as well as on the anomalous behavior of a large vessel in the Adriatic Sea, and how surveillance sensors data, in particular AIS reported information and SAR images, might be beneficial for CUIs monitoring. All Sentinel-1 SAR images presented in the following use cases are publicly available through the Copernicus Open Access Hub [81], which provides complete, free, and open access to all Sentinel missions. Raw SAR images have been preprocessed with the ESA Sentinel application platform software [82], following the guidelines provided in [83]. AIS data related to the Nord Stream and Shetland Islands accidents has been retrieved by the NATO STO CMRE database, which collects AIS contacts from both governmental and public sources. AIS data related to the Adriatic use case, instead, has been provided by MarineTraffic, a leading provider of ship tracking and maritime intelligence.
NORD STREAM We focus on examining the maritime situational picture over the Baltic Sea and, specifically, around the Nord Stream gas pipelines. By performing data association 14
between detected ships in Sentinel-1 SAR images and trajectories provided by the AIS, we look for anomalous behaviors since some days before the explosions detected on 26 September 2022 [1]. As an example, Figure 4 depicts a Sentinel-1 SAR image with the circles indicating detected ships, while solid lines representing vessels’ trajectories that cover a time interval from 10 minutes before to 10 minutes after the SAR image acquisition time. These trajectories are obtained by interpolating the AIS positions reported by the ships themselves, and filling gaps in the data of at most 6 hours. The interpolation requires the availability of AIS positions both before and after the image acquisition time; when this information is not available, e.g., during (quasi) real-time operations, the vessel’s position at the time of the image acquisition can be inferred from the available past data [84]. The combined use of AIS data and satellite images (SAR, optical, MSP, and HPS) allows giving an identity to a picture of a vessel that could complement AIS for MS and that, otherwise, would remain unknown. On the other hand, the unavailability of AIS data for a detected ship within a satellite image can be caused by incorrect or lost data, or might highlight an anomalous behavior that requires further investigation. The association between AIS trajectories and SAR detections in Figure 4(a) is achieved by solving a specific assignment problem (clearly, other solutions are available in the MTT literature, discussed in the previous section). The assignment cost between any SAR detection–AIS trajectory pair is the relative distance between the location of the SAR detection and the interpolated position of the vessel at the time of the image acquisition if such a distance is below 3 km, and it is assumed infinite otherwise. The association is then represented by the color of the circle that matches the color of the AIS trajectory which it is associated with; white circles represent SAR detections that are not associated with any AIS trajectory. We observe that the majority of the vessels detected in Figure 4 (a) are associated with an AIS trajectory. Figure 4(b) and (c) present two examples of associated detections and not-associated detections, respectively. The first panel shows three detected ships, each within a colored circle. The color of the circle matches that of the AIS trajectory, obtained as the interpolation of the AIS reported positions represented by the crosses. The stars indicate the interpolated positions of the vessels at the time of the image acquisition; the offset between the shape of a vessel as detected by the SAR and its interpolated AIS position is due to the unknown Doppler frequency generated by the motion of the vessel itself, and it is thus related to its velocity [85]. Figure 4(c), instead, shows a single detected ship—enclosed in the white circle—which is not associated with any AIS trajectory; the closest available AIS interpolated position, indeed, is more than 3 km away. This AIS trajectory, however, presents a relevant characteristic, that is, a gap of several hours (yet less than 6
IEEE A&E SYSTEMS MAGAZINE
OCTOBER 2023
Soldi et al.
Figure 4. Nord Stream use case. (a) Sentinel-1 SAR image acquired over the Baltic Sea some days before the explosions. Colored solid lines are vessels’ trajectories obtained by interpolating the AIS positions provided by the vessels themselves. Circles are ships as detected within the SAR image: the color of the circle identifies the detection as coming from a vessel whose AIS trajectory has the same color; a white circle indicates that the detection is not associated with any AIS trajectory. Panels (b) and (c) present two details of the full SAR image showing examples of associated detections and not-associated detections, respectively. Here, the crosses depict AIS reported positions, and the stars are the interpolated positions of the vessels at the time of the image acquisition.
hours) in the AIS data before and after the acquisition of the SAR image. On one hand, this could suggest that the detected object and the AIS trajectory depicted in Figure 4(c) do refer to the same vessel; on the other hand, the gap in the AIS data and the detection of the ship in a position that deviates from a OCTOBER 2023
linear path, might be an indication of an anomalous behavior. In this context, automatic anomaly detection techniques, as the ones described in the “Anomaly detection” section relying upon the available measurements—i.e., AIS data before and after the acquisition of the SAR image, as well as the
IEEE A&E SYSTEMS MAGAZINE
15
Monitoring of Critical Undersea Infrastructures: The Nord Stream and Other Recent Case Studies
Figure 5. Nord Stream use case. Full trajectory of the vessel of interest a few days before the explosions. The trajectory spans several days and its color (cyan to magenta) is related to the vessel’s speed (slower to faster); gray arrows indicate the direction of the vessel at the beginning and at the end of the trajectory. The locations of the gas leakage are marked by yellow diamonds.
ship detections in the SAR image itself—might be crucial to detect such a stealth deviation from the linear path [32], [33]. In particular, an hypothesis testing procedure built on the changes of the mean velocity parameter of the Ornstein– Uhlenbeck (OU) process, that well describes the kinematic of nonmaneuvering ships [84], [86], might be run to detect whether or not a deviation from the nominal route happened within the specific time period. We note that even though such methodology could enable advanced anomaly detection capabilities to support S3A, its large scale implementation in real time is still very challenging, mainly because most of the available current sensors are not engineered to support autonomy, as their data frequency (e.g., the revisit time of most satellite sensors) is not high enough for the real-time decisionmaking needed by an autonomous surveillance system [52]. Also, it is required to resolve conflicts between heterogeneous sensors and to interpret and support decisions in ambiguous situations. Proceeding further into the analysis of AIS trajectories, the behavior of a specific vessel, which has spent a significant amount of time drifting and loitering near the explosion sites, appears significant. Figure 5 shows its full trajectory— over several days—before the Nord Stream accident. The colors, from cyan to magenta, reflect the speed of the vessel. Extracts of the trajectory, each spanning several hours, are reported in Figure 6. Speeds below three knots are reported in black because large vessels, such as the one considered, are hard to be maneuvered at such low speeds and tend to drift. These portions of the trajectory may suggest that the ship is following a search path, that is, it is maneuvering to 16
approach the pipeline, and moving away from it while drifting. Note that such behavior might also be compatible with other scenarios as, for example, loitering while waiting for orders. Nevertheless, it is worth mentioning that the region where the Nord Stream accident occurred is not designed as a stationary area. The presence of the vessel in the area is corroborated by its detection in two Sentinel-1 SAR images; Figure 6(a) and (b) reports the locations of the vessel as detected in the SAR images, while Figure 7(a) and (b) shows the details of the mentioned Sentinel-1 SAR images. Both the SAR images are acquired while the vessel is drifting: this is confirmed by the vessel’s speeds reported in Figure 6(a) and (b) at the time of the acquisition (below three knots) and by the orientation of the vessel as reported in Figure 7. Finally, note that the slight offset between the vessel’s trajectory (dashed line) and the shape of the ship as acquired by the SAR, particularly evident in Figure 7(b), is compatible with the vessel’s AIS reference point. Beyond the aforementioned kinematic characteristics, two important remarks are useful to assess the potential implication of the vessel in the Nord Stream explosions. The first one regards the vessel’s ownership that results difficult to establish; this, in turn, might raise concern about the legitimacy of the activities in which the vessel is involved in. The second remark is related to the operations that can be conducted by the vessel: according to several navy officers, consulted by the authors, the vessel would have been perfectly able to support and coordinate a sabotage operation using specific instrumentation, such as an ROV.
IEEE A&E SYSTEMS MAGAZINE
OCTOBER 2023
Soldi et al.
Figure 6. Nord Stream use case. Extracts of the full trajectory of the vessel of interest reported in Figure 5. Each panel reports a portion of the full trajectory that spans several hours. The colors (cyan to magenta) are related to the vessel’s speed (slower to faster); speeds below three knots are all reported in black. Gray arrows indicate the direction of the vessel at the beginning and at the end of the trajectory, and yellow diamonds mark the locations of the gas leakage. Panel (a) and (b) also report with a red dot the location of the vessel as detected in two SAR images acquired, respectively, at time TA and at time TB. Additional details on these SAR detections are presented in Figure 7.
As discussed in the previous sections, all the aforementioned information needs to be ingested and automatically processed to asses an overall risk associated with a suspicious ship.
SHETLAND ISLANDS On 20 October 2022, the news reported that the south segment of the SHEFA-2 cable, connecting the Shetland Islands to the U.K. mainland via the Orkney Islands, was cut; the north segment of the same underwater cable, connecting the Shetland Islands to the Faroe Islands, was also damaged a week earlier, thus causing a major communication outage on the islands for several days [11]. The area where the cable OCTOBER 2023
lies is interested in an intense fishing activity conducted by multiple trawlers. It is therefore widely thought that the damage to the SHEFA-2 cable was accidentally caused by a fishing vessel, as also documented in the past, and not the result of a sabotage [87]. In order to better understand the causes of this accident, we analyze the AIS data collected in the region several days before the cutoff was reported. In particular, we select AIS trajectories (cf., the “Coastal Radars and AIS” section) of the vessels which spent more than 6 hours within a 10 km distance from the SHEFA-2 cable, and whose average speed over ground was lower than four knots. From this analysis, it results that multiple boats were actually engaged in fishing activities. The trajectory
IEEE A&E SYSTEMS MAGAZINE
17
Monitoring of Critical Undersea Infrastructures: The Nord Stream and Other Recent Case Studies
Figure 7. Nord Stream use case. Details of two Sentinel-1 SAR images showing the vessel of interest. (a) SAR image of the vessel located as marked in Figure 6(a); (b) SAR image of the vessel located as marked in Figure 6(b). TA and TB are the acquisition times of the SAR images, respectively, in (a) and (b). Magenta dots represent the vessel’s locations as reported by the AIS, while the blue stars indicate the vessel’s interpolated locations at time TA and time TB . Dashed lines reproduce the vessel’s trajectories obtained by interpolating the AIS reported locations.
of one of these fishing boats is reported in Figure 8, and demonstrates that the path of the underwater cable was crossed several times in few days. Over the same period, a cargo vessel was loitering in the area crossing the cable’s path few times; however, it is also reported that this vessel left the area few days before the accident. Although the damage of the SHEFA-2 cable is widely considered to be accidental, it is a clear example of how a malicious coordinated attack targeting multiple links of a network could isolate a territory, thus causing major disruptions.
ADRIATIC SEA At the end of 2022, several Italian news agencies reported the anomalous behavior of a large vessel entering the Strait of Otranto, Adriatic Sea, and stopping over the underwater pipeline TAP and the underwater cable OteGlobe for a few hours [3]. Figure 9 shows its full trajectory as reported by the AIS from 19 December to 23 December, and a detail of its route in the Strait of Otranto from the evening of 20 December to the afternoon
Figure 8. Shetland Islands use case. The main panel shows the full trajectories of a fishing vessel and a cargo vessel in the region of interest since few weeks before the SHEFA-2 cable cutoff. The inserts show details of the trajectories spanning five days. The color of the trajectories (cyan to magenta, and green to yellow) is related to the vessels’ speed (slower to faster).
18
IEEE A&E SYSTEMS MAGAZINE
OCTOBER 2023
Soldi et al. of 22 December. The illustration also reports the paths of the TAP and the OteGlobe as reported by open sources, which, however, may deviate from their actual routes. The dashed black lines highlight a gap of more than 21 hours in the AIS data available for the vessel of interest on 21 December. Given that the use of the AIS system is mandatory at all times, this gap might suggest an intentional switch off of the AIS system, probably to hide its transit over the OteGlobe cable and other critical infrastructures. On this same day, a Sentinel-1 SAR image was acquired over the strait about 3 hours after the latest available AIS position. Given this large time interval, the vessel’s position at the SAR acquisition time is inferred from the latest available AIS position by means of a longterm prediction tool for nonmaneuvering targets that uses the OU process to model the ship’s kinematic [84], [86]; the vessel-of-interest’s long-term predicted position is depicted as an orange star, and its uncertainty is represented by an orange circle. The SAR image is reported in Figure 10 along with all the detected ships in the area, represented by circles, and all the vessels’ trajectories as reported by the AIS; by
performing the same AIS/SAR association procedure mentioned in the “Nord Stream” section, the color of a circle matches the color of the AIS trajectory which it is associated to, while white circles represent SAR detections that are not associated to any AIS trajectory. Longterm predicted position and uncertainty of the vessel of interest are reported in the top-right panel of Figure 10. Interestingly, the dashed orange circle representing the uncertainty of the prediction encloses two detected vessels that appear to be stationary in close proximity of the underwater cable OteGlobe; these vessels are not associated with any other AIS trajectories, thus suggesting that one of them might actually be the large vessel reported by the news agencies.
CONCLUSION CUIs are fundamental in many fields of everyday life, ranging from telecommunications to energy, economy, and finance. Following the explosions affecting the Nord Stream gas pipelines on 26 September 2022, there has been
Figure 9. Adriatic Sea use case. Full trajectory of the vessel of interest from 19 December to 23 December, 2022. The color of the trajectory (cyan to magenta) is related to the vessel’s speed (slower to faster); gray arrows indicate the direction of the vessel when entering and exiting the Strait of Otranto. The orange star in the smaller panel represents the vessel’s long-term predicted position at the acquisition time of the Sentinel-1 SAR image recorded over the strait on 21 December ; the SAR image is shown in Figure 10.
OCTOBER 2023
IEEE A&E SYSTEMS MAGAZINE
19
Monitoring of Critical Undersea Infrastructures: The Nord Stream and Other Recent Case Studies
Figure 10. Adriatic Sea use case. Sentinel-1 SAR image acquired over the Strait of Otranto on 21 December, 2022. Colored solid lines are vessels’ trajectories obtained by interpolating the AIS positions provided by the vessels themselves. Circles are ships as detected within the SAR image: the color of the circle identifies the detection as coming from a vessel whose AIS trajectory has the same color; a white circle indicates that the detection is not associated with any AIS trajectory. Top-right panel reports the vessel-of-interest’s long-term predicted position and its uncertainty with an orange star and dashed circle, respectively.
increasing attention by authorities and international agencies, toward the improvement of CUIs resilience. In this article, we have shown how analyses of the data from AIS and satellite data, e.g., SAR, in conjunction with contextual information, such as bathymetry and weather data, can highlight anomalous and suspicious behaviors. Moreover, we have presented how AI and IF can fuse information generated from a large variety of assets, e.g., AIS and SAR, and use the learned knowledge to anticipate future behaviors and/or identify threats and critical situations toward CUIs.
problem, and MarineTraffic for providing the real-world AIS data set used in Figures 9–10. Financial support from the Office of the NATO Chief Scientist and the NATO Allied Command Transformation (ACT) is greatly acknowledged.
REFERENCES [1] BBC, “Nord Stream: Ukraine accuses Russia of pipeline terror attack,” Sep. 2022. [Online]. Available: https:// www.bbc.com/news/world-europe-63044747 [2] M. Jia et al., “The Nord Stream pipeline gas leaks released approximately 220,000 tonnes of methane into the atmos-
ACKNOWLEDGMENTS
phere,” Environ. Sci. Ecotechnol., vol. 12, Oct. 2022,
The views expressed in this research paper are the responsibility of the authors and do not necessarily reflect the opinions of the NATO. The authors would like to thank Dr. Alvarez for his hints and the useful discussions about the operational 20
Art. no. 100210, doi: 10.1016/j.ese.2022.100210. [3] “La flotta russa a Otranto minaccia il gasdotto Tap,” la Repubblica, Dec. 2022. [Online]. Available: https://www.
IEEE A&E SYSTEMS MAGAZINE
repubblica.it/esteri/2022/12/22/news/otranto_flotta_russia_ gasdotto_tap-380270375/
OCTOBER 2023
Soldi et al.
[4] G. Soldi et al., “Space-based global maritime surveillance.
[17] R. Albert, H. Jeong, and A.-L. Barabasi, “Error and attack
Part I: Satellite technologies,” IEEE Aerosp. Electron.
tolerance of complex networks,” Nature, vol. 406,
Syst. Mag., vol. 36, no. 9, pp. 8–28, Sep. 2021, doi: 10.1109/MAES.2021.3070862. [5] K. Hicke and K. Krebber, “Towards efficient real-time submarine power cable monitoring using distributed fibre optic
no. 6794, pp. 378–382, Jul. 2000, doi: 10.1038/35019019. [18] Z. Kong and E. M. Yeh, “Resilience to degree-dependent and cascading node failures in random geometric networks,” IEEE Trans. Inf. Theory, vol. 56, no. 11,
acoustic sensors,” in Proc. 25th Opt. Fiber Sensors Conf.,
pp.
Jeju, South Korea, 2017, pp. 1–4, doi: 10.1117/12.2267474.
TIT.2010.2068910.
5533–5546,
Nov.
2010,
doi:
10.1109/
[6] R. H. Littlefield, K. Soenen, G. Packard, and J. Kaeli,
[19] A. E. Motter and Y.-C. Lai, “Cascade-based attacks on
“Seafloor cable based navigation and monitoring with
complex networks,” Phys. Rev. E, vol. 66, no. 6, Dec.
autonomous underwater vehicles,” in Proc. OCEANS,
2002, Art. no. 065102, doi: 10.1103/PhysRevE.66.065102.
Seattle, WA, USA, 2019, pp. 1–5, doi: 10.23919/ OCEANS40490.2019.8962587.
[20] T. Tanizawa, G. Paul, R. Cohen, S. Havlin, and H. E. Stanley, “Optimization of network robustness to waves of targeted and random attacks,” Phys. Rev. E, vol. 71, no. 4,
[7] G. Soldi et al., “Space-based global maritime surveillance. Part II: Artificial intelligence and data fusion techniques,”
Apr.
IEEE Aerosp. Electron. Syst. Mag., vol. 36, no. 9, pp. 30–
PhysRevE.71.047101.
2005,
Art.
no.
047101,
doi:
10.1103/
[21] S. Rinaldi, J. Peerenboom, and T. Kelly, “Identifying,
42, Sep. 2021, doi: 10.1109/MAES.2021.3070884. [8] D. Gaglione, P. Braca, G. Soldi, F. Meyer, F. Hlawatsch,
understanding, and analyzing critical infrastructure inter-
and M. Z. Win, “Fusion of sensor measurements and tar-
dependencies,” IEEE Control Syst. Mag., vol. 21, no. 6,
get-provided information in multitarget tracking,” IEEE Trans. Signal Process., vol. 70, pp. 322–336, Dec. 2022,
pp. 11–25, Dec. 2001, doi: 10.1109/37.969131. [22] “UNCLOS: The Law of the Sea in the 21st century,” House of Lords, International Relations and Defence Committee’s
doi: 10.1109/TSP.2021.3132232.
2nd Report of Session 2021–2022, Feb. 2022.
[9] D. Gaglione et al., “Bayesian information fusion and multitarget tracking for maritime situational awareness,” IET Radar
[23] G. Vivone, P. Braca, and J. Horstmann, “Knowledge-based
Sonar Navigation, vol. 14, no. 12, pp. 1845–1857, Dec. 2020,
multitarget ship tracking for HF surface wave radar sys-
doi: 10.1049/iet-rsn.2019.0508.
tems,” IEEE Trans. Geosci. Remote Sens., vol. 53, no. 7,
[10] P. Kowalski and A.-L. Jousselme, “Context-awareness for
pp. 3931–3949, Jul. 2015, doi: 10.1109/TGRS.2014.
information correction and reasoning in evidence theory,” Int. J. Approx. Reason., vol. 153, pp. 29–48, Feb. 2023,
2388355. [24] G. F. Aureliano, High Frequency Over-the-Horizon Radar: Fundamental Principles, Signal Processing, and Practical
doi: 10.1016/j.ijar.2022.11.009.
Applications. New York, NY, USA: McGraw Hill, 2013.
[11] BBC, “Damaged cable leaves Shetland cut off from mainland,” Oct. 2022. [Online]. Available: https://www.bbc.com/
[25] M. Q. Li, J. Xu, X. Zhou, L. C. Qian, T. Long, and M. M.
news/uk-scotland-north-east-orkney-shetland-63326102
Bian, “OTHR highly maneuvering target detection via
[12] C. Bueger, T. Liebetrau, and J. Franken, “Security threats
generalized Radon-Fourier transform,” in Proc. CIE Int.
to undersea communications cables and infrastructure—
Conf. Radar, Guangzhou, China, 2016, pp. 1–4,
Consequences for the EU,” Eur. Parliament—In-Depth Anal., Jun. 2022.
doi: 10.1109/RADAR.2016.8059163. [26] P. Braca, S. Maresca, R. Grasso, K. Bryan, and J. Horstmann,
[13] A. Brzozowski, “NATO seeks ways of protecting undersea
“Maritime surveillance with multiple over-the-horizon HFSW
cables from Russian attacks,” Euractiv, Oct. 2022.
radars: An overview of recent experimentation,” IEEE Trans.
[Online].
Aerosp. Electron., vol. 30, no. 12, pp. 4–18, Dec. 2015,
Available:
https://www.euractiv.com/section/
doi: 10.1109/MAES.2015.150004.
defence-and-security/news/nato-seeks-ways-of-protecting-
[27] A. Ponsford and J. Wang, “A review of high frequency surface
undersea-cables-from-russian-attacks/ [14] C. Gallagher and S. Carswell, “Russian naval drill to still take
wave radar for detection and tracking of ships,” Turk. J. Elect.
place over vital cables, experts believe,” The Irish Times, Jan. 2022. [Online]. Available: https://www.irishtimes.com/news/
Eng. Comput. Sci., vol. 18, no. 3, pp. 409–428, May 2010, doi: 10.3906/elk-0912-331.
environment/russian-naval-drill-to-still-take-place-over-vital-
[28] S. Maresca, P. Braca, J. Horstmann, and R. Grasso, “Maritime surveillance using multiple high-frequency sur-
cables-experts-believe-1.4789421 [15] P. Pourbeik, P. Kundur, and C. Taylor, “The anatomy of a
face-wave radars,” IEEE Trans. Geosci. Remote Sens.,
power grid blackout—Root causes and dynamics of recent
vol. 52, no. 8, pp. 5056–5071, Aug. 2014, doi: 10.1109/ MAES.2015.150004.
major blackouts,” IEEE Power Energy Mag., vol. 4, no. 5, pp. 22–29, Sep./Oct. 2006, doi: 10.1109/MPAE.2006.1687814.
[29] International Association of Marine Aids to Navigation
[16] J. V. Milanovic and W. Zhu, “Modeling of interconnected critical infrastructure systems using complex network the-
and Lighthouse Authorities, “IALA technical clarifications to Recommendation ITU-R M.1371-1,” Tech. Rep.
ory,” IEEE Trans. Smart Grid, vol. 9, no. 5, pp. 4637–4648,
ITU-R WP8B, Mar. 2006. [Online]. Available: https://
Sep. 2018, doi: 10.1109/TSG.2017.2665646.
www.itu.int/md/R03-WP8B-C-0393
OCTOBER 2023
IEEE A&E SYSTEMS MAGAZINE
21
Monitoring of Critical Undersea Infrastructures: The Nord Stream and Other Recent Case Studies
[30] International Maritime Organization, “International Convention for the Safety of Life at Sea (SOLAS),” Dec. 2000.
[44] D. B. Reid, “An algorithm for tracking multiple targets,” IEEE Trans. Autom. Control, vol. 24, no. 6, pp. 843–854, Dec. 1979, doi: 10.1109/TAC.1979.1102177.
[Online]. Available: https://wwwcdn.imo.org/localresources/ en/KnowledgeCentre/IndexofIMOResolutions/MSCResoluti ons/MSC.99(73).pdf
[45] C. Chong, S. Mori, and D. B. Reid, “Forty years of multiple hypothesis tracking—A review of key developments,”
[31] “Commission Directive 2011/15/EU,” Feb. 2011. [Online].
in Proc. 21st Int. Conf. Inf. Fusion, Cambridge, U.K., 2018, pp. 452–459, doi: 10.23919/ICIF.2018.8455386.
Available: https://eur-lex.europa.eu/eli/dir/2011/15/oj [32] E. d’Afflisio, P. Braca, L. M. Millefiori, and P. Willett,
[46] R. P. S. Mahler, “Multitarget Bayes filtering via first-order
“Detecting anomalous deviations from standard maritime
multitarget moments,” IEEE Trans. Aerosp. Electron.
routes using the Ornstein-Uhlenbeck process,” IEEE
Syst., vol. 39, no. 4, pp. 1152–1178, Oct. 2003, doi: 10.1109/TAES.2003.1261119.
Trans. Signal Process., vol. 66, no. 24, pp. 6474–6487, Dec. 2018, doi: 10.1109/TSP.2018.2875887. [33] E. d’Afflisio, P. Braca, and P. Willett, “Malicious AIS spoof-
[47] R. Mahler, “PHD filters of higher order in target number,” IEEE Trans. Aerosp. Electron. Syst., vol. 43, no. 4,
ing and abnormal stealth deviations: A comprehensive statis-
pp.
tical framework for maritime anomaly detection,” IEEE
TAES.2007.4441756.
Trans. Aerosp. Electron. Syst., vol. 57, no. 4, pp. 2093–2108,
1523–1543,
2007,
doi:
10.1109/
[48] B.-T. Vo, B.-N. Vo, and A. Cantoni, “The cardinality balanced
Aug. 2021, doi: 10.1109/TAES.2021.3083466.
multi-target
multi-Bernoulli
filter
and
its
implementations,” IEEE Trans. Signal Process., vol. 57,
[34] P. Braca, P. Willett, K. LePage, S. Marano, and V.
no.
Matta, “Bayesian tracking in underwater wireless sensor networks with port-starboard ambiguity,” IEEE Trans. Signal Process., vol. 62, no. 7, pp. 1864–1878,
Oct.
2,
pp.
409–423,
Feb.
2009,
doi:
10.1109/
TSP.2008.2007924. [49] G. Soldi, F. Meyer, P. Braca, and F. Hlawatsch, “Self-tuning algorithms for multisensor-multitarget tracking using
Apr. 2014, doi: 10.1109/TSP.2014.2305640. [35] G. Ferri et al., “Cooperative robotic networks for
belief propagation,” IEEE Trans. Signal Process., vol. 67,
underwater surveillance: An overview,” IET Radar
no. 15, pp. 3922–3937, Aug. 2019, doi: 10.1109/ TSP.2019.2916764.
Sonar Navigation, vol. 11, no. 12, pp. 1740–1761,
[50] R. Laxhammar, “Anomaly detection for sea surveillance,”
Dec. 2017, doi: 10.1049/iet-rsn.2017.0074. [36] M. Brambilla et al., “Cooperative localization and multitar-
in Proc. 11th Int. Conf. Inf. Fusion, Cologne, Germany,
get tracking in agent networks with the sum-product algorithm,” IEEE Open J. Signal Process., vol. 3, pp. 169–195,
2008, pp. 1–8. [51] A. Sidibe and G. Shu, “Study of automatic anomalous behaviour detection techniques for maritime vessels,” J.
Mar. 2022, doi: 10.1109/OJSP.2022.3154684.
Navigation, vol. 70, no. 4, pp. 847–858, Mar. 2017,
[37] M. P. Hayes and P. T. Gough, “Synthetic aperture sonar: A
doi: 10.1017/S0373463317000066.
review of current status,” IEEE J. Ocean. Eng, vol. 34, no.
3,
pp.
207–224,
Jul.
2009,
doi:
10.1109/
[52] N. Forti, E. d’Afflisio, P. Braca, L. M. Millefiori, S. Carniel, and P. Willett, “Next-Gen intelligent situational
JOE.2009.2020853. [38] F. Cheng, B. Chi, N. J. Lindsey, T. C. Dawe, and J. B.
awareness systems for maritime surveillance and autono-
Ajo-Franklin, “Utilizing distributed acoustic sensing and ocean bottom fiber optic cables for submarine structural
mous navigation [Point of View],” Proc. IEEE, vol. 110, no. 10, pp. 1532–1537, Oct. 2022, doi: 10.1109/
characterization,”
Sci.
Rep.,
vol.
11,
Mar.
JPROC.2022.3194445.
2021,
[53] R. O. Lane, D. A. Nevell, S. D. Hayward, and T. W. Bea-
Art. no. 5613, doi: 10.1038/s41598-021-84845-y. [39] A. Trafford et al., “Distributed acoustic sensing for active
ney, “Maritime anomaly detection and threat assessment,”
offshore shear wave profiling,” Sci. Rep., vol. 12, Jun.
in Proc. 13th Int. Conf. Inf. Fusion, Edinburgh, U.K., 2010, pp. 1–8, doi: 10.1109/ICIF.2010.5711998.
2022, Art. no. 9691, doi: 10.1038/s41598-022-13962-z. [40] Y. Shang et al., “Research progress in distributed acoustic
[54] M. Riveiro, G. Pallotta, and M. Vespe, “Maritime anomaly
sensing techniques,” Sensors, vol. 22, no. 16, Aug. 2022, Art. no. 6060, doi: 10.3390/s22166060.
detection: A review,” WIREs Data Min. Knowl. Discov., vol. 8, no. 5, Sep. 2018, Art. no. e1266, doi: 10.1002/ widm.1266.
[41] L. M. Millefiori, P. Braca, K. Bryan, and P. Willett, “Adaptive filtering of imprecisely time-stamped measurements with
[55] M. Ester, H.-P. Kriegel, J. Sander, and X. Xu, “A density-
application to AIS networks,” in Proc. 18th Int. Conf. Inf.
based algorithm for discovering clusters in large spatial
Fusion, Washington, DC, USA, 2015, pp. 359–365.
databases with noise,” in Proc. KDD-96, Portland, OR, USA, 1996, pp. 226–231.
[42] F. Meyer et al., “Message passing algorithms for scalable multitarget tracking,” Proc. IEEE, vol. 106, no. 2, pp. 221–
[56] M. Anneken, F. de Rosa, A. Kr€oker, A.-L. Jousselme, S.
259, Feb. 2018, doi: 10.1109/JPROC.2018.2789427. [43] Y. Bar-Shalom, P. K. Willett, and X. Tian, Tracking and
Robert, and J. Beyerer, “Detecting illegal diving and other suspicious activities in the North Sea: Tale of a successful
22
Data Fusion: A Handbook of Algorithms. Storrs, CT,
trial,” in Proc. 20th Int. Radar Symp., Ulm, Germany,
USA: YBS Publising, 2011.
2019, pp. 1–10, doi: 10.23919/IRS.2019.8768132.
IEEE A&E SYSTEMS MAGAZINE
OCTOBER 2023
Soldi et al.
[57] J. Dabrowski and J. De Villiers, “Maritime piracy situation
[69] M. Vespe, I. Visentini, K. Bryan, and P. Braca,
modelling with dynamic Bayesian networks,” Inf. Fusion,
“Unsupervised learning of maritime traffic patterns for
vol. 23, pp. 116–130, May 2015, doi: 10.1016/j.
anomaly detection,” in Proc. DFTT-12, London, U.K.,
inffus.2014.07.001. [58] R. O. Lane, D. A. Nevell, S. D. Hayward, and T. W. Bea-
2012, doi: 10.1049/cp.2012.0414. [70] C. Zor and J. Kittler, “Maritime anomaly detection in ferry
ney, “Maritime anomaly detection and threat assessment,”
tracks,” in Proc. IEEE Int. Conf. Acoust., Speech, Signal
in Proc. 13th Int. Conf. Inf. Fusion, Edinburgh, Scotland,
Process., New Orleans, LA, USA, 2017, pp. 2647–2651, doi: 10.1109/ICASSP.2017.7952636.
2010, pp. 1–8, doi: 10.1109/ICIF.2010.5711998. [59] B. Ristic, B. L. Scala, M. Morelande, and N. Gordon,
[71] K. Kowalska and L. Peel, “Maritime anomaly detection
“Statistical analysis of motion patterns in AIS data: Anom-
using Gaussian process active learning,” in Proc. 15th Int.
aly detection and motion prediction,” in Proc. 11th Int.
Conf. Inf. Fusion, Singapore, 2012, pp. 1164–1171.
Conf. Inf. Fusion, Cologne, Germany, 2008, pp. 1–7, doi: 10.1109/ICIF.2008.4632190.
[72] S. Mascaro, A. E. Nicholson, and K. B. Korb, “Anomaly detection in vessel tracks using Bayesian networks,” Int. J.
[60] X. Wang, X. Liu, B. Liu, E. N. de Souza, and S. Matwin,
Approx. Reason., vol. 55, no. 1, pp. 84–98, Jan. 2014,
“Vessel
route
anomaly
detection
with
doi: 10.1016/j.ijar.2013.03.012.
Hadoop
MapReduce,” in Proc. IEEE Int. Conf. Big Data, Wash-
[73] J. Llinas and J. Scrofani, “Foundational technologies for
ington, DC, USA, 2014, pp. 25–30, doi: 10.1109/
activity-based intelligence—A review of the literature,”
BigData.2014.7004464.
Naval Postgraduate School, Tech. Rep. NPS-EC-14-001, Feb. 2014.
[61] R. Zhen, J. Yongxing, Q. Hu, Z. Shao, and N. Nikitakos, “Maritime anomaly detection within coastal waters based on vessel trajectory clustering and naı¨ve Bayes classifier,”
[74] G. Levchuk, O. Mendoza-Schrock, and W. Shebilske, “Adversarial behavior recognition from layered and persis-
J. Navigation, vol. 70, no. 3, pp. 648–670, Jan. 2017,
tent sensing systems,” in Proc. SPIE-09, vol. 7347, Orlando, FL, USA, Apr. 2009, doi: 10.1117/12.818676.
doi: 10.1017/S0373463316000850. [62] N. Forti, E. d’Afflisio, P. Braca, L. M. Millefiori, P. Willett,
[75] G. Shafer, A Mathematical Theory of Evidence. Princeton, NJ, USA: Princeton Univ. Press, 1976.
and S. Carniel, “Maritime anomaly detection in a real-world scenario: Ever given grounding in the Suez Canal,” IEEE
[76] D. Mercier, E. Lefevre, and F. Delmotte, “Belief functions
Trans. Intell. Transp. Syst., vol. 23, no. 8, pp. 13904–13910,
contextual discounting and canonical decompositions,” Int.
Aug. 2022, doi: 10.1109/TITS.2021.3123890. [63] N. Forti, L. M. Millefiori, and P. Braca, “Hybrid Bernoulli
J. Approx. Reason., vol. 53, no. 2, pp. 146–158, Feb. 2012, doi: 10.1016/j.ijar.2011.06.005.
filtering for detection and tracking of anomalous path
[77] F. Pichon, D. Mercier, E. Lefevre, and F. Delmotte,
deviations,” in Proc. 21st Int. Conf. Inf. Fusion, Cam-
“Proposition and learning of some belief function contex-
bridge, U.K., 2018, pp. 1178–1184, doi: 10.23919/
tual correction mechanisms,” Int. J. Approx. Reason.,
ICIF.2018.8455567.
vol.
72,
pp.
4–42,
May
2016,
doi:
10.1016/j.
ijar.2015.12.012.
[64] N. Forti, L. M. Millefiori, P. Braca, and P. Willett, “Random finite set tracking for anomaly detection in the presence of
[78] P. P. Shenoy, “Using Dempster-Shafer’s belief function
clutter,” in Proc. IEEE Radar Conf., Florence, Italy, 2020, pp. 1–6, doi: 10.1109/RadarConf2043947.2020.9266705.
theory in expert systems,” in Advances in the DempsterShafer Theory of Evidence, M. Fedrizzi, J. Kacprzyk, and
[65] N. Forti, L. M. Millefiori, P. Braca, and P. Willett, “Bayesian
R. R. Yager, Eds. New York, NY, USA: Wiley, 1994, pp. 395–414.
filtering for dynamic anomaly detection and tracking,” IEEE Trans. Aerosp. Electron. Syst., vol. 58, no. 3, pp. 1528–1544,
[79] P.
Kowalski,
M.
Zocholl,
and
A.-L.
Jousselme,
“Explaining the impact of source behaviour in evidential
Jun. 2022, doi: 10.1109/TAES.2021.3122888.
reasoning,” Inf. Fusion, vol. 81, pp. 41–58, May 2022,
[66] F. Mazzarella, M. Vespe, A. Alessandrini, D. Tarchi, G.
doi: 10.1016/j.inffus.2021.11.007.
Aulicino, and A. Vollero, “A novel anomaly detection approach to identify intentional AIS on-off switching,” Expert Syst. Appl., vol. 78, pp. 110–123, Jul. 2017,
[80] P. Kowalski and A.-L. Jousselme, “Reasoning with conceptual graphs and evidential networks for multi-entity maritime threat assessment,” in Proc. 15th Int. Conf. Inf.
doi: 10.1016/j.eswa.2017.02.011.
Fusion, Link€oping, Sweden, 2022, pp. 1–8, doi: 10.23919/
[67] K. Tserpes, K. Chatzikokolakis, D. Zissis, G. Spiliopoulos,
FUSION49751.2022.9841301.
and I. Kontopoulos, “Real-time maritime anomaly detection: Detecting intentional AIS switch-off,” Int. J. Big
[81] ESA, “Copernicus open access hub,” [Online]. Available: https://scihub.copernicus.eu
Data Intell., vol. 7, no. 2, pp. 85–96, May 2020,
[82] ESA, “Sentinel application platform.” [Online]. Available:
doi: 10.1504/IJBDI.2020.10029526. [68] F. Katsilieris, P. Braca, and S. Coraluppi, “Detection of malicious AIS position spoofing by exploiting radar
https://step.esa.int/main/download/snap-download [83] F. Filipponi, “Sentinel-1 GRD preprocessing workflow,” in
information,” in Proc. 16th Int. Conf. Inf. Fusion, Istanbul,
Proc. Int. Electron. Conf. Remote Sens., May/Jun. 2019,
Turkey, 2013, pp. 1196–1203.
p. 11, doi: 10.3390/ECRS-3-06201.
OCTOBER 2023
IEEE A&E SYSTEMS MAGAZINE
23
Monitoring of Critical Undersea Infrastructures: The Nord Stream and Other Recent Case Studies
[84] L. M. Millefiori, P. Braca, K. Bryan, and P. Willett,
[86] L. M. Millefiori, P. Braca, and P. Willett, “Consistent esti-
“Modeling vessel kinematics using a stochastic mean-revert-
mation of randomly sampled Ornstein–Uhlenbeck process
ing process for long-term prediction,” IEEE Trans. Aerosp.
long-run mean for long-term target state prediction,” IEEE
Electron. Syst., vol. 52, no. 5, pp. 2313–2330, Oct. 2016, doi: 10.1109/TAES.2016.150596.
Signal Process. Lett., vol. 23, no. 11, pp. 1562–1566, Nov. 2016.
[85] M. D. Graziano, A. Renga, and A. Moccia, “Integration of
[87] A. Martin, “Fishing vessel, not sabotage, to blame for
24
automatic identification system (AIS) data and single-chan-
Shetland Island submarine cable cut,” The Record, Oct.
nel synthetic aperture radar (SAR) images by SAR-based
2022. [Online]. Available: https://therecord.media/fishing-
ship velocity estimation for maritime situational awareness,”
vessel-not-sabotage-to-blame-for-shetland-island-
Remote Sens., vol. 11, no. 19, Sep. 2019, Art. no. 2196.
submarine-cable-cut
IEEE A&E SYSTEMS MAGAZINE
OCTOBER 2023
Feature Article:
DOI. No. 10.1109/MAES.2023.3266179
Efficient Transmitter Selection Strategies for Improved Information Gathering of Aerial Vehicle Navigation in GNSS-Denied Environments Alexander A. Nguyen , University of California, Irvine, CA 92697 USA Zaher M. Kassas , The Ohio State University, Columbus, OH 43210 USA
INTRODUCTION Modern aerial vehicle navigation systems, whether lowaltitude unmanned aerial vehicles (UAVs) or high-altitude aircraft, rely on global navigation satellite system (GNSS) signals [1]. However, relying on GNSS alone does not yield a continuous flow of resilient position, velocity, and time estimates. In recent years, GNSS radio frequency interference (RFI) incidents have increased dramatically, threatening the safety of flight operations [2], calling for a reliable alternative to GNSS signals in the event that these signals become unusable [3]. Signals of opportunity (SOPs) [4], [5], [6], [7] have been the subject of extensive research, where they have shown promise to be a standalone navigation alternative to GNSS. SOPs can be terrestrial-based (e.g., FM radio [8], [9], cellular [10], [11], [12], [13], and digital television [14], [15]) or space-based (e.g., low Earth orbit (LEO) satellites [16], [17], [18]). Among terrestrial SOPs, cellular signals have shown the most promise for aerial vehicle navigation [19], achieving submeter-level accuracy in a standalone [20] and differential [21] navigation fashion. Assessing terrestrial SOPs on low-altitude UAVs and high-altitude aircraft has been considered in the context of channel modeling, communication, and navigation [22], [23], [24], [25]. Of particular note is the recent week-long flight campaign by the Autonomous Systems Perception, Intelligence, and Navigation (ASPIN) Laboratory in collaboration with the U.S. Air Force (USAF) to study the potential of
Authors’ current address: Alexander A. Nguyen is with the University of California, Irvine, CA 92697 USA (email: [email protected]). Zaher M. Kassas is with The Ohio State University, Columbus, OH 43210 USA (email: [email protected]). Manuscript received 11 November 2022; accepted 6 April 2023, and ready for publication 12 April 2023. Review handled by Daniel O’Hagan. 0885-8985/23/$26.00 ß 2023 IEEE 26
cellular SOPs for high-altitude aircraft navigation. This campaign, called “SNIFFER: Signals of opportunity for Navigation In Frequency-Forbidden EnviRonments,” revealed that terrestrial cellular SOPs can be acquired and tracked at altitudes reaching 23,000 ft above ground level and at horizontal distances of more than 100 km away, and could yield meterlevel accurate navigation solutions without GNSS [26], [27]. At high altitudes, it was recently discovered that more than a hundred cellular SOPs can be acquired and tracked, from which pseudorange measurements can be extracted [28]. Tracking all such SOPs simultaneously could be formidable on platforms with limited size, weight, power, and cost (SWaP-C) or unnecessary, since tracking a subset of the SOPs could yield a comparable performance. As such this article considers the transmitter selection problem, where an aerial vehicle is tasked with selecting a subset of the SOPs, to minimize the receiver’s computational strain. Figure 1 illustrates a real-world scenario in which the problem of transmitter selection was encountered. Here, the white pins denote M ¼ 57 cellular SOPs which the aerial vehicle-mounted receiver was able to acquire and track. A similar problem to transmitter selection has been studied in the literature in the context of sensor selection for target tracking [29], [30]. Sensor selection problems are typically formulated as an integer programming (IP) problem, which is difficult to solve in a computationally efficient fashion, while finding the optimal solution via exhaustive search becomes formidable for large sensor networks. To circumvent this, several heuristic and suboptimal algorithms have been proposed. Some approaches formulate the sensor selection problem as a convex or nonconvex optimization problem [31], [32], [33], [34], [35]. Others have approached this problem as a greedy sensor selection by leveraging the notion of submodularity [36], [37] or by utilizing the Fisher information matrix (FIM) [38], [39]. It is rather difficult to explicitly solve the transmitter selection optimization problem efficiently for a large number of transmitters due to the integer constraints. As such, this article aims to extend the findings in [40] by proposing two
IEEE A&E SYSTEMS MAGAZINE
OCTOBER 2023
Image licensed by Ingram Publishing
suboptimal, computationally efficient transmitter selection strategies, termed opportunistic greedy selection (OGS) and one-shot selection (OSS). The proposed strategies are simple, yet highly effective at selecting the “best” transmitters without explicitly solving the IP problem. The proposed strategies exploit the additive, iterative properties of the FIM, where the OGS selects the most informative transmitters in finite iterations (i.e., recursively), while the OSS selects in one iteration (i.e., batch). Numerical simulations are presented analyzing the performance of the proposed selection strategies, where it is concluded that OGS performs closely to the optimal selection, while executing in a fraction of the optimal selection’s time. Experimental results are also presented for a USAF high-altitude aircraft navigating without GNSS in a rural and a semiurban region. The effectiveness of the selected SOPs on the navigation performance is also demonstrated. The position
root-mean-squared error (RMSE) with the optimal, OGS, and OSS were 4.53, 6.28, and 7.13 m in the rural region; and 5.83, 6.08, and 6.70 m in the semiurban region for an aircraft traversing a trajectory of 1.48 and 1.22 km, respectively. It is important to highlight that the SOP selection subset was found to be valid over a trajectory of several kilometers, since the aerial vehicle-to-SOP geometry is approximately stationary for sufficiently faraway SOPs. The rest of this article is organized as follows. The “Problem Description” section overviews the considered problem. The “Problem Formulation” section formulates the transmitter selection problem. The “Transmitter Selection Framework” section presents the selection strategies. The “Selection Strategy Analysis” section analyzes the proposed selection strategies. We then present the “Simulation Results,” section. The “Experimental Results” section provides
Figure 1. Motivating example. M ¼ 57 terrestrial SOP transmitters (white) in the environment relative to the aerial vehicle’s selection point (black cross). What is the “best” (most informative) K < M subset of SOPs for navigation?
OCTOBER 2023
IEEE A&E SYSTEMS MAGAZINE
27
Efficient Transmitter Selection Strategies for Improved Information Gathering
Figure 2. Problem description. An aerial vehicle is equipped with a receiver capable of extracting pseudorange measurements from terrestrial SOPs. During flight, the aerial vehicle experiences GNSS outage. A nearby uplink station sends SOP map data to the aerial vehicle, which contains the locations of M terrestrial SOPs. The aerial vehicle’s selects the “best” (most informative) K < M to use to continue navigating.
experimental results for an aerial vehicle flying in two different regions. Finally, we present the “Conclusions” section.
PROBLEM DESCRIPTION To motivate the transmitter selection problem, consider the scenario depicted in Figure 2, in which an aerial vehicle is navigating in an environment comprising M terrestrial SOP transmitters. During flight, the aerial vehicle experiences GNSS outage (e.g., due to RFI or spoofing). The aerial vehicle is assumed to have a map of the SOP locations (e.g., loaded prior to flight or transmitted from a nearby uplink station). The aerial vehicle is assumed to be equipped with receivers capable of extracting pseudorange measurements from the SOPs, which will be used instead of GNSS to navigate the aerial vehicle. Due to SWaP-C constraints (e.g., limited payload, processing power, etc.), the aerial vehicle can only use signals from (K < M) SOPs. What is the “best” (most informative) SOP subset to select?
PROBLEM FORMULATION This section formulates the transmitter selection problem.
is the 3D position vector of the ith SOP, dtr is the aerial vehicle-mounted receiver’s clock bias, dtsi is the ith SOP’s clock bias, c is the speed of light, and vsi is the ith SOP’s measurement noise, which is modeled as a zeromean white Gaussian sequence with variance s 2si and is assumed to be independent across all SOPs. The dynamics of the clock error is described in Appendix A.
FISHER INFORMATION MATRIX The proposed transmitter selection strategies aim to choose the most informative measurements, which motivates adopting the FIM defined as " # @ln pðzzjx xÞ @ln pðzzjx xÞ T Iðx xÞ ¼ E @x x @x x where, pðzzjx xÞ is the likelihood function of the measurements z parameterized by the states x. Since the noise in the measurement model (1) is assumed to be independent across all SOPs, the FIM can be written as the prior FIM plus a summation of the information content associated with each measurement, that is, M X 1 @hi ðx xÞ @hi ðx xÞ T Iðx xÞ ¼ I0 ðx xÞ þ s2 @x x @x x i¼1 si
PSEUDORANGE MEASUREMENT MODEL
¼ I0 ðx xÞ þ
Consider an aerial vehicle equipped with an onboard receiver capable of extracting pseudorange measurements from M terrestrial SOPs in the environment. The pseudorange measurement made by the receiver to the ith SOP, after discretization and mild approximations, is modeled as zsi ðkÞ ¼ rr ðkÞ rsi 2 þ c dtr ðkÞ dtsi ðkÞ þvsi ðkÞ |fflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflffl{zfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflffl} hi ½x ðkÞ
(1) T
where, rr ¼ ½xr ; yr ; zr is the three-dimensional (3D) position vector of the aerial vehicle, rsi ¼ ½xsi ; ysi ; zsi T 28
M X
Ii ðx xÞ:
(2)
i¼1
The additive property of information from different sources [41] will be utilized in the proposed transmitter selection strategies. Denoting the (prior) information content associated with a subset of SOPs as I0 ðx xÞ and the information associated with the ith SOP as Ii ðx xÞ, the (posterior) information content associated with updating the SOP subset to include the ith SOP is defined as Iposterior;i ðx xÞ ¼ I0 ðx xÞ þ Ii ðx xÞ. Appendix B relates the FIM to the horizontal dilution of precision (HDOP) metric, commonly used in positioning and navigation.
IEEE A&E SYSTEMS MAGAZINE
OCTOBER 2023
Nguyen and Kassas
ESTIMATION FRAMEWORK Generally, one needs to estimate the state vector x, which includes the aerial vehicle’s position rr and velocity r_ r as well as relative clock error states fx xclk;i gM i¼1 between the vehicle-mounted receiver and each SOP, namely h iT x , rTr ; r_ Tr ; DxTclk;1 ; . . . ; DxTclk;M _ r dt _ s Þ T: Dxclk;i , c ðdtr dtsi Þ; c ðdt i Formulating the transmitter selection problem with x 2 R6þ2 M results in a large-scale optimization problem. To scale down the problem, two simplifications are made. First, it is noted that terrestrial SOPs suffer from poor geometric diversity in the vertical direction (particular as seen by high-altitude aircraft). Therefore, relying exclusively on SOPs for 3D navigation leads to a large vertical dilution of precision [42], [43]. Hence, it is assumed that the aerial vehicle is equipped with an altimeter to determine its altitude. As such, in what follows, the problem is formulated to only consider the 2D (planar) aerial vehicle states. Second, only the position states of the aerial vehicle will be considered, leading to the redefined state vector x0 2 R2 . It will be demonstrated in the “Effect of Timing on the Optimal Transmitter Selection” section that this simplification, which ignores the timing states, results in a negligible increase in position uncertainty (on the order of submeter). A static, weighted nonlinear least-squares (WNLS) estimator is employed on the redefined state vector x0 . The resulting Jacobian matrix Hr is given by 2 T T 3 Hr r
6 6 ¼6 4
r r rr s1 krr r rr s1 k2
.. .
rT rT r r sM
7 7 7: 5
(3)
krr r rr sM k2
The WNLS estimation error covariance matrix is given by h i1 T 1 Prr , P1 (4) 0;rr r þ Hr r R Hr r where, a prior of x0 may be given, denoted by x^0 , with an associated initial estimation error covariance ðP0;rrr ¼ 2 2 I1 0;rr r Þ and R ¼ diag½s s1 ; . . . ; s sM .
OPTIMAL TRANSMITTER SELECTION PROBLEM The optimal transmitter selection problem can be cast as the optimization problem minimize w
J ðw wÞ
subject to 1TM w ¼ K wi 2 f0; 1g;
i ¼ 1; . . . ; M
where, J ðw wÞ denotes a desired cost function [e.g., A-, D-, and E-optimality criterion or dilution of precision (DOP) [44], [45]], wi is a binary decision variable which determines whether to accept or reject the ith measurement, w ¼ ½w1 ; . . . ; wM T is a vector of the binary decision variables, OCTOBER 2023
1M 2 RM is a vector of ones, and K is the selection subset’s cardinality. This optimization problem is computationally involved to solve in real time due to the integer constraints. Instead of solving the abovementioned optimization problem, two efficient transmitter selection strategies are proposed in the next section.
TRANSMITTER SELECTION FRAMEWORK The proposed transmitter selection framework selects the most informative SOP subset to minimize the aerial vehicle’s position error uncertainty. According to the simplification discussed in the “Estimation Framework” section, only the information contribution from the ith SOP to the position states, denoted Irr ;i , is used to evaluate the cost function J ðw wÞ. Ergo, the cost function is defined as the A-optimality criterion: trace of the posterior position estimation error covariance (equivalently, trace of the inverse of FIM) h i1 J ðwÞ , tr I0;rrr þ Hr0Tr diagðw wÞHr0 r " #1 M X ¼ tr I0;rrr þ wi Irr ;i (5) i¼1 T where, Hr0 r , ðR1 a Þ Hr , Ra is the upper triangular Cholesky factorized measurement covariance (i.e., R ¼ RTa Ra ), and I0;rrr is the prior FIM corresponding to the receiver’s position states (see Figure 3). Algorithm 1 summarizes each of the proposed transmitter selection strategy’s steps.
Algorithm 1. Transmitter Selection Strategies Input: Prior FIM, FIM associated with each measurement, map of all SOPs, and number of SOPs to be selected Output: SOP selection subset and FIM for the selected SOPs 1: Define an empty set for SOP selection 2: Perform an exhaustive search to select the two SOPs with the largest information content 3: Update the prior FIM and SOP selection subset One-Shot Selection (OSS) 4: Compute the posterior FIM for all SOPs, excluding those already selected 5: Choose the K 2 SOPs which minimize the receiver’s average position error uncertainty 6: Compute the FIM for the selected SOPs (i.e., prior FIM plus all selected SOP’s FIM) and update the SOP selection subset 7: Return SOP selection subset and FIM for the selected SOPs Opportunistic Greedy Selection (OGS) for K 2 iterations 8: Compute the posterior FIM for all SOPs, excluding those already selected 9: Choose one SOP which minimizes the receiver’s average position error uncertainty 10: Redefine the prior FIM [i.e., (current) prior FIM plus selected SOP’s FIM] and update the SOP selection subset end for
IEEE A&E SYSTEMS MAGAZINE
29
Efficient Transmitter Selection Strategies for Improved Information Gathering
Figure 3.
^ 0 2 R2 ) in an environment comprised of M terrestrial SOPs. Receiver estimating its two position states (i.e., x
SELECTION STRATEGY ANALYSIS This section will compare the selection subsets of the proposed transmitter selection strategies and provide an upper bound on the FIM for the selected range-only measurements.
OSS VERSUS OGS SELECTION SUBSET COMPARISON The OSS and OGS will yield identical selection if the information content is scalar valued, that is, x 2 R is constrained to one dimension. This can be readily shown as follows. First, assume that the posterior FIM Iposterior;i ¼ I0;r þ Irr ;i are ordered from smallest to largest, such that J ðIw1 Þ J ðIw2 Þ . . . J ðIwK Þ . . . J ðIwM Þ, where wi is a vector of zeros with a one at the ith element. Therefore, the OSS will yield the optimal selection set S ¼ f1; 2; . . . ; Kg, where the information content associated with the selected transmitters is denoted as I ¼ d0 þ s21 þ . . . þ s2K . On the s1 sK other hand, the OGS recursively selects one transmitter (i.e., i ¼ 1; . . . ; M), which minimizes the cost function to yield " i ¼ argmin i
g0 þ
3 i þ s 2s3 s 2si
#1 :
This selection process is performed at each iteration. By virtue of the simplifying assumption, the minimum argument (resulting in the smallest cost function) is selected in ascending order as i ¼ 4; i ¼ 5; . . . ; i ¼ K since J ðw wi Þ J ðw wi Þ; 8i n fSg. Therefore, the transmitter selection subset will be S ¼ f1; 2; . . . ; Kg, where the information content associated with the selected transmitters is I ¼ d0 þ s21 þ . . . þ s2K , which is identical to the s1 sK OSS. 30
For the 2D case, that is, x 2 R2 , the OSS and OGS will yield different selections. To show this, one proceeds similarly to the scalar case, noting that in the ith iteration of the OGS
i ¼ argmin i2 =S
02 a2i 0 B6 g 11 þ s 2si trB @4 0 g 12 þ asi2bi si
g 012 þ asi2bi si
g 022
b2i
þ s2
31 1 7 C 5 C A:
si
This implies that the optimization problem’s solution is heavily dependent on the cross terms associated with the ith SOP’s posterior FIM. More specifically, each state’s information content is coupled with one another. This affects the 2 2 FIM inverse via the determinant term which incorporates the mutually shared information shared (cross terms) between the position state estimates. Therefore, J ðw wi ÞCJ ðw wi Þ; 8i n fSg, yielding a different selection subset than the OSS.
COMPUTATIONAL COMPLEXITY OF OSS VERSUS OGS If one attempts to find the optimal selection strategy by exhaustively searching, that is, M choose K algorithm: M M! K ¼ ðMKÞ!K! , the computational complexity is exponential, namely OðM K Þ for a large enough M, since M M K 1 K1 K ¼ K! ð1ð1 M Þ ð1 M ÞÞ where, K is assumed 1 K1 fixed and 1ð1 M Þ ð1 M Þ ! 1 as M ! 1. In contrast, the OGS computational complexity is OðM 2 Þ þ OðK 2Þ OðM 2 Þ, while the OSS computational complexity is OðM 2 Þ þ Oð1Þ OðM 2 Þ, for a large enough M, both of which are quadratic. These can be derived by noting that the exhaustive search step’s com 2 plexity is M2 ¼ M 2!M , where 2!1 ðM 2 MÞ ! M 2 as
IEEE A&E SYSTEMS MAGAZINE
OCTOBER 2023
Nguyen and Kassas
Figure 4. (a) BPP realization with M = 22 SOPs. (b) Parameterization of the ith SOP’s position.
M ! 1. For OGS, the recursive search step’s computational complexity is defined as OðK 2Þ because it requires K 2 iterations to complete. For OSS, the batch selection step’s computational complexity is defined as Oð1Þ because it only requires a single iteration to complete.
SIMULATION RESULTS This section presents simulation results demonstrating the efficacy of the proposed OGS and OSS strategies, which are compared against the optimal selection, whose solu tion is obtained by exhaustive search M K .
Table 1.
OPTIMAL SELECTION, OSS, AND OGS STRATEGY COMPARISON
Transmitter Selection Environment Simulation Settings
The aerial vehicle is assumed to have initial access to GNSS signals, leading to knowledge of its initial position estimates ^rr ð0Þ, after which the aerial vehicle looses access to GNSS. During GNSS availability, the aerial vehicle chooses the “best” K < M SOPs to use for navigation once GNSS signals are cut off. The cellular SOP network was modeled as a binomial point process (BPP), where the horizontal positions of the SOPs are independently and uniformly distributed over an annular region centered at the aerial vehicle’s current position O, that is, BO ðdmin ; dmax Þ ¼ pðd2max d2min Þ[46], where dmax is the maximum distance for which ranging signals can be detected by the receiver and dmin is the minimum distance required for the far-field assumption to hold (see Figure 4). The location of the ith SOP with respect to the aerial vehicle can be parameterized in terms of the range Ri and bearing angle ui . The simulation environment considered M ¼ 22 terrestrial SOPs, of which K 2 f6; 7; . . . ; 14g are to be selected. For each K, 103 Monte Carlo (MC) realizations were generated according to the simulation settings summarized in Table 1. The randomized MC realizations were the clock’s process noise, measurement noise, and SOPs’ OCTOBER 2023
locations. The three selection strategies (optimal, OSS, and OGS) were performed for each realization. Figure 5 compares the transmitter selection strategy performance for each K. The medium-sized green, blue, and red dots represent the cost function values for the optimal selection J ðw w Þ, OGS J ðw wOGS Þ, and OSS J ðw wOSS Þ, respectively, averaged over all MC realizations. The tiny green, blue, and red dots represent the cost function value for each MC realization. It can be seen that, on one hand, the OGS yielded very close solution to the optimal value (the medium green and red dots are nearly on top of each other). The zoom, in the following, shows the difference between both solutions [i.e., J ðw w Þ J ðw wOGS Þ]. On the other
Parameter
Value
M
22
K
f6; 7; . . . ; 14g
rr ð0Þ
½0; 0T
P0;rrr
102 I22
r^r ð0Þ
N ½rrr ð0Þ; Pr0;rrr
fRi ; ui g r si xclk;i ð0Þ fx xsi ð0ÞgM i¼1
fU½5; 80; 000 m, U½p; p rad g ½Ri cosðui Þ; Ri sinðui ÞT c ½9; 0:9T ½rrTsi ð0Þ; xTclk;i ð0ÞT
fP0;clk;si gM i¼1
diag½30 103 ; 0:3 103
f^ xclk;si ð0ÞgM i¼1
N ½x xclk;i ð0Þ; Px0;clk;si
fh0;r ; h2;r g
f8:0 1020 ; 4:0 1023 g
fh0;si ; h2;si gM i¼1
f2:6 1022 ; 4:0 1026 g
fs 2si gM i¼1
10 m2
T
0.01 s
IEEE A&E SYSTEMS MAGAZINE
31
Efficient Transmitter Selection Strategies for Improved Information Gathering
Figure 5.
Cost function point cloud with 103 MC realizations for optimal selection (green), OGS (red), and OSS (blue). The average cost function values over all MC realizations is shown as medium-sized dots, while the cost function value for each MC realization is represented as tiny dots. The top boxed values represent J ðw w Þ J ðw wOSS Þ, while bottom boxed values (in the zoom in) represent J ðw w Þ J ðw wOGS Þ.
minimize (5) with the FIM
hand, the OSS yielded solutions that were further than the optimal solution. Table 2 shows the average cost function values J ðw wÞ over all MC realizations along with the corresponding
1s. Note that the average cost function values for the OGS and optimal selection strategies are very close to each other with a low standard deviation, whereas the OSS strategy is prone to worse selection performance (i.e., larger cost function value) with a higher standard deviation.
Iðx x0 Þ ¼ I0;rrr ðx x0 Þ þ
M X wi a2i s 2 a i bi i¼1 si
a i bi : b2i
Next, the FIM that considers the full state vector x (comprising position, velocity, and timing states) was used to find the optimal selection, that is, minimize (5), but with the FIM Iðx xÞ ¼ I0 ðx xÞ 2 PM wi a2i PM wi ai bi i¼1 s 2s i¼1 s 2s 6 i i 6 6 PM wi ai bi PM wi b2i 6 i¼1 2 i¼1 s 2s 6 s si i 6 6 w1 a1 w1 b1 6 s 2s s 2s1 þ6 6 w a1 w 2 b2 2 2 6 6 s 2s2 s 2s2 6 6 .. .. 6 6 . . 4 w a w b
EFFECT OF TIMING ON THE OPTIMAL TRANSMITTER SELECTION A simulation was conducted to justify the simplification invoked in the “Estimation Framework” section, whereby only the aerial vehicle’s position states were considered, while ignoring the timing states. To this end, 250 MC realizations were generated, for each of which, the optimal transmitter selection strategy M was performed to K
K K s 2sK
K K s 2sK
w1 a1 s 2s1
w2 a2 s 2s2
...
w1 b1 s 2s1
w2 b2 s 2s2
...
w1 s 2s1
0
0
0
w2 s 2s2
0
.. . 0
.. .
..
0
0
wM aM s 2sM
.
3
7 7 wM bM 7 7 2 s sM 7 7 7 0 7 7: 7 0 7 7 7 .. 7 7 . 7 5 w K
s 2sK
Table 2.
Average Cost Function Values for the Transmitter Selection Strategies with K ¼ 6 14 6
7
8
9
10
11
12
13
14
J ðw w Þ ½ s
6.45 ½ 0:001
5.56 ½ 0:005
4.88 ½ 0:006
4.35 ½ 0:009
3.92 ½ 0:01
3.57 ½ 0:01
3.28 ½ 0:01
3.03 ½ 0:02
2.82 ½ 0:02
J ðw wOGS Þ ½ s OGS
6.47 ½ 0:02
5.62 ½ 0:02
4.89 ½ 0:01
4.38 ½ 0:01
3.93 ½ 0:01
3.59 ½ 0:02
3.29 ½ 0:02
3.04 ½ 0:02
2.83 ½ 0:03
J ðw wOSS Þ ½ s OSS
10.08 ½ 0:76
9.13 ½ 0:99
8.19 ½ 1:13
7.26 ½ 1:19
6.37 ½ 1:16
5.58 ½ 1:08
4.87 ½ 0:96
4.28 ½ 0:84
3.77 ½ 0:70
K
32
IEEE A&E SYSTEMS MAGAZINE
OCTOBER 2023
Nguyen and Kassas Table 3.
Reduction in Average Position Uncertainty Due to Optimizing the FIM Iðx xÞ instead of Iðx x0 Þ 6
7
8
9
10
11
12
13
14
0.2630
0.2593
0.2643
0.2568
0.2862
0.2183
0.2781
0.2178
0.2475
K Uncertainty reduction [m2 ]
Table 3 tabulates the reduction in position uncertainty (averaged over all MC realizations for each K value) upon including the timing error states in the FIM. It can be noted that the reduction in position uncertainty is small (on the order of submeter), which justifies the considered simplification.
EXPERIMENTAL RESULTS This section demonstrates the efficacy of the proposed algorithms to select a “manageable” subset of terrestrial SOPs to navigate an aircraft in a real-world environment.
HARDWARE AND SOFTWARE SETUP The SNIFFER flight campaign took place on a Beechcraft C12 Huron (called Ms. Mabel), a fixed-wing U.S. Air Force aircraft, flown by members of the USAF Test Pilot School (TPS) over two different regions as follows. i) A rural region located in Edwards, California, USA. ii) A semiurban region located in Palmdale, California, USA. The C-12 aircraft was equipped with a quad-channel universal software radio peripheral (USRP)-2955, three
consumer-grade 800/1900 MHz Laird cellular antennas, GPS antenna, a solid-state drive for data storage, PCIe cable, and a laptop computer running ASPIN Laboratory’s software-defined radio (SDR), called MATRIX: Multichannel Adaptive TRansceiver Information eXtractor, for real-time monitoring of the cellular signals [27]. The MATRIX SDR produced the navigation observables: Doppler frequency, carrier phase, and pseudorange, along with the corresponding carrier-to-noise ratio (C=N0 ). The experimental hardware setup is shown in Figure 6.
TRANSMITTER SELECTION AND NAVIGATION FILTER The OGS and OSS selection strategies were performed in each of the two regions. To demonstrate the efficacy of the selected transmitters for aircraft navigation, the pseudorange measurements from the selected transmitters are fused with altimeter measurements via an extended Kalman filter (EKF), as described in [27]. The navigation solution was computed over a flight segment during which the selection strategies remained valid. The EKF’s initial state vector was set as ^_ 1 ð0Þ; . . . ; cdt ^ 1 ð0Þ; cdt ^ K ð0Þ; x^ð0Þ ¼ ½^rr ð0ÞT ; ^r_ r ð0ÞT ; c dt _ T ^ cdtK ð0Þ with a corresponding initial estimation error
Figure 6. Hardware setup equipped to the C-12 aircraft.
OCTOBER 2023
IEEE A&E SYSTEMS MAGAZINE
33
Efficient Transmitter Selection Strategies for Improved Information Gathering
Figure 7. Rural region transmitter selection results with OGS and OSS strategies during the aircraft’s flight. (Map data: Google Earth.)
covariance Pð0Þ ¼ diag½102 I33 ; 10 I33 ; 108 ; 10; 8 . . . ; 10 ; 10. The clock error states of each SOP was initialized using the pseudorange measurements from the initial two time epochs. Specifically, the clock bias was initialized as c^dti ð0Þ ¼ zsi ð0Þ krrr ð0Þ rsi k2 and ^_ i ð0Þ the clock drift was initialized as cdt 1 ¼ T ½zsi ð1Þ zsi ð0Þ krrr ð1Þ rsi k2 þ krrr ð0Þ rsi k2 . The aircraft’s dynamics was assumed to evolve according to the simple, yet effective, velocity random walk model [47], with power spectra of the continuoustime acceleration noise in the East (E), North (N), and Up (U) directions set as q~E ¼ q~N ¼ 5 m2 /s3 and q~U ¼ 103 m2 /s3 , respectively. The receiver’s clock covariance Qclk;r was set to correspond to a low-quality temperature-compensated crystal oscillator (TCXO) with h0;r ¼ 2:0 1019 s and h2;r ¼ 2:0 1020 s1 . The SOPs’ clock covariance Qclk;si was set to correspond to a typical-quality oven-controlled crystal oscillator with h0;si ¼ 8:0 1020 s and h2;si ¼
4:0 1023 s1 . The time-varying measurement covariance R was proportional to the inverse of C=N0 and the sampling time was T ¼ 0:01 s.
FLIGHT REGION 1: RURAL The rural region was comprised of M ¼ 57 terrestrial cellular SOPs, where the aircraft was tasked with selecting the most “informative” K ¼ 15 SOPs to use for navigation. Figure 7 shows: i) selected SOPs from OGS (red pins), ii) selected SOPs from OSS (yellow pins), iii) selected SOPs from both OSS and OGS strategies (orange pins), and iv) and nonselected SOPs (white pins). Table 6 compares the snapshot performance (A-, D-, and E-optimality and HDOP metrics) of the OGS versus OSS selection. Upon selecting the SOPs, the aircraft navigated along the green trajectory in Figure 7 for 1.48 km. It should be noted the optimal solution (i.e., global minimizer) for
Table 4.
Experiment 1: Navigation Solution Performance in the Rural Region Selection Type
Pos. RMSE [m]
Vel. RMSE [m/s]
Max pos. error [m]
Max vel. error [m/s]
Run time [ms]
105 MC Runs
[4.53, 71.55]
[0.98, 7.61]
[10.50, 125.06]
[5.90, 11.46]
–
OGS
6.28
1.44
10.50
6.38
19.30
OSS
7.13
1.39
10.50
6.38
16.50
34
IEEE A&E SYSTEMS MAGAZINE
OCTOBER 2023
Nguyen and Kassas
Figure 8. Semiurban region transmitter selection results with optimal selection, OGS, and OSS strategies during the aircraft’s flight. (Map data: Google Earth.)
terrestrial SOP selection is infeasible to compute using the M K selection strategy due to its formidable run time. In light of this, 105 MC runs were performed in an attempt to capture a range of best-to-worst selections. Table 4 summarizes the navigation performance when using the transmitters selected by the OGS, OSS, and MC realizations. It is worth noting that the OGS returned relatively comparable performance to the best-case MC realization. It is also worth noting the worst-case MC realization yielding much larger error than the best-case realization (i.e., the variance is rather large), which further motivates the importance of transmitter selection.
FLIGHT REGION 2: SEMIURBAN The semiurban region was comprised of M ¼ 18 terrestrial cellular SOPs, where the aircraft was tasked with
selecting the most “informative” K ¼ 9 SOPs to use for navigation. The number of SOPs in this region was small enough to compute the optimal solution via the M K selection strategy to determine the global minimizer. Figure 8 shows: i) selected SOPs from M K (green pins), ii) selected SOPs from OGS (violet pins), iii) selected SOPs from OSS (yellow pins), iii) selected SOPs from both OSS and OGS strategies (orange pins), iv) selected SOPs from optimal, OSS, and OGS (blue pins), and v) nonselected SOPs (white pins). Table 7 compares the snapshot performance (A-, D-, and E-optimality and HDOP metrics) of the optimal versus OGS and OSS. Upon selecting the SOPs, the aircraft navigated along the green trajectory in Figure 8 for 1.22 km. Table 5 summarizes the navigation performance when using the transmitters selected by the optimal, OGS, and OSS. Note that the navigation performance with the OGS strategy is close
Table 5.
Experiment 2: Navigation Solution Performance in the Semiurban Region Selection type
Pos. RMSE [m]
Vel. RMSE [m/s]
Max pos. error [m]
Max vel. error [m/s]
Run time [ms]
Optimal Selection
5.84
1.45
10.80
5.90
700.30
OGS
6.08
1.42
10.80
5.90
5.30
OSS
6.70
1.35
10.80
5.90
3.90
OCTOBER 2023
IEEE A&E SYSTEMS MAGAZINE
35
Efficient Transmitter Selection Strategies for Improved Information Gathering Table 6.
Experiment 1: Snapshot Performance Metrics in Rural Region After Transmitter Selection Selection type
tr[Prr ]
log½detðPrr Þ
max ½Prr
HDOP
OGS
200.99
9.20
100.00
0.54
OSS
200.99
9.20
100.00
0.54
Table 7.
Experiment 2: Snapshot Performance Metrics in Semiurban Region After Transmitter Selection Selection type
tr[Prr ]
log½detðPrr Þ
max ½Prr
HDOP
Optimal selection
200.99
9.20
100.00
0.74
OGS
200.99
9.20
100.00
0.88
OSS
200.99
9.20
100.00
0.94
to that of the optimal selection, whereas the OSS strategy performed slightly worse. This further motivates using the computationally efficient OGS selection strategy instead of the computationally expensive optimal selection strategy, over this valid selection region.
CONCLUSION
_ cdt; cdt
T
;
1 T 0 1
Fclk ¼
2
Qclk ¼ c
3
2
Sw~ dt T þ Sw~ dt_ T3
Sw~ dt_ T2
Sw~ dt_ T2
Sw~ dt_ T
2
# :
(7)
The terms Sw~ dt and Sw~ dt_ are the clock bias and drift process noise power spectral densities, respectively, which can be related to the power-law coefficients, fhai g2ai ¼2 , which have been shown through laboratory experiments to characterize the power spectral density (PSD) of the fractional frequency deviation of an oscillator from nominal frequency according to Sw~ dt h20 and Sw~ dt_ 2p2 h2 . The receiver’s and SOPs’ process noise covariances Qclk;r and fQclk;si gM i¼1 are calculated from (7) using the PSDs associated with the receiver’s and SOPs’ oscillator quality, respectively.
APPENDIX B
APPENDIX A
RELATIONSHIP BETWEEN WEIGHTED HDOP AND INFORMATION CONTENT
CLOCK ERROR DYNAMICS The aerial vehicle-mounted receiver’s and SOP’s clock error states are assumed to evolve according to
36
_ is the clock drift, c is the where, dt is the clock bias, dt speed of light, T is the constant sampling interval, and wclk is the process noise, which is modeled as a discretetime white noise sequence with covariance "
This article proposed computationally efficient transmitter selection strategies to select the most informative terrestrial SOPs to use when navigating an aerial vehicle. The strategies exploited the additive, iterative properties of the FIM to minimize the vehicle’s average position error variance. Simulation results showed the OGS performance to be very close to the optimal selection, while executing in a fraction of the optimal selection’s time. Experimental results in a realworld environment were presented showing the efficacy of the OGS and OSS strategies in navigating a U.S. Air Force high-altitude aircraft with terrestrial cellular SOPs. The achieved position RMSE with the optimal, OGS, and OSS solutions were 4.53, 6.28, and 7.13 m in the rural region; and 5.83, 6.08, and 6.70 m in the semiurban region for an aircraft traversing a trajectory of 1.48 and 1.22 km, respectively.
xclk ðk þ 1Þ ¼ Fclk xclk ðkÞ þ wclk ðkÞ
xclk ,
(6)
DOP states how errors in the measurement will affect errors in the final estimates of the unknown quantities. The weighted HDOP matrix for the measurement vector z0 , ½z0s1 ; . . . ; z0sM T with an associated Jacobian matrix
IEEE A&E SYSTEMS MAGAZINE
OCTOBER 2023
Nguyen and Kassas H defined in (3) and measurement covariance R ¼ 1 T 1 diag½s 2s1 ; . . . ; s 02 sM , is defined as Dw , ½H R H , which has the form " Dw ¼
s 2x
s 2xy
s 2xy
s 2y
# (8)
:
pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi The weighted HDOP is trðDw Þ ¼ s 2x þ s 2y [48]. The weighted HDOP matrix can be related to information content by the inverse of the estimation error covariance matrix as 2 D1 w ¼
K X j¼1
¼
1 6 6 s 2sj 4
K X 1 2 s j¼1 sj
"
ðxr xsj Þ2
ðxr xsj Þðyr ysj Þ
ðxr xsj Þ2 þðyrysj Þ2
ðxr xsj Þ2 þðyrysj Þ2
ðxr xsj Þðyr ysj Þ
ðyr ysj Þ2
ðxr xsj Þ2 þðyrysj Þ2
a2j
aj bj
a j bj
b2j
#
3 7 7 5
ðxr xsj Þ2 þðyrysj Þ2
where, aj and bj are variables which define the position r r rr s unit vectors (i.e., ½aj ; bj T ¼ krrr rrs jk ). j 2
In addition, the weighted HDOP matrix can be related to the information content in a closed form, defined by K X 1 Dw ¼ L 2 s j¼1 sj
"
b2j
aj bj
aj bj
a2j
# (9)
Khalife, Joshua Morales, Kimia Shamaei, Mahdi Maaref, Kyle Semelka, MyLinh Nguyen, and Trier Mortlock for their help with preparing for data collection. DISTRIBUTION STATEMENT A. Approved for public release; Distribution is unlimited. 412TW-PA-20146. This work was supported in part by the National Science Foundation (NSF) under Grant 1929571, and in part by the Air Force Office of Scientific Research (AFOSR) under Grant FA9550-22-1-0476. This work was also supported in part by the Laboratory Directed Research and Development program at Sandia National Laboratories, a multimission laboratory managed and operated by National Technology and Engineering Solutions of Sandia LLC, a wholly owned subsidiary of Honeywell International Inc. for the U.S. Department of Energy’s National Nuclear Security Administration, under contract DENA0003525. This paper describes objective technical results and analysis. Any subjective views or opinions that might be expressed in the paper do not necessarily represent the views of the U.S. Department of Energy or the United States Government. SAND2022-13901 C.
REFERENCES [1] R. Sabatini et al., “Avionics systems panel research and innovation perspectives,” IEEE Aerosp. Electron. Syst. Mag., vol. 35, no. 12, pp. 58–72, Dec. 2020, doi: 10.1109/
where 2( L¼4
K a2 X j j¼1
)(
s 2sj
K b2 X j s2 j¼1 sj
)
( )2 31 K X aj bj 5 s 2sj j¼1
MAES.2020.3033475. [2] EUROCONTROL, Aviation Intelligence Unit, “Does radio frequency interference to satellite navigation pose an increasing threat to network efficiency, cost-effectiveness
is a constant value, corresponding to the information content of the position states, which is found in all HDOP terms. Finally, the HDOP constant can be defined as vffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi ) u ( K X 1 u 2 2 t HDOP ¼ L bj þ aj : (10) s2 j¼1 sj
and ultimately safety?,” Tech. Rep., Mar. 2021. [Online]. Available:
https://www.eurocontrol.int/sites/default/files/
2021-03/eurocontrol-think-paper-9-radio-frequencyintereference-satellite-navigation.pdf [3] International Civil Aviation Organization (ICAO), “An urgent need to address harmful interferences to GNSS,” Tech. Rep., May 2019. [Online]. Available: https://www.iata. org/contentassets/e45e5219cc8c4277a0e80562590793da/ address-harmful-interferences-gnss.pdf
ACKNOWLEDGMENTS
[4] J. Raquet et al., “Part D: Position, navigation, and timing
The authors would like to thank Ali Abdallah for his help with data processing as well as Tucker Haydon, Nadim Khairallah, and Mohammad Neinavaie for helpful discussions. The authors would like to thank Edwards Air Force Base (AFB) and Holloman AFB for inviting the ASPIN Laboratory to conduct experiments on U.S. Air Force aircraft in the “SNIFFER: Signals of opportunity for Navigation In Frequency-Forbidden EnviRonments” flight campaign. The authors would like to thank the U.S. Air Force pilots Chiawei Lee, Juan Jurado, Steven Wachtel, Jacob Duede, Zachary Hoeffner, Thomas Hulsey, and Rachel Quirarte and Republic of Singapore Air Force pilot RunXuan Tay. The authors would also like to thank Joe OCTOBER 2023
using radio signals-of-opportunity,” Position, Navigation, and Timing Technologies in the 21st Century, J. Morton, F. van Diggelen, J. Spilker Jr., and B. Parkinson, Eds. New York, NY, USA: Wiley, 2021, vol. 2, ch. 35–43, pp. 1115–1412, doi: 10.1002/9781119458555.ch35. [5] J. Mortier, G. Pages, and J. Vila-Valls, “Robust TOAbased UAS navigation under model mismatch in GNSSdenied harsh environments,” Remote Sens., vol. 12, no. 18, pp. 2928–2947, Sep. 2020, doi: 10.3390/rs12182928. [6] N. Souli, P. Kolios, and G. Ellinas, “Online relative posi-
IEEE A&E SYSTEMS MAGAZINE
tioning of autonomous vehicles using signals of opportunity,” IEEE Trans. Intell. Veh., vol. 7, no. 4, pp. 873– 885, Dec. 2022, doi: 10.1109/TIV.2021.3124727.
37
Efficient Transmitter Selection Strategies for Improved Information Gathering
[7] Z. Kassas, J. Khalife, A. Abdallah, and C. Lee, “I am not
[18] C. Huang, H. Qin, C. Zhao, and H. Liang, “Phase - time
afraid of the GPS jammer: Resilient navigation via signals
method: Accurate doppler measurement for iridium NEXT
of opportunity in GPS-denied environments,” IEEE
signals,” IEEE Trans. Aerosp. Electron. Syst., vol. 58,
Aerosp. Electron. Syst. Mag., vol. 37, no. 7, pp. 4–19, Jul. 2022, doi: 10.1109/MAES.2022.3154110.
no. 6, pp. 5954–5962, Dec. 2022, doi: 10.1109/ TAES.2022.3180702.
[8] M. Psiaki and B. Slosman, “Tracking of digital FM OFDM
[19] J. Khalife and Z. Kassas, “Opportunistic UAV navigation
signals for the determination of navigation observables,”
with carrier phase measurements from asynchronous cellu-
in Proc. ION GNSS Conf., 2019, pp. 2325–2348,
lar signals,” IEEE Trans. Aerosp. Electron. Syst., vol. 56,
doi: 10.33012/2019.17120.
no. 4, pp. 3285–3301, Aug. 2020, doi: 10.1109/
[9] X. Chen, Q. Wei, F. Wang, Z. Jun, S. Wu, and A. Men,
TAES.2019.2948452.
“Super-resolution time of arrival estimation for a symbi-
[20] J. Khalife and Z. Kassas, “On the achievability of sub-
otic FM radio data system,” IEEE Trans. Broadcast., vol. 66, no. 4, pp. 847–856, Dec. 2020, doi: 10.1109/
meter-accurate UAV navigation with cellular signals exploiting loose network synchronization,” IEEE Trans. Aerosp. Electron. Syst., vol. 58, no. 5, pp. 4261–4278, Oct.
TBC.2019.2957666.
2022, doi: 10.1109/TAES.2022.3162770.
[10] K. Shamaei and Z. Kassas, “Receiver design and time of arrival estimation for opportunistic localization with 5 G
[21] J. Khalife and Z. Kassas, “Differential framework for sub-
signals,” IEEE Trans. Wireless Commun., vol. 20, no. 7,
meter-accurate vehicular navigation with cellular signals,”
pp. 4716–4731, Jul. 2021, doi: 10.1109/TWC.2021.
IEEE Trans. Intell. Veh., vol. 8, no. 1, pp. 732–744, Jan. 2023, doi: 10.1109/TAES.2022.3162770.
3061985. [11] P. Wang, Y. Wang, and J. Morton, “Signal tracking algorithm with adaptive multipath mitigation and experimental results
[22] N. Schneckenburger et al., “Measurement of the l-band air-to-ground channel for positioning applications,” IEEE
for LTE positioning receivers in urban environments,” IEEE
Trans. Aerosp. Electron. Syst., vol. 52, no. 5, pp. 2281–
Trans. Aerosp. Electron. Syst., vol. 58, no. 4, pp. 2779–2795,
2297, Oct. 2016, doi: 10.1109/TAES.2016.150451. [23] W. Khawaja, I. Guvenc, D. Matolak, U. Fiebig, and N.
Aug. 2022, doi: 10.1109/TAES.2021.3139569. [12] I. Lapin, G. Granados, J. Samson, O. Renaudin, F.
Schneckenburger, “A survey of air-to-ground propagation
Zanier, and L. Ries, “STARE: Real-time software
channel modeling for unmanned aerial vehicles,” IEEE
receiver for LTE and 5 G NR positioning and signal mon-
Commun. Surv. Tut., vol. 21, no. 3, pp. 2361–2391, Jul.–
itoring,” in Proc. Workshop Satell. Navigation Technol., 2022, pp. 1–11, doi: 10.1109/NAVITEC53682.
Sep. 2019, doi: 10.1109/COMST.2019.2915069. [24] E. Kim and Y. Shin, “Feasibility analysis of LTE-based UAS navigation in deep urban areas and DSRC
2022.9847544.
augmentation,” Sensors, vol. 19, no. 9, pp. 4192–4207,
[13] C. Yang, M. Arizabaleta-Diez, P. Weitkemper, and T.
Apr. 2019, doi: 10.3390/s19194192.
Pany, “An experimental analysis of cyclic and reference signals of 4G LTE for TOA estimation and positioning in
[25] B. Stevens and M. Younis, “Detection algorithm for cellu-
mobile fading environments,” IEEE Aerosp. Electron.
lar synchronization signals in airborne applications,” IEEE
Syst. Mag., vol. 37, no. 9, pp. 16–41, Sep. 2022,
Access, vol. 9, pp. 55555–55566, Apr. 2021, doi: 10.1109/
doi: 10.1109/MAES.2022.3186650. [14] C. Yang and A. Soloviev, “Mobile positioning with signals of
ACCESS.2021.3071674. [26] Z. Kassas et al., “Received power characterization of ter-
opportunity in urban and urban canyon environments,” in
restrial cellular signals on high altitude aircraft,” in Proc.
Proc. IEEE/ION Position, Location, Navigation Symp., 2020,
IEEE Aerosp. Conf., 2022, pp. 1–8, doi: 10.1109/ AERO53065.2022.9843492.
pp. 1043–1059, doi: 10.1109/PLANS46316.2020.9109876. [15] T. Hong, J. Sun, T. Jin, Y. Yi, and J. Qu, “Hybrid position-
[27] Z. Kassas et al., “Assessment of cellular signals of oppor-
ing with DTMB and LTE signals,” in Proc. Int. Wireless
tunity for high-altitude aircraft navigation,” IEEE Aerosp.
Commun.
303–307,
Electron. Syst. Mag., vol. 37, no. 10, pp. 4–19, Oct. 2022,
doi: 10.1109/IWCMC51323.2021.9498758. [16] Z. Kassas, N. Khairallah, and S. Kozhaya, “Ad Astra: Simul-
doi: 10.1109/MAES.2022.3187142. [28] Z. Kassas et al., “Protecting the skies: GNSS-less accurate
taneous tracking and navigation with megaconstellation LEO
aircraft navigation with terrestrial cellular signals of oppor-
satellites,” IEEE Aerosp. Electron. Syst. Mag., 2023.
tunity,” in Proc. ION GNSS Conf., 2022, pp. 1014–1025,
Mobile
Comput.,
2021,
pp.
doi: 10.33012/2022.18579.
[17] M. Hartnett, “Performance assessment of navigation
38
using carrier Doppler measurements from multiple
[29] L. Kaplan, “Local node selection for localization in a dis-
LEO constellations,” Master’s thesis, Air Force Insti-
tributed sensor network,” IEEE Trans. Aerosp. Electron.
tute of Technology, Wright-Patterson AFB, OH, USA,
Syst., vol. 42, no. 1, pp. 136–146, Jan. 2006, doi: 10.1109/
2022.
TAES.2006.1603410.
IEEE A&E SYSTEMS MAGAZINE
OCTOBER 2023
Nguyen and Kassas
[30] N. Cao, S. Choi, E. Masazade, and P. Varshney, “Sensor
[39] F. Wang, X. Bai, B. Guo, and C. Liu, “Dynamic clustering
selection for target tracking in wireless sensor networks
in wireless sensor network for target tracking based on the
with uncertainty,” IEEE Trans. Signal Process., vol. 64,
fisher information of modified kalman filter,” in Proc.
no. 20, pp. 5191–5204, Oct. 2016, doi: 10.1109/ TSP.2016.2595500.
IEEE Int. Conf. Syst. Informat., 2016, pp. 696–700, doi: 10.1109/ICSAI.2016.7811042.
[31] S. Joshi and S. Boyd, “Sensor selection via convex opti-
[40] A. Nguyen and Z. Kassas, “Transmitter selection for improved
mization,” IEEE Trans. Signal Process., vol. 57, no. 2,
information gathering in aerial vehicle navigation with terrestrial signals of opportunity,” in Proc. ION Int. Tech. Meeting,
pp. 451–462, Feb. 2009, doi: 10.1109/TSP.2008.2007095. [32] V. Kekatos and G. Giannakis, “Selecting reliable sensors
2022, pp. 723–734, doi: 10.33012/2022.18262.
via convex optimization,” in Proc. IEEE Int. Workshop
[41] G. Mutambara, Decentralized Estimation and Control for
Signal Process. Adv. Wireless Commun., 2010, pp. 1–5,
Multisensor Systems. Boca Raton, FL, USA: CRC Press,
doi: 10.1109/SPAWC.2010.5670891. [33] S. Liu, S. Chepuri, M. Fardad, E. Masazade, G. Leus, and
1998. [42] J. Morales, J. Khalife, and Z. Kassas, “Opportunity for
P. Varshney, “Sensor selection for estimation with corre-
accuracy,” GPS World Mag., vol. 27, no. 3, pp. 22–29,
lated measurement noise,” IEEE Trans. Signal Process., vol. 64, no. 13, pp. 3509–3522, Jul. 2016, doi: 10.1109/
Mar. 2016. [43] M. Maaref, J. Khalife, and Z. Kassas, “Aerial vehicle protec-
TSP.2016.2550005.
tion level reduction by fusing GNSS and terrestrial signals of
selection for TDOA-based source localization in wireless
opportunity,” IEEE Trans. Intell. Transp. Syst., vol. 22, no. 9, pp. 5976–5993, Sep. 2021, doi: 10.1109/TITS.2021.3095184.
sensor networks,” IEEE Trans. Veh. Technol., vol. 69, no. 10, pp. 12031–12042, Oct. 2020, doi: 10.1109/
[44] D. Ucinski, Optimal Measurement Methods for Distributed Parameter System Identification. Boca Raton, FL, USA:
[34] Z. Dai, G. W. X. Jin, and X. Lou, “Nearly optimal sensor
CRC Press, 2005.
TVT.2020.3011118. [35] V. Cerone, S. Fosson, and D. Regruto, “A non-convex
[45] N. Blanco-Delgado, F. Nunes, and G. Seco-Granados,
adaptive regularization approach to binary optimization,”
“Relation between GDOP and the geometry of the satellite
in Proc. IEEE Conf. Decis. Control, 2021, pp. 3844–3849,
constellation,” in Int. Conf. Localization GNSS, 2011, pp. 175–180, doi: 10.1109/ICL-GNSS.2011.5955282.
doi: 10.1109/CDC45484.2021.9683667. [36] A. Hashemi, M. Ghasemi, H. Vikalo, and U. Topcu,
[46] S. Aditya, H. Dhillon, A. Molisch, R. Buehrer, and H.
“Randomized greedy sensor selection: Leveraging weak submodularity,” IEEE Trans. Autom. Control,
Behairy, “Characterizing the impact of SNR heterogeneity on time-of-arrival-based localization outage probability,”
vol. 66, no. 1, pp. 199–212, Jan. 2021, doi: 10.1109/
IEEE Trans. Wireless Commun., vol. 18, no. 1, pp. 637– 649, Jan. 2019, doi: 10.1109/TWC.2018.2883726.
TAC.2020.2980924. [37] Y. Saito et al., “Determinant-based fast greedy sensor
[47] X. Li and V. Jilkov, “Survey of maneuvering target track-
selection algorithm,” IEEE Access, vol. 9, pp. 68535–
ing. Part I: Dynamic models,” IEEE Trans. Aerosp. Elec-
68551, 2021, doi: 10.1109/ACCESS.2021.3076186.
tron. Syst., vol. 39, no. 4, pp. 1333–1364, Oct. 2003, doi: 10.1109/TAES.2003.1261132.
[38] X. Shen and P. Varshney, “Sensor selection based on generalized information gain for target tracking in large sensor networks,” IEEE Trans. Signal Process.,
[48] D. Won et al., “Weighted DOP with consideration on elevation-dependent range errors of GNSS satellites,” IEEE
vol. 62, no. 2, pp. 363–375, Jan. 2014, doi: 10.1109/
Trans. Instrum. Meas., vol. 61, no. 12, pp. 3241–3250,
TSP.2013.2289881.
Dec. 2012, doi: 10.1109/TIM.2012.2205512.
OCTOBER 2023
IEEE A&E SYSTEMS MAGAZINE
39
IEEE AESS
SCHOLARSHIP IEEE AEROSPACE AND ELECTRONIC SYSTEMS
PROGRAM $10,000 + CERTIFICATE
DEADLINE: DECEMBER 1
AESS awards one Graduate and one Undergraduate-level scholarship annually. Visit the website for requirements and eligibility. UNDERGRADUATE LEVEL Electrical Engineering
GRADUATE LEVEL Systems Engineering
CONTACT US: [email protected] ieee-aess.org/scholarship
2023 Aerospace & Electronic Systems Society Organization and Representatives OFFICERS
VP Member Services – Lorenzo Lo Monte VP Publications – Lance Kaplan VP Technical Operations – Michael Braasch
President – Mark Davis President-Elect – Sabrina Greco Past President – Walt Downing Secretary – Kathleen Kramer Treasurer – Mike Noble VP Conferences – Braham Himed VP Education – Alexander Charlish VP Finance – Peter Willett VP Industry Relations – Steve Butler
OTHER POSITIONS
Undergraduate Student Rep – Abir Tabarki Graduate Student Rep – Jemma Malli Young Professionals Program Coordinator – Philipp Markiton Operations Manager – Amanda Osborn
BOARD OF GOVERNORS 2023 Members-at-Large
2021-2023 Laura Anitori Steve Butler Michael Cardinale Alexander Charlish Stefano Coraluppi Braham Himed Lorenzo Lo Monte Peter Willett
2022-2024 Alfonso Farina Maria Sabrina Greco Hugh Griffiths Puneet Kumar Mishra Laila Moreira Bob Rassa Michael Noble Roberto Sabatini
STANDING COMMITTEES & CHAIRS Awards – Fulvio Gini M. Barry Carlton Award – Gokhan Inalhan Harry Rowe Mimno Award – Daniel O’Hagan Warren D. White Award – Scott Goldstein Pioneer Award – Daniel Tazartes Fred Nathanson Award – Braham Himed Robert T. Hill Best Dissertation Award – Alexander Charlish AESS Early Career Award – George T. Schmidt AESS Judith A. Resnik Space Award – Maruthi Akella Chapter Awards – Kathleen Kramer Distinguished Service Award – Peter Willett Industrial Innovation Award – Mike Noble Engineering Scholarship – Bob Rassa Chapter Program Coordinator – Kathleen Kramer Constitution, Organization & Bylaws – Hugh Griffiths Education – Alexander Charlish Distinguished Lecturer Program – Alexander Charlish Fellow Evaluation – Hugh Griffiths Fellow Search – George T. Schmidt History – Alfonso Farina International Director Liaison – Joe Fabrizio Member Services – Lorenzo Lo Monte Nominations & Appointments – Walt Downing Publications – Lance Kaplan Systems Magazine – Daniel O’Hagan Transactions – Gokhan Inalhan Tutorials – W. Dale Blair QEB – Francesca Filippinni; Philipp Markiton Strategic Planning – Sabrina Greco Student Activities –Kathleen Kramer Technical Operations – Michael Braasch Avionics Systems – Roberto Sabatini Cyber Security – Aloke Roy Glue Technologies for Space Systems – Claudio Sacchi Gyro & Accelerometer Panel – Jason Bingham Navigation Systems Panel – Michael Braasch Radar Systems Panel – Laura Anitori Visions and Perspectives (ad hoc) – Joe Dauncey
2023-2025 William Dale Blair Arik Brown Joe Fabrizio Francesca Filippini Wolfgang Koch Luke Rosenberg Marina Ruggieri George Schmidt
CONFERENCE LIAISONS IEEE Aerospace Conference – Claudio Sacchi IEEE AUTOTESTCON – Bob Rassa, Dan Walsh, Walt Downing IEEE International Carnahan Conference on Security Technology – Gordon Thomas IEEE/AIAA Digital Avionics Systems Conference – Kathleen Kramer IEEE Radar Conference – Kristin Bing IEEE/ION Position, Location & Navigation Symposium – Michael Braasch IEEE/AIAA/NASA Integrated Communications Navigation & Surveillance – Aloke Roy IEEE International Workshop for Metrology for Aerospace – Pasquale Daponte FUSION – W. Dale Blair REPRESENTATIVES TO IEEE ENTITIES Journal of Lightwave Technology – Michael Cardinale Nanotechnology Council – Yvonne Gray Sensors Council – Paola Escobari Vargas, Peter Willett Systems Council – Bob Rassa, Michael Cardinale IEEE Women in Engineering Committee – Kathleen Kramer
Please send corrections or omissions for this page to the Operations Manager at [email protected].
Visit our website at ieee-aess.org.
History Column:
DOI. No. 10.1109/MAES.2023.3295349
William Sealy Gosset–“Student” Hugh Griffiths , University College London, London WC1E 6BT, U.K.
he derived, are known as the Student-t distribution and William Sealey Gosset was a statistician, born in 1876 the Student-t test. in Canterbury, U.K. He studied natural sciences and Joan Fisher Box, the daughter of Ronald Fisher and mathematics at New College, Oxford University, then, the second wife (of three) of George Box (another in 1899, joined the brewery of Arthur Guinness famous statistician), described Gossett as “an extraordiin Dublin, Ireland, where he held the position of narily appealing individual, head brewer. He spent the generous to a fault, humble, rest of his 38-year career at enthusiastic in the pursuit Guinness. of his varied interests, and In his role at the brewhelpful.” ery, he was responsible for Gosset met and correassessing the quality of the sponded with two of the foreraw ingredients of the beer, and of the beer itself, and most academic statisticians he turned his mathematical of the early twentieth centalent to the problem of tury: Ronald Fisher of how to do this with a small Cambridge University, Karl number of samples. He Pearson of University Coldeveloped the fundamental lege London, and Gossett theory behind this problem, spent a sabbatical year workand sought to publish the ing in Pearson’s department results, in a paper titled at University College Lon“The probable error of a don. Both Fisher and Pearson mean.” This may easily be worked on the application of downloaded, and it will be advanced statistical techniseen that it is beautifully ques to agriculture, and William Sealy Gosset (1876–1937). Public domain. and clearly written. Fisher took up a position However, the brewery at the Rothamsted Research did not want to reveal their Institute, which exists to this day and undertakes important identity (and, hence, the fact that they were using research in this field. He is now known for the concepts of such methods), nor that of the author, so they insisted maximum likelihood estimation and of Fisher informathat he publish under a pseudonym. He chose tion, which forms the basis of the Cramer–Rao lower the name “Student,” and the statistical distribution bound in statistical estimation. Apparently, Fisher and derived in the paper (essentially a generalization of the normal distribution) and the associated test that Pearson were not especially friendly with each other, but Gosset corresponded freely with both. Fisher and Pearson both now attract some criticism for their work and opinions on eugenics. Author’s current address: Hugh Griffiths, University Gosset’s work is significant because it shows the College London, London WC1E 6BT, U.K. (e-mail: origins of some of the statistical methods in (for [email protected]). ple) radar clutter distributions and radar detection theManuscript received 29 June 2023; accepted 3 July 2023, ory that we now take for granted. It is interesting that, and ready for publication 27 July 2023. even now, Gosset’s true identity as the source of some Review handled by Daniel O’Hagan. of these ideas is not fully appreciated. 0885-8985/23/$26.00 ß 2023 IEEE 42
IEEE A&E SYSTEMS MAGAZINE
OCTOBER 2023
Further information may be found in: Student, “The probable error of a mean,” Biometrika, vol. 6, no. 1, pp. 1–25, Mar. 1908. [Online]. Available: https://www.york.ac.uk/depts/maths/histstat/student.pdf S. Zabell, “‘On Student’s 1908 article ‘The probable error of a mean,’” J. Amer. Statis. Assoc., Mar. 2008. [Online]. Available: https://www.jstor.org/stable/27640017 J. Fisher Box, “Gosset, Fisher, and the t distribution,” The American Statistician, vol. 35, no. 2, pp. 61–66, May 1981. [Online]. Available: https://doi.org/ 10.2307/2683142 J. Fisher Box, R. A. Fisher, The Life of a Scientist, New York, NY, USA: Wiley, 1978.
Ronald Fisher (1890–1962) in 1913. Public domain.
OCTOBER 2023
IEEE A&E SYSTEMS MAGAZINE
43
The Virtual Dis.nguished Lecturer Program (VDLP) allows us to serve the AESS par.cipants and the aerospace and electronic systems community the opportunity to hear from our respected Dis.nguished Lecturers from around the world – both live and on demand. Registra.on is free for all webinars. If you are unable to aXend the "live" virtual events, the presenta.ons will be available aZer the event.
Scan to Register ieee-aess.org/vdl
+100
DL talk Btles to choose from in a variety of technical fields
38
available DLs providing lectures online and in person at your next meeBng, chapter event, and more
UPCOMING WEBINARS TUES. 10 OCTOBER 11 AM ET/3 PM UTC
Measurement Extrac.on for a Point Target from an Op.cal Sensor
Yaakov Bar-Shalom
THURS. 12 OCTOBER 11 AM ET/3 PM UTC
Adap.ve Radar Detec.on
Antonio De Maio
THURS. 26 OCTOBER 11 AM ET/3 PM UTC
Mul.ple-Hypothesis Tracking
Stefano P. Coraluppi
WED. 15 NOVEMBER 11 AM ET/3 PM UTC
Ontological Decision-Making Support for Air Traffic Management Towards Trustworthy Autonomy: How AI Can Help Address Fundamental Learning and Adapta.on Challenges Sensor Loca.on Op.miza.on for Effec.ve and Robust Beamforming Filter Design for Radar Tracking of Maneuvering Targets Iner.al Naviga.on Computa.on: History, Present and Beyond
WED. 29 NOV. 11 AM ET/4 PM UTC THURS. 7 DEC. 11 AM ET/4 PM UTC THURS. 14 DEC. 11 AM ET/4 PM UTC THURS. 21 DEC. 11 AM ET/4 PM UTC
Carlos C. Insaurralde
Full schedule is available on ieee-aess.org/vdl
Gokhan Inalhan Wei Liu Dale Blair Yuanxin Wu
2023-2024 Aerospace & Electronic Systems Society Meetings and Conferences The information listed on this page was valid as of 1 September 2023. Please check the respective conference websites for the most up-to-date information. DATE
MEETING
LOCATION
CONFERENCE WEBSITE
1-5 Oct. 2023
IEEE/AIAA 42nd Digital Avionics Systems Conference (DASC)
Barcelona, Spain
2023.dasconline.org
2-6 Oct. 2023
13th European Space Power Conference (ESPC)
Elche, Spain
atpi.eventsair.com/espc2023
2-6 Oct. 2023
European Data Handling & Data Processing Conference (EDHPC)
Juan-LesPins, France
atpi.eventsair.com/edhpc-conference
11-15 Oct. 2023
IEEE International Carnahan Conference on Security Technology (ICCST)
Pune, India
site.ieee.org/iccst
6-8 Nov. 2023
IEEE International Conference on Microwaves, Communications, Antennas, Biomedical Engineering and Electronic Systems (COMCAS)
Tel Aviv, Israel
www.comcas.org
6-10 Nov. 2023
IEEE International Radar Conference (RADAR)
Sydney, Australia
www.radar2023.org
27-29 Nov. 2023
IEEE Symposium Sensor Data Fusion and International Conference on Multisensor Fusion and Integration (SDF-MFI)
Bonn, Germany
www.fkie.fraunhofer.de/de/Veranstalt ungen/sdf2023.html
21-24 Jan. 2024
IEEE Radio and Wireless Symposium (RWS)
San Antonio, TX, USA
www.radiowirelessweek.org
2-9 March 2024
IEEE Aerospace Conference (AERO)
Big Sky, MT, USA
www.aeroconf.org
23-25 April 2024
IEEE Integrated Communications, Navigation and Surveillance Conference (ICNS)
Herndon, VA, USA
i-cns.org
24-26 April 2024
International Conference on Global Aeronautical Engineering and Satellite Technology (GAST)
Marrakesh, Morocco
gast24.sciencesconf.org/resource/ac ces
6-10 May 2024
IEEE Radar Conference (RadarCon’f24)
Denver, CO, USA
2024.ieee-radarconf.org
27-28 May 2024
Security for Space Systems (3S)
Noordwijk, Netherlands
ieeeaess.org/event/conference/2024security-space-systems-3s
3-5 June 2024
2024 11th International Workshop on Metrology for AeroSpace
Lublin, Poland
www.metroaerospace.org
5-7 June 2024
2024 International Conference on Localization and GNSS
Antwerp, Belgium
events.tuni.fi/icl-gnss2024
For a full list of AESS-sponsored conferences, visit ieee-aess.org/conferences. For corrections or omissions, contact [email protected]. VP Conferences, Braham Himed