International Conference on Artificial Intelligence: Advances and Applications 2019: Proceedings of ICAIAA 2019 (Algorithms for Intelligent Systems) 981151058X, 9789811510588

Citation preview

Algorithms for Intelligent Systems Series Editors: Jagdish Chand Bansal · Kusum Deep · Atulya K. Nagar

Garima Mathur · Harish Sharma · Mahesh Bundele · Nilanjan Dey · Marcin Paprzycki Editors

International Conference on Artificial Intelligence: Advances and Applications 2019 Proceedings of ICAIAA 2019

Algorithms for Intelligent Systems Series Editors Jagdish Chand Bansal, Department of Mathematics, South Asian University, New Delhi, Delhi, India Kusum Deep, Department of Mathematics, Indian Institute of Technology Roorkee, Roorkee, Uttarakhand, India Atulya K. Nagar, Department of Mathematics and Computer Science, Liverpool Hope University, Liverpool, UK

This book series publishes research on the analysis and development of algorithms for intelligent systems with their applications to various real world problems. It covers research related to autonomous agents, multi-agent systems, behavioral modeling, reinforcement learning, game theory, mechanism design, machine learning, meta-heuristic search, optimization, planning and scheduling, artificial neural networks, evolutionary computation, swarm intelligence and other algorithms for intelligent systems. The book series includes recent advancements, modification and applications of the artificial neural networks, evolutionary computation, swarm intelligence, artificial immune systems, fuzzy system, autonomous and multi agent systems, machine learning and other intelligent systems related areas. The material will be beneficial for the graduate students, post-graduate students as well as the researchers who want a broader view of advances in algorithms for intelligent systems. The contents will also be useful to the researchers from other fields who have no knowledge of the power of intelligent systems, e.g. the researchers in the field of bioinformatics, biochemists, mechanical and chemical engineers, economists, musicians and medical practitioners. The series publishes monographs, edited volumes, advanced textbooks and selected proceedings.

More information about this series at http://www.springer.com/series/16171

Garima Mathur Harish Sharma Mahesh Bundele Nilanjan Dey Marcin Paprzycki •

•

• •

Editors

International Conference on Artificial Intelligence: Advances and Applications 2019 Proceedings of ICAIAA 2019

123

Editors Garima Mathur Department of Electronics and Communication Engineering Poornima College of Engineering Jaipur, Rajasthan, India Mahesh Bundele Poornima College of Engineering Jaipur, Rajasthan, India

Harish Sharma Department of Computer Science and Engineering Rajasthan Technical University Kota, Rajasthan, India Nilanjan Dey Department of Information Technology Techno India College of Technology Durgapur, West Bengal, India

Marcin Paprzycki Systems Research Institute Polish Academy of Science Warsaw, Poland

ISSN 2524-7565 ISSN 2524-7573 (electronic) Algorithms for Intelligent Systems ISBN 978-981-15-1058-8 ISBN 978-981-15-1059-5 (eBook) https://doi.org/10.1007/978-981-15-1059-5 © Springer Nature Singapore Pte Ltd. 2020, corrected publication 2021 This work is subject to copyright. All rights are reserved by the Publisher, whether the whole or part of the material is concerned, specifically the rights of translation, reprinting, reuse of illustrations, recitation, broadcasting, reproduction on microfilms or in any other physical way, and transmission or information storage and retrieval, electronic adaptation, computer software, or by similar or dissimilar methodology now known or hereafter developed. The use of general descriptive names, registered names, trademarks, service marks, etc. in this publication does not imply, even in the absence of a specific statement, that such names are exempt from the relevant protective laws and regulations and therefore free for general use. The publisher, the authors and the editors are safe to assume that the advice and information in this book are believed to be true and accurate at the date of publication. Neither the publisher nor the authors or the editors give a warranty, expressed or implied, with respect to the material contained herein or for any errors or omissions that may have been made. The publisher remains neutral with regard to jurisdictional claims in published maps and institutional affiliations. This Springer imprint is published by the registered company Springer Nature Singapore Pte Ltd. The registered company address is: 152 Beach Road, #21-01/04 Gateway East, Singapore 189721, Singapore

Preface

The Conference on Artificial Intelligence: Advances and Applications-2019 is the first International Conference that has been organized with publication support of Springer. It is jointly organized by Poornima College of Engineering, Jaipur, Poornima Institute of Engineering and Technology, Jaipur, Rajasthan Technical University, Kota, Rajasthan, India, and Jawaharlal Nehru Technological University Hyderabad under RTU (ATU) TEQIP III. The objective of the conference was to bring together researchers from academia, government agencies, research laboratories and the corporate sector from all over the world to present their research works carried out in artificial intelligence domain. Artificial intelligence research is leading to evolve in the area of machine learning, deep learning applications in health care, agriculture, business and security, etc. It covers research in core concepts of computer networks, intelligent system design and deployment, real-time systems, WSN, sensors and sensor nodes, SDN, NFV, etc. It is useful for students, faculty and industry working on AI applications. AI is an emerging technology that shall be used in almost all applications/technologies/systems in one or other form. On behalf of RTU Kota, Poornima College of Engineering, Poornima Institute of Engineering and Technology, Jaipur, and JNTU, Hyderabad, I am pleased to welcome all the readers of Proceedings of International Conference on Artificial Intelligence: Advances and Applications-ICAIAA 2019. This conference has provided an environment to conduct intellectual discussions and exchange ideas that are instrumental in shaping the future of artificial intelligence. It has been planned in two tracks, viz. Track 1: Artificial Intelligence and Track 2: Intelligent Sensor, Devices & Application. The conference could get very good response from four countries such as India, USA, UAE and Nigeria. A total of 188 papers were received and were reviewed by 125 reviewers from different reputed institutions across India. A total of 59 papers were accepted for presentation in the conference from two countries, out of which 42 papers were registered and presented during June 28–29, 2019. I am thankful to RTU Kota for giving this opportunity to organize this conference under TEQIP III project.

v

vi

Preface

ICAIAA 2019 included following keynotes and plenary talks:

Dr. Jagdish Chand Bansal Assistant Professor (Senior Grade), Department of Mathematics, 327, Akbar Bhawan, Chanakyapuri, New Delhi, 110021, e-mail: [email protected] Talk Title: Artificial Intelligence & Environment

Dr. Swagatam Das Associate Professor, Electronics and Communication Sciences Unit of the Indian Statistical Institute, Kolkata, India Talk Title: Predictive and Generative Deep Inference Models: Perspectives and Recent Trends

Preface

vii

Dr. Nilanjan Dey Professor, Department of Information Technology, Techno India College of Technology, Kolkata, India Talk Title: The Silent Thief of Sight—CAD for Glaucoma

Dr. Kottakkaran Sooppy Nisar Associate Professor, Department of Mathematics, College of Arts & Sciences, Prince Sattam bin Abdul-Aziz University, Kingdom of Saudi Arabia Talk Title: Some Recent Trends in Mathematics: Fractional Calculus, Optical Communications, and Cryptosystems

viii

Preface

Dr. B. Vishnu Vardhan Professor, Department of Computer Science and Engineering, JNTUH College of Engineering Hyderabad Talk Title: Research Avenues in Data Mining

Dr. Praveen Agarwal Vice Principal, Anand International College of Engineering, Near Kanota, Agra Road, Jaipur, Rajasthan, India Talk Title: Mathematical Modelling of Transmission Dynamics of Nipah Virus via Fractional Order Approach

Preface

ix

Dr. Nishcal Verma Professor, Department of Electrical Engineering and Inter-Disciplinary Program in Cognitive Science, IIT Kanpur Talk Title: Deep Learning & Fuzzy System

Dr. Anupam Yadav Assistant Professor Grade-I, Department of Mathematics, National Institute of Technology Jalandhar Talk Title: Artificial Electric Field Algorithm (AEFA) and Its Applications

x

Preface

Dr. Deepak Garg Director Leadingindia.ai, Professor and Chair, Computer Science Engineering, Bennett University, Director, NVIDIA-Bennett Research Centre for Artificial Intelligence Talk Title: Cognitive Intelligence: Opportunities and Challenges

Steering Committee Aleksandra Mileva, Goce Delcev University STIP, Macedonia Carlos M. Travieso, Gonzalez University of Las Palmas de Gran Canaria, Spain Raghunath K. Shevgaonkar, Bennett University, India Marcin Paprzycki, Systems Research Institute, Polish Academy of Sciences, Warsaw, Poland Sedat Akleylek, Ondokuz Mayis University Samsun, Turkey Sureswaran Ramadass, USM University Penang, Malaysia Youcef Soufi, University of Tabessa, Algeria Lalit Kumar Goel, NTU Nanyang, Singapore Pinnamaneni Bhanu, Kelenn Technology Antony, France Armin Aberle, SERIS National University of Singapore, Singapore Sandeep Sancheti, SRM University, Chennai, India Daniele Riboni, University of Cagliari, Italy Wan Young Chung, Pukyong National University Busan, South Korea Maria Ganzha, System Research Institute Polish Academy of Sciences and IBS PAN, Warsaw, Poland

Preface

Advisory Committee Vijay K. Bhargav, University of British Columbia Ramjee Prasad, Aalborg University, Denmark Mangalore Anantha Pai, University of Illinois Urbana–Champaign Vinod Kumar, ALCATEL Boulogne-Billancourt, France Sastri Kota, University of Oulu, Finland Udai Desai, IIIT Raipur, India Sanjay K. Bose, IIT Guhawati, India Ajit K. Chaturvedi, IIT Kanpur, India Abhay Karandikar, IIT Bombay, India Timothy A. Gonsalves, IIT Mandi, India Baylon G. Fernandes, IIT Bombay, India Ali Al-Sherbaz, University of Northampton, UK Atheer Matroud, University of Otago, New Zealand George Tsaramirsis, King Abdulaziz University Jeddah, Saudi Arabia Nidaa A. Abbas, University of Babylon, Iraq Ramesh C. Bansal, Queensland University, Brisbane, Australia RaedAbd-Alhameed, University of Bradford, UK Sattar B. Sadkhan, University of Babylon, Iraq Sedat Akleylek, Ondokuz Mayis University Samsun, Turkey Sureswaran Ramadass, USM University Penang, Malaysia William Puech, University Montpellier, France Youcef Soufi Mail, University of Tabessa, Algeria Wan Young Chung, Pukyong National University Busan, South Korea Youcef Soufi, University of Tbessa, Algeria Li Zhiwu, Macau University of Science and Technology, China Abhishek Ukil, Nanyang Technological University, Singapore Akshay Rathore, Concordia University Montreal, Canada Lalit Kumar Goel, Nanyang Technological University, Singapore Pinnamaneni Bhanu, Kelenn Technology Antony, France Armin Aberle, SERIS National University of Singapore Thomas Zimmer, University of Bordeaux, France Sebastien Fregonese, University of Bordeaux, France Shuliang Wang, Beijing Institute of Technology, China Bijaya Ketan Panigrahi, IIT Delhi, India G. Nath Pillai, IIT Roorkee, India

xi

xii

Preface

Organizing Commitee General Chairs Kusum Deep, Professor, IIT Roorkee Dhirendra Mathur, RTU, Kota Mahesh Bundele, Principal and Director, PCE, Jaipur Organizing Chairs D. K. Sambariya, RTU, Kota Harish Sharma, RTU, Kota Dinesh Goyal, Principal and Director, PIET, Jaipur Organizing Secretaries Irum Alvi, RTU, Kota Garima Mathur, PCE, Jaipur Ajay Sharma, GECJ, Jhalawar Publicity Chairs Nirmala Sharma, RTU, Kota Ajay Khuteta, PCE, Jaipur Sandeep Poonia, Amity University, Jaipur Registration Chairs Pankaj Dhemla, Vice Principal, PCE Punit Mathur, PIET, Jaipur Publication and Media Chairs Virendra Sangtani, PCE, Jaipur Rekha Jain, PIET, Jaipur Devendra Somwanshi, PCE, Jaipur

Technical Program Committee Chairs Marcin Paprzycki, Systems Research Institute, Polish Academy of Sciences, Warsaw, Poland Maria Ganzha, System Research Institute Polish Academy of Sciences and IBS PAN, Warsaw, Poland

Preface

xiii

Nilanjan Dey, Assistant Professor (Senior Grade) Department of Information Technology, Techno India College of Technology, Kolkata Co-chairs Emmanuel Pilli, MNIT Jaipur, India Pinaki Mitra, IIT Guwahati, India We are thankful to all the members of the steering committee, advisory committee, chairs and co-chairs of the technical program committee for their cooperation and support. Jaipur, India

Mahesh Bundele

Contents

1

2

3

4

5

Miniaturized Single-Layer Asymmetric CPW-Fed Antenna for UWB Applications . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Kirti Vyas, Dilip Gautam and Rajendra Prasad Yadav

1

Four Elements MIMO Antenna Array Having Band Notching Properties and High Isolation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Kirti Vyas, Dilip Gautam and Rajendra Prasad Yadav

13

Designing a Nonlinear Tri Core Photonic Crystal Fiber for Minimizing Dispersion and Analyzing it in Various Sensing Applications . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Sunil Sharma, Lokesh Tharani and Ravindra Kumar Sharma Hiding an Image Using Multi-object Vector Steganography and Variable Length Mixed Key Cryptography . . . . . . . . . . . . . . . Manju Kumari Gupta and Vipra Bohara Customer Attrition Estimation Modelling Based on Predominant Attributes Using Multi-layered Feed-Forward Neural Network . . . Vaishnavi Sidhamshettiwar, Yash Gaba, Rutika Jadhav and Kiran Gawande

6

A New Methodology to Implement ASCII-Based Cryptography . . . Shubha Agarwal

7

Sentiment Analysis of Product Reviews of Ecommerce Websites . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Shubhojit Sarkar and Souparna Palit

8

Novel FPGA-Based Hardware Design of Canonical Signed Digit Matrix Multiplier and Its Comparative Analysis with Other Multipliers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Ritik Koul, Mukul Yadav and Kriti Suneja

21

29

37

47

55

65

xv

xvi

9

Contents

Design of CMOS Instrumentation Amplifier Using Three-Stage Operational Amplifier for Low Power Signal Processing . . . . . . . . Shubham Saurabh, Mujahid Saifi, Shylaja V. Karatangi and Amrita Rai

10 A Novel Image Encryption Technique Using Arnold Transform and Asymmetric RSA Algorithm . . . . . . . . . . . . . . . . . . . . . . . . . . Gaurav Kumar Soni, Himanshu Arora and Bhavesh Jain 12 Designing of Band Reject Filter for Radio Astrophysics . . . . . . . . . Mudita Vats, Smriti Sachan, Shilpa Choudhary, Aprana Mishra and Vidit Shukla

77

83 91

13 Novel Hardware Design of Correlation Function and Its Application on Binary Matrix Factorization Based Features . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 101 Mayank Jain, Rahul Saini, Manish and Kriti Suneja 14 Design and Implementation of Biometrically Activated Self-Defence Device for Women’s Safety . . . . . . . . . . . . . . . . . . . . . 113 Sumiran Mehra, Shaurya Deep Singh, Subhangini Kumari, Shylaja Vinaykumar Karatangi, Reshu Agarwal and Amrita Rai 15 Facial Expression Recognition Using Random Forest Classifier . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 121 Kamlesh Tiwari and Mayank Patel 16 Design and Optimization of PV–Wind–DG and Grid-Based Hybrid Energy System for an Educational Institute in India . . . . . 131 Narendra Gothwal, Tanuj Manglani, Devendra Kumar Doda, Devendra Somwanshi and Mahesh Bundele 17 Fast Walsh–Hadamard Transform-Based Artificial Intelligent Technique for Transmission Line Fault Detection and Faulty Phase Recognition . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 141 Gaurav Kapoor, Vikas Soni and Jitendra Yadvendra 18 New Designs and Analysis of Multi-Core Photonic Crystal Fiber Using Ellipse with Different Radiuses and Angles . . . . . . . . . . . . . . 151 Trimesh Kumar, Mayank Sharma and Brijraj Singh Solanki 19 Air Quality Status in Jaipur and Nearby Areas of Rajasthan . . . . . 161 Harshit Tiwari, Meena Tekriwal and Rekha Nair 20 Recognizing MNIST Handwritten Data Set Using PCA and LDA . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 169 Ruksar Sheikh, Mayank Patel and Amit Sinhal

Contents

xvii

21 Optimal Design of a Stand-Alone Hybrid System Using GA, PSO and ABC . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 179 Jaswant Suthar, Virendra Sangtani and Ajay Kumar Bansal 22 Equal Distribution Based Load Balancing Technique for Fog-Based Cloud Computing . . . . . . . . . . . . . . . . . . . . . . . . . . . 189 Mandeep Kaur and Rajni Aron 23 Design of a Compact MIMO Antenna for RADAR Applications Using DGS Technology . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 199 Harshal Nigam, Monika Mathur and Mukesh Arora 24 Portfolio Management Using Artificial Intelligence . . . . . . . . . . . . . 207 Rakshit Gupta, Yogesh Mahajan, Punit Mukesh Ahuja and Jyoti Ramteke 25 Improved Fingerprint Recognition System Using Filtering Techniques . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 217 Devender Kumar Dhaked, Manish Sharma and Manish Mathuria 26 Image Security Using Triple Key Chaotic Encryption and SPIHT Compression Technique in Steganography . . . . . . . . . . . . . . . . . . . 227 Ajay Khunteta, Preeti Sharma, Sunil Pathak and Ajit Noonia 27 Power Quality Improvement in a Grid-Tied SPV System Using Fractional-Order Proportional Integral Controller . . . . . . . . 237 Pankaj Gakhar and Manoj Gupta 28 Conditional Status Detection for Factory Management System with Optimized Predictive Modeling . . . . . . . . . . . . . . . . . . . . . . . . 247 Sanil A. Raut and Vidya N. More 29 Obstacle Detection Approach for Robotic Wheelchair Navigation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 261 Devendra Somwanshi and Mahesh Bundele 30 An Effective Duplicate Removal Algorithm for Text Documents . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 269 Amit Jha, Devendra Somwanshi and Mahesh Bundele 31 Simulation and Optimization of Solar Photovoltaic–Wind–Diesel Generator Stand-alone Hybrid System in Remote Village of Rajasthan, India . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 279 Rahul Ranjan, Devendra Kumar Doda, Mahendra Lalwani and Mahesh Bundele 32 Reconfigurable Microstrip Patch Array Antenna: Design and Performance Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . 287 Ashok Kajla and Devendra Somwanshi

xviii

Contents

33 Lamp-Shaped Frequency Reconfigurable Microstrip Patch Array Antenna Design and Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 295 Ashok Kajla and Devendra Somwanshi 34 Static Cluster PBDA Localization Algorithm for Wireless Nanosensor Networks in Terahertz Communication Band . . . . . . . 303 Shruti Sharma and Deepak Bhatia 35 Design and Simulation of 3D MEMS Actuating System for Optical Scanning Application . . . . . . . . . . . . . . . . . . . . . . . . . . 311 Shivani Rathore and Deepak Bhatia 36 Design and Optimization of a Grid-Connected Hybrid Solar Photovoltaic-Wind Generation System for an Institutional Block . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 319 Devendra Somwanshi and Anmol Chaturvedi 37 ABCD Features Extraction-Based Melanoma Detection and Classification . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 327 Devendra Somwanshi, Anmol Chaturvedi and Pushpendra Mudgal 38 Design and Analysis of Modified Bully Algorithm for Leader Election in Distributed System . . . . . . . . . . . . . . . . . . . 337 Jayshree Surolia and Mahesh M. Bundele 39 Monitoring and Traffic Optimization Using Vertical Controller in Multi-domain SDN . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 349 Anmol Chaturvedi, Devendra Somwanshi, Mahesh Bundele and Charu Dubey 40 Detection of Induction Motor Bearing Fault Using Time Domain Analysis and Feed-Forward Neural Network . . . . . . . . . . . . . . . . . 359 Amit Shrivastava 41 A Review Paper on Sarcasm Detection . . . . . . . . . . . . . . . . . . . . . . 371 Ravinder Ahuja and S. C. Sharma 42 Handling Double Intensifiers in Feature-Level Sentiment Analysis Based on Movie Reviews . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 383 Pushpendra Mudgal and Ajay Khunteta Correction to: Conditional Status Detection for Factory Management System with Optimized Predictive Modeling . . . . . . . . . . . . . . . . . . . . . Sanil A. Raut and Vidya N. More

C1

Author Index . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 393

About the Editors

Dr. Garima Mathur is currently working as Head of Department, Poornima College of Engineering, Jaipur since 1st July 2016. Previously she was working as Head of Department in Electronics & Communication Engineering at Jaipur Engineering College, Jaipur, Rajasthan, India since Oct. 2000 to Dec. 2015. She has total 18 years experience in teaching and research. She is working on various research projects sponsored by various agencies. She did his doctoral degree in Performance Evaluation of Modified Sphere Decoding Scheme for MIMO Systems. Dr. Mathur published and presented International, National Journals, Conferences, Symposium and Seminar more than 30 research papers. She has guided more than 15 M.Tech. Dissertation thesis. Her area of interest is Wireless channel, Channel modeling, and Ad hoc Networks. Dr. Mathur is a life member of the Institution of Electronics and Telecommunication Engineers (IETE), The Indian Society for Technical Education (ISTE), India, and many other professional bodies.

xix

xx

About the Editors

Dr. Harish Sharma is an Associate Professor at Rajasthan Technical University, Kota in Department of Computer Science & Engineering. He has worked at Vardhaman Mahaveer Open University Kota, and Government Engineering College Jhalawar. He received his B.Tech and M.Tech degree in Computer Engg. from Govt. Engineering College, Kota and Rajasthan Technical University, Kota in 2003 and 2009 respectively. He obtained his Ph.D. from ABV Indian Institute of Information Technology and Management, Gwalior, India. He is secretary and one of the founder member of Soft Computing Research Society of India. He is a life time member of Cryptology Research Society of India, ISI, Kolkata. He is an Associate Editor of “International Journal of Swarm Intelligence (IJSI)” published by Inderscience. He has also edited special issues of the journals “Memetic Computing” and “Journal of Experimental and Theoretical Artificial Intelligence”. His primary area of interest is nature inspired optimization techniques. He has contributed in more than 45 papers published in various international journals and conferences. Dr. Mahesh Bundele is currently working as Principal and Director of Poornima College of Engineering, Jaipur since 1st September 2018. Previously he was working as Professor in Computer Engineering and Dean (Research & Development) at Poornima University, Jaipur, Rajasthan, India since 2011. He was heading Advanced Studies Research Center and responsible for design, development and execution of research curriculum and processes for Masters in Engineering and Doctoral Program in School of Engineering, Sciences and Management. He has total 31 years experience in teaching and research. He has developed many unique research methodology concepts and implemented. He is the mentor and controller of quality research and publications at the University. He is also responsible for inculcation of innovative and critical analysis concepts across the University and across the Poornima Foundation involving three other campuses. He is working on various research projects sponsored by various agencies. He did his doctoral

About the Editors

xxi

degree in Wearable Computing and guiding research in Pervasive & Ubiquitous Computing, Computer Networks, Software Defined Networking. His areas of interests are also Wireless Sensor Networks, Algorithmic research, Mathematical Modeling, and Smart grids. He has earlier worked at different levels such as Lecturer, Associate Professor and Professor & Head of Computer Engineering & Information Technology and lastly as Principal of Babasaheb Naik College of Engineering, Pusad, Yavatmal, MS, India from 1986 to 2011. He has worked on many government and non government research projects and guided bachelors and master degree students for their research. He has more than 30 publications in reputed journals and conferences. He has been reviewer of few IEEE Transactions. He is actively involved in IEEE activities in Rajasthan Subsection and Delhi Section and holding the responsibility on Standing Committee of IEEE Delhi Section for Technical & Professional activities for controlling quality of conferences and publications in IEEE. He is also holding position of membership development chair at IEEE Rajasthan Subsection. Dr. Nilanjan Dey is an Assistant Professor in Department of Information Technology at Techno India College of Technology, Kolkata. He has completed his PhD. in 2015 from Jadavpur University. He is a Visiting Fellow of Wearables Computing Laboratory,Department of Biomedical Engineering Univeristy of Reading, UK, Visiting Professor of College of Information and Engineering, Wenzhou Medical University, P.R. China and Duy Tan University, Vietnam. He has held honorary position of Visiting Scientist at Global Biomedical Technologies Inc., CA, USA (2012-2015). He is the Editor-in-Chief of International Journal of Ambient Computing and Intelligence, IGI Global, series Co-Editor of Springer Tracts in Nature-Inspired Computing, Springer, Advances in Ubiquitous Sensing Applications for Healthcare (AUSAH), Elsevier and the series editor of Intelligent Signal processing and data analysis, CRC Press. He has authored/edited more than 40 books with Elsevier, Wiley, CRC and Springer, and

xxii

About the Editors

published more than 350 research articles. His main research interests include Medical Imaging, Machine learning, Bio-inspired computing, Data Mining etc. He is a life member of Institute of Engineers (India). Dr. Marcin Paprzycki is an associate professor at the Systems Research Institute, Polish Academy of Sciences. He has an MS from Adam Mickiewicz University in Poznan, Poland, a Ph.D. from Southern Methodist University in Dallas, Texas, and a D.Sc. Degree from the Bulgarian Academy of Sciences. He is a senior member of IEEE, a senior member of ACM, a Senior Fulbright Lecturer, and an IEEE CS Distinguished Visitor. He has contributed to more than 450 publications and was invited to the program committees of over 500 international conferences. He is on the editorial boards of 12 journals and a book series.

Chapter 1

Miniaturized Single-Layer Asymmetric CPW-Fed Antenna for UWB Applications Kirti Vyas, Dilip Gautam and Rajendra Prasad Yadav

1 Introduction In 2002, FCC declared that UWB spectrum (3.1–10.6 GHz) can be used for wide range of applications such as data communications, radar, sensors, positioning and imaging [1]. Some of feedline techniques used for antenna are microstrip slot antenna [2–4], symmetric CPW [5–10], asymmetric CPW [8, 11], microstrip line with via hole [12, 13], microstrip line [14–20] and microstrip-line-fed antennas having an electrically conductive adhesive coated on the substrate [21]. UWB antennas proposed in [2–4, 12–21] are quite big in size and have metallization on both sides of substrate. Easy implementation of the shunt and the series connections on single side of substrate is not possible with these antennas. However, some work has been done on the coplanar feed antennas [5–10], but these antennas are quite bigger in size which can be miniaturized further. The design approaches used in antennas [5, 6, 17 and 18] were complicated and resulted in complex structure which can introduce fabrication errors. However, few of the antennas barely cover a little part UWB spectrum [10, 18 and 20]. The complex antenna structure proposed in [5, 18] has resulted in distorted radiation patterns for the E-plane and H-plane. Here, a compact single-layer UWB antenna is proposed using modification in ground plane of the coplanar feed structure. Stubs of optimum dimensions are added to the asymmetric CPW-fed antenna to control flow of current in the antenna which resulted in wideband characteristic for the miniaturized antenna. Methodology of miniaturization, antenna design and analysis is explained later in this paper. The simulation and measured experimental results are found in good agreement. K. Vyas (B) · D. Gautam Arya College of Engineering and I.T., Kukas, Jaipur, Rajasthan, India e-mail: [email protected] R. P. Yadav MNIT, Jaipur, Rajasthan, India © Springer Nature Singapore Pte Ltd. 2020 G. Mathur et al. (eds.), International Conference on Artificial Intelligence: Advances and Applications 2019, Algorithms for Intelligent Systems, https://doi.org/10.1007/978-981-15-1059-5_1

1

2

K. Vyas et al.

2 Design and Analysis of Proposed Antenna The design layout of proposed UWB antenna (miniaturized) is publicized in Fig. 1. The radiator is printed on inexpensive FR4 substrate (relative permittivity ‘1r ’ = 4.4), uniform thickness of 1.6 mm and having the loss tangent value 0.02. The antenna is uniplanar, having metallization only on one side of the substrate and has overall volume of 18 × 16 × 1.6 mm3 . Figure 1 depicts optimized dimensions of proposed asymmetric CPW-fed UWB antenna. Ground plane for the antenna geometry is asymmetric coplanar waveguide having an extended stub (inverted ‘L’ shaped) added to right side of ground plane. The radiator of the antenna comprises of conventional rectangular radiator modified with an extended quasi-circular patch at the bottom for impedance matching. The antenna is fed by asymmetric coplanar waveguide structure consisting of metallic strip in the center with width ‘W f ’ and gap ‘g’ amid central conducting strip and the ground plane. Figure 2 shows the fabricated proposed antenna. Antenna dimensions are optimized by various parametric simulations from CST MWS. The optimized length (in mm) of both ground planes is ‘L g ’ = 4.5, and the width of ground plane on left and right sides are ‘W 1 ’ = 5.8 and ‘W 2 ’ = 8, respectively. Various other parameters of antenna are ‘L’ = 18, ‘W ’ = 16 ‘W p ’ = 11.6, ‘L p ’ = 8, ‘L s2 ’ = 7, ‘W s2 ’ = 0.5, ‘R’ = 5.3, ‘W s1 ’ = 2.4, ‘L s1 ’ = ‘L’−‘L g ’ = 13.5, ‘W f ’ = 1.8, ‘g’ = 0.3. Various development stages of proposed antenna are shown in Fig. 3. In order to design the antenna primarily, a printed rectangular monopole antenna (PRMA) fed by 50  asymmetric CPW line is printed on FR4 substrate of dimension 18 × 16 × 1.6 mm3 as publicized in Fig. 3a. Lower cut-off frequency of printed monopole antenna is given by equation [22]: Fig. 1 Configuration of UWB antenna

1 Miniaturized Single-Layer Asymmetric CPW-Fed Antenna …

3

Fig. 2 Fabricated prototype

Fig. 3 Design evolution of proposed asymmetric UWB antenna a Antt.-1; b Antt.-2; c Antt.-3; and d Antt.-4

fl =

7.2 GHz W +r + p

(1)

Here, W is width of equivalent cylindrical-shaped monopole, having radius ‘r’ = L/2π and ‘length of feed line is p’ between bottom of the rectangular patch and ground plane. For proposed antenna (see Fig. 3a), W = 1.16 cm, r = 0.8/2π cm and p = 0.15 cm. The calculated lower operating frequency ‘f l ’ is 5 GHz which exactly matches the simulated results. This antenna exhibits small impedance bandwidth centered at around 7.25 GHz ranging from 5 to 9.5 GHz as shown in Fig. 4. This small bandwidth is the results of abrupt truncation of the feedline leading to discontinuity in the feed point by a straight bottom of the PRMA [23]. In order to reduce this discontinuity, the bottom side of the rectangular radiator is modified, and a quasicircular patch is added as shown in Fig. 3b. This resultant antenna-2 generates two

4

K. Vyas et al.

closely spaced resonance modes at about 7 and 8.86 GHz which overlap to give improved impedance bandwidth from 5.5 to 11.1 GHz (shown in Figs. 4, 5). In third stage due to added extended open-circuited vertical stub antenna-3 configuration creates closely spaced resonance modes at about 3.7, 5.4 and 9.65 GHz; and achieves improvement in impedance matching in 3–10.7 GHz range. To enhance impedance matching for targeted 7–8 GHz band; an additional horizontal stub is attached on top of the modified ground plane shown in Fig. 3d. This final stage (Antt.4) shows resonance modes at around 3.2/5.7/10.7 GHz which overlap to achieve the impedance bandwidth from 2.7 to 12 GHz and hence covering the entire UWB range. Current distribution at showing various modes at resonance frequencies 3.2, 5.7 and 10.7 GHz is shown in Fig. 6. Fundamental mode of the proposed antenna is shown Fig. 4 Simulated S 11 curves

Fig. 5 Simulated and measured S 11 results

1 Miniaturized Single-Layer Asymmetric CPW-Fed Antenna …

5

Fig. 6 Stimulated surface current distributions of proposed antenna: a 3 GHz, b 5.8 GHz and c 10.5 GHz

in Fig. 6a that creates resonance at around 3.2 GHz leading to current distribution pattern along ‘y’ direction of the quasi-circular patch. Figure 6b shows the current distribution of the second resonance mode at 5.7 GHz (i.e., at first harmonic of the fundamental mode). Figure 6c shows the current distribution at third resonant mode at 10.7 GHz which is a second harmonic of the fundamental mode. The harmonics of the fundamental modes have overlapped and achieved huge impedance bandwidth covering the complete UWB range as shown in Fig. 4.

3 Results with Discussions Fabricated antenna is measured for value of S 11 using Agilent’s PNA E8364C. Comparison among measured and simulated S 11 values is given in Fig. 5. The antenna covers impedance bandwidth in 2.7–12 GHz range for magnitude of S 11 < −10 dB. Next section presents a detailed study of modification techniques used for the antenna design.

3.1 Influences on Impedance Bandwidth Parametric analyses for impedance bandwidth of proposed antenna are simulated in CST MWS. Here, we analyze the improvement in the impedance bandwidth due to the proposed technique.

3.1.1

Study for Modifications in Ground Plane

Since the modified ground plane in this proposed antenna comprises two stubs added to conventional ground plane, parametric analysis is applied on the width and lengths

6

K. Vyas et al.

Fig. 7 Effect by variation of ‘L s1 ’ and ‘W s1 ’ on proposed antenna (all dimensions in mm)

of both stubs to show the overall effect on impedance bandwidth of antenna. Figure 7a depicts the S 11 value for different stub lengths such as ‘L s1 ’ (in mm) = 11.5, 12.5, 13.5, and Fig. 7b presents S 11 value for different stub widths, i.e., ‘W s1 ’ (in mm) = 1.4, 2.4, 3.5. It can be noticed that for the values of ‘L s1 ’ below 13.5 mm, proposed antenna gives poor impedance match in low-frequency band ranging from 3.5 to 4.5 GHz because the resonance modes created by ‘L s1 ’ = 11.5 mm, 12.5 mm are not closely spaced at low frequencies and hence do not overlap to cover UWB range. The stub length of ‘L s1 ’ = 13.5 mm creates sharp resonance mode at 3.2 GHz which has overlapped with the resonance mode at 5.7 GHz to provide better impedance matching. With stub width ‘W s1 ’ = 3.4 mm (see Fig. 7b), the antenna has resonance mode at only one frequency at 6.5 GHz and does not cover entire ultra-wideband. ‘W s1 ’ = 1.4 mm results in non-overlapping bands due to widely spaced resonance modes at 2.9 and 6.5 GHz which degrades impedance match of the antenna in 3.5– 4.5 GHz band; whereas with ‘W s1 ’ = 2.4 mm impedance match is achieved for entire UWB region. So ‘W s1 ’ = 2.4 and ‘L s1 ’ = 13.5 mm are chosen as optimized dimensions for the first stub. The far-field measurements of proposed antenna are done in anechoic chamber by means of Agilent PNA-L network analyzer N5234A. Measured patterns of proposed UWB antenna (E-plane (yz-plane) and H-plane (xz-plane)) of the antenna are presented in Fig. 8a–f for four different frequencies at 3.2, 5.7 and 9 and 10.7 GHz, respectively. The E-plane radiation patterns are having bidirectional radiation making ‘eight’ shaped figure comparable to typical dipole antenna. Radiation patterns obtained in the H-plane are omnidirectional at 3.2, 5.7 and 9 and 10.7 GHz, respectively. Figure 9 gives magnitude and the phase variations of pulse transmission coefficient S 21 against frequency. Phase variation is noticed linear in the entire operating band. The linearity in phase variation against frequency guarantees that all frequency components of input signal have the same delay. Figure 10 depicts the simulated group delay variation with frequency for both face-to-face and the side-to-side alignment of the antennas for distance of 30 cm in open-space environment. It can be noted

1 Miniaturized Single-Layer Asymmetric CPW-Fed Antenna …

3.2 GHz (E Plane)

5.7 GHz (E Plane)

9 GHz (E plane)

3.2 GHz (H plane)

5.7 GHz (Hplane)

9 GHz (H plane)

Fig. 8 Measured patterns (hard line: co-polar, dashed line: cross-polar component)

7

8

K. Vyas et al.

Fig. 9 Magnitude and phase of S 21 verses frequency plot

Fig. 10 Group delay analysis

from Fig. 10 that group delay variation is restricted to lower than 1 ns and has a flat response in entire UWB range. Measured gain (dB) curve of proposed UWB antenna is represented in Fig. 11. Gain variation is within 3 dB for UWB region. The antenna dimensions and performances of various UWB printed antennas are compared in Table 1. Proposed antenna achieves good impedance bandwidth characteristics with a small size. Antenna in [11] achieved a compact profile, and however, it failed to cover the entire ultra-wide bandwidth. All other antennas are bigger than the proposed antenna, and some [19, 21] fail to cover the entire ultra-wideband. The proposed antenna design outperforms available printed antenna structures presented in the literature in comparison with dimensions while achieving satisfactory standards of electrical parameters for operation in ultrawideband.

1 Miniaturized Single-Layer Asymmetric CPW-Fed Antenna …

9

Fig. 11 Gain verses frequency curve Table 1 Performance comparison from some recently reported UWB antennas References

Active area (mm2 )

Feeding method

Substrate

Operating band (GHz)

Percentage reduction in active area

[2]

20 × 20

Microstrip slot

FR–4

3–13.6

28%

[4]

26 × 38.5

Microstrip slot

Rogers RT6010

2.63–14.2

71%

[7]

43 × 28

Symmetric CPW

FR–4

2.27–2.57 and 3.1–14

[11]

16 × 14

Asymmetric CPW

FR–4

3.75–19

Does not cover lower UWB

[12]

28 × 20

Microstrip line with via hole

FR–4

3.1–10.6

48.57%

[14]

27 × 36

Microstrip line

FR–4

2.8–14

70.37%

[15]

30 × 30

Microstrip line

FR–4

2.39–2.49 & 3.1–11.4

[17]

36 × 38

Microstrip line

FR–4

2.86–13.3

78.94%

[18]

38 × 38

Microstrip line

FR–4

2.95–14.28

80.05%

76.07%

68%

(continued)

10

K. Vyas et al.

Table 1 (continued) References

Active area (mm2 )

Feeding method

Substrate

Operating band (GHz)

Percentage reduction in active area

[20]

30 × 30

Microstrip line

FR–4

2.2–11

68%

This work

18 × 16

Asymmetric CPW

FR–4

2.7–12

Smallest of all antennas

4 Conclusion An original technique is projected to miniaturize conventional coplanar-fed rectangular patch antenna. We have introduced the technique of modifications in ground plane and radiator geometry of the antenna for the desired UWB communication. A simple approach of adding an inverted stub in ‘L’ shape at one of ground plane of asymmetric CPW-fed antenna is introduced for better impedance matching. Proposed antenna comprises small electrical dimensions of 0.16 λl × 0.14 λl × 0.015 λl at least frequency of operation. Magnitude of S 11 is below −10 dB in 2.7–12 GHz band, thus covering more than the desired ultra-wideband ranging from 3.1 to 10.6 GHz. The peak gain is 3.6 dB at 7 GHz in the ultra-wideband range. The antenna has satisfactory electrical performance, and it can be easily used in portable ultra-wideband devices. Acknowledgements The authors thank Dr. Deepak Bhatnagar from the University of Rajasthan, Jaipur, and Dr. K. Vaibhav Srivastava from IIT Kanpur, for their facilities.

References 1. Federal Communications Commission (2002) Revision of part 15 of the communication’s rules regarding ultra-wideband transmission systems, ET-Docket 98–153, FCC 02–48 2. Kakhki MB, Rezaei P (2017) Reconfigurable microstrip slot antenna with DGS for UWB applications. Int J Microwave Wirel Technol 9(7):1517–1522 3. Ojaroudi N (2014) Application of protruded strip resonators to design an UWB slot antenna with WLAN band-notched characteristic. Prog In Electromagn Res C 47:111–117 4. Al-Zuhairi DT, Gahl JM, Al-Azzawi A, Islam NE (2017) Simulation design and testing of a dielectric embedded tapered slot UWB antenna for breast cancer detection. Prog In Electromagn Res C 79:1–15 5. Zhang C, Zhang J, Li L (2014) Triple band-notched UWB antenna based on SIR-DGS and fork-shaped stubs. Electron Lett 50(2):67–69 6. Singhal S, Pandey A, Singh AK (2015) CPW-fed circular-shaped fractal antenna with three iterations for UWB applications. Int J Microwave Wirel Technol 9(2):373–379 7. Abdulhasan RA, Alias R, Ramli KN (2017) A Compact CPW fed UWB antenna with quad band notch characteristics for ISM band applications. Prog In Electromagn Res M 62:79–88

1 Miniaturized Single-Layer Asymmetric CPW-Fed Antenna …

11

8. Vyas K, Sharma AK, Singhal PK (2014) Design and analysis of two novel CPW-Fed dual band-notched UWB antennas with modified ground structures. Prog In Electromagn Res C 49:159–170 9. Xu K, Zhu Z, Li H, Huangfu J, Li C (2013) A printed single-layer UWB monopole antenna with extended ground plane stubs. IEEE Antennas Wirel Propag Lett 12:237–240 10. Vyas K, Sanyal G, Sharma A, Singhal P (2013) Gain enhancement over a wideband in CPW-fed compact circular patch antenna. Int J Microwave Wirel Technol 6(5):497–503 11. Singhal S, Singh AK (2017) Asymmetrically CPW-fed hourglass shaped UWB monopole antenna with defected ground plane. Wirel Pers Commun 94(3):1685–1699 12. Takemura N, Ichikawa S (2017) Broadbanding of printed bell-shaped monopole antenna by using short stub for UWB applications. Prog In Electromagn Res C 78:57–67 13. Lin S, Cai RN, Huang GL, Wang JX (2011) A miniature UWB semi-circle monopole printed antenna. Prog In Electromagn Res Lett 23:157–163 14. Mewara HS, Jhanwar D, Sharma MM, Deegwal JK (2018) A printed monopole ellipsoidal UWB antenna with four band rejection characteristics. Int J Electron Commun 83:222–232 15. Yang B, Qu S (2017) A compact integrated Bluetooth UWB dual-band notch antenna for automotive communications. Int J Electron Commun 80:104–113 16. Liu J, Yin Y (2016) Triple band-notched UWB Antenna using novel asymmetrical resonators. Int J Electron Commun 70(12):1630–1636 17. Mewara HS, Deegwal JK, Sharma MM (2018) A slot resonators based quintuple band-notched Y-shaped planar monopole ultra-wideband antenna. Int J Electron Commun 83:470–478 18. Saraswat RK, Kumar M (2016) Miniaturized slotted ground UWB antenna loaded with metamaterial for WLAN and WiMAX applications. Prog In Electromagn Res B 65:65–80 19. Behdad N, Li M, Yusuf Y (2013) A very low-profile, omnidirectional, ultra wideband antenna. IEEE Antennas Wirel Propag Lett 12:280–283 20. Sanyal R, Patra A, Sarkar P, Chowdhury S (2015) Frequency and time domain analysis of a novel UWB antenna with dual band-notched characteristics. Int J Microwave Wirel Technol 9(2):427–436 21. Reyes-Vera E, Arias-Correa M, Giraldo-Muñoz A, Cataño-Ochoa D, Santa-Marín J (2017) Development of an improved response ultra-wideband antenna based on conductive adhesive of carbon composite. Prog In Electromagn Res C 79:199–208 22. Ray KP (2008) Design aspects of printed monopole antennas for ultra-wide band applications. Int J of Antennas Propag 2008:1–8 23. Ray KP, Ranga Y, Gabhale P (2007) Printed square monopole antenna with semicircular base for ultra-wide bandwidth. Electron Lett 43(5):13–14

Chapter 2

Four Elements MIMO Antenna Array Having Band Notching Properties and High Isolation Kirti Vyas, Dilip Gautam and Rajendra Prasad Yadav

1 Introduction In the present scenario, lot of research work is going on in finding the solution to enhance the channel capacity and improve the quality of communication for UWB applications. In this direction, the MIMO antennas have proved themselves as a good solution to this problem. However, the MIMO antenna designing is very challenging as it requires more than −15 dB isolation as per standard recommendations. One of the design challenges for MIMO antennas are that it requires low envelop correlation coefficient and high diversity gain. In the literature, a lot of UWB MIMO antennas [1–6] are designed but they are only two element configurations. Feeding technique is also important while designing UWB antennas as CPW-fed antennas have the advantage that they are uniplanar and can be easily integrated with MMICs. Scarce literature is available on four elements UWB MIMO antenna [7–11] with uniplanar designs [12]. These MIMO antennas have limitations as antenna reported in ref 10 does not cover entire UWB, and some antennas [8, 11] are larger in size. Some antennas do not have band notching characteristics [8, 9, 10, and 12] and suffer from narrowband interference. It is desired to have band notching characteristics in UWB antenna so as to mitigate the interference from coexisting narrowband applications. The isolation reported in [7–9, 11 and 12] is also low and can be improved further. In order to achieve high isolation with band-notched characteristic, this paper presents a novel design of four elements UWB MIMO antenna array. Here, we propose an UWB MIMO antenna with rejected WLAN band through CSRR and improved isolation greater than −20 dB in almost entire band. K. Vyas (B) · D. Gautam Arya College of Engineering and I.T., Kukas, Jaipur, Rajasthan, India e-mail: [email protected] R. P. Yadav MNIT, Jaipur, Rajasthan, India © Springer Nature Singapore Pte Ltd. 2020 G. Mathur et al. (eds.), International Conference on Artificial Intelligence: Advances and Applications 2019, Algorithms for Intelligent Systems, https://doi.org/10.1007/978-981-15-1059-5_2

13

14

K. Vyas et al.

2 MIMO Antenna Design The single element CPW-fed antenna used for design of MIMO array is shown in Fig. 1. The antenna geometry comprises the rectangular radiator with a circular bottom and covers 18 × 16 × 1.6 mm3 . The antenna has an extended ground inverted ‘L’-shaped stub for achieving UWB range. The optimized length (in mm) of both ground planes is ‘L g ’ = 4.5, and the widths of the ground plane on left and right sides are ‘W 1 ’ = 5.8 and ‘W 2 ’ = 8, respectively. Various other parameters of the antenna are ‘L’ = 18, ‘W ’ = 16, ‘W p ’ = 11.6, ‘L p ’ = 8, ‘Ls2 ’ = 7, ‘W s2 ’ = 0.5, ‘R’ = 5.3, ‘W s1 ’ = 2.4, ‘L s1 ’ = ‘L–L g ’ = 13.5, ‘W f ’ = 1.8, ‘g’ = 0.3 and ‘g1 ’ = 0.8. To the best of our knowledge, this antenna is smallest single-layer antenna which covers entire ultra-wideband. The antenna design used for the MIMO is presented in Fig. 2. MIMO antenna array is 44 × 44 x 1.6 mm3 and is printed on FR4.

Fig. 1 Layout of single element

2 Four Elements MIMO Antenna Array Having Band Notching …

15

Fig. 2 Proposed MIMO antenna array

3 Discussions with Results Simulated S 11 , S 22 , S 33 and S 44 of antenna are drawn in Fig. 3. The surface currents on the antenna at 5.5 GHz center-notched frequency with CSRR explaining the bandnotched characteristics are demonstrated in Fig. 4a–d. These current distributions are shown with single port excited at one time and all rest terminated using 50  impedance. The current scale shown at right-hand side of the figure is scaled from 0 to 12 A/m representing the current intensity within the antenna elements when individual ports are excited at 5.5 GHz. Antenna elements at 5.5 GHz have maximum oppositely directed current distribution on surface around the periphery of the CSRR.

Fig. 3 S 11 , S 22 , S 33 and S 44 versus frequency curve of the MIMO antenna

16

K. Vyas et al.

Fig. 4 Surface current distribution at 5.5 GHz

Due to this current distribution, the impedance of the antenna increases which results in band-notched characteristics. The isolation values between antenna elements are shown in Fig. 5. It can be noticed that isolation achieved between individual antennas is more than −20 dB except between the port 1 and port 2 where minimum isolation value is −18.1 dB for 7.3–8.7 GHz and for 10.3–10.6 GHz band. The radiation patterns with the port 1 excited to and rest ports terminated with matched impedances are publicized in Fig. 6a–f. These patterns are monopole type with H plane having unidirectional pattern and E plane having shape of figure of eight. The patterns at high frequency are distorted due to the generation of higher modes. Figure 7 shows the gain verses frequency curve of the individual antenna element. Significant band notching can be seen at notched WLAN band. The gain of the antenna is moderate, and maximum peak gain is 4.9 dB at 7.5 GHz.

2 Four Elements MIMO Antenna Array Having Band Notching …

17

Fig. 5 S parameter verses frequency curve

(a) 6 GHz

(d) 6 GHz

(b) 8 GHz H Planes

(c) 10 GHz

(e) 8 GHz

(f) 10 GHz

E Planes Fig. 6 Radiation patterns with port 1 excited

18

K. Vyas et al.

Fig. 7 Gain verses frequency curve of individual antenna

ECC is a crucial parameter to find diversity performance of a MIMO antenna. In MIMO antennas, ECC is found by using Eq. (2): ∗ ∗ S12 + S21 S22 |2 |S11     ρe =  1 − |S11 |2 + |S21 |2 1 − |S22 |2 + |S12 |2

(1)

ECC value below 0.5 guarantees good diversity performance, and here we are having maximum ECC value as 0.006 which is quite good (Fig. 8). Diversity gain obtained is close to ideal 10 dB for nearly complete UWB band with the lowest value of 9.97 dB at center band-notched frequency. Diversity gain is calculated by using the equation:  (2) Diversity gain  Dg = 10 × 1 − |ρe |

4 Conclusion The paper reports a novel four element MIMO antenna array with enhanced isolation through a novel decoupling structure. The antenna covers 3–12 GHz band with more than −20 dB isolation within all elements in entire band. The maximum ECC value of the four element antenna is 0.006 with minimum 9.97 dB diversity gain. The antenna has moderate gain and can be used in UWB MIMO applications. In future, the authors will try to avoid interference from other narrow band applications such as WiMax and aeronautical radio navigation (ARN) bands by applying band notching characteristics.

2 Four Elements MIMO Antenna Array Having Band Notching …

19

Fig. 8 a Envelop correlation coefficient (ECC) verses frequency plot b Diversity gain verses frequency plot

Acknowledgements The authors thank Dr. Deepak Bhatnagar from University of Rajasthan, Jaipur, for allowing using their research work.

References 1. Sipal D, Abegaonkar MP, Koul SK (2016) Compact band-notched UWB antenna for MIMO applications in portable wireless devices. Microw Opt Technol Lett 58(6):1390–1394 2. Latif F, Tahir FA, Khan MU, Sharawi MS (2017) An ultra-wide band diversity antenna with band-rejection capability for imaging applications. Microw Opt Tech Lett 59(7):1661–1668

20

K. Vyas et al.

3. Gorai A, Dasgupta A, Ghatak R (2018) A compact quasi-self-complementary dual band notched UWB MIMO antenna with enhanced isolation using hilbert fractal slot. Int J Electr Commun 94:36–41 4. Li J-F, Wu D-L, Wu Y-J (2017) Dual band-notched UWB MIMO antenna with uniform rejection performance. Prog Electromagnet Res M 54:103–111 5. Banerjee J, Karmakar A, Ghatak R, Poddar DR (2017) Compact CPW-fed UWB MIMO antenna with a novel modified Minkowski fractal defected ground structure (DGS) for high isolation and triple band-notch characteristic. J Electromagnet Waves Appl 31(15):1550–1565 6. Zhang S, Lau BK, Sunesson A, He S (2012) Closely-packed UWB MIMO/diversity antenna with different patterns and polarizations for USB dongle applications. IEEE Trans Antennas Propag 60(9):4372–4380 7. Kiem NK, Phuong HNB, Chien DN (2014) Design of compact 4 × 4 UWB-MIMO antenna with WLAN band rejection. Int J Antennas Propag 2014 8. Wu W, Yuan B, Wu A (2018) A quad-element UWB-MIMO antenna with band-notch and reduced mutual coupling based on EBG structures. Int J Antennas Propag 2018 9. Yu JF, Liu XL, Shi XW, Wang ZA (2014) A compact four-element UWB MIMO antenna with QSCA implementation. Prog Electromagnet Res Lett 50:103–109 10. Liu X-L, Wang Z-D, Yin Y-Z, Ren J, Wu J-J (2015) A compact ultrawideband MIMO antenna using QSCA for high isolation. IEEE Antennas Wirel Propag Lett 13:1497–1500 11. Khan MS, Capobianco AD, Asif S, Iftikhar A, Ijaz B, Braaten BD (2015) Compact 4X4 UWB-MIMO antenna with WLAN band rejected operation. Electron Lett 51:1048–1050 12. Yang B, Chen M, Li L (2018) Design of a four-element WLAN/LTE/UWB MIMO antenna using half-slot structure. Int J Electr Commun 93:354–359

Chapter 3

Designing a Nonlinear Tri Core Photonic Crystal Fiber for Minimizing Dispersion and Analyzing it in Various Sensing Applications Sunil Sharma, Lokesh Tharani and Ravindra Kumar Sharma

1 Introduction Nowadays, optical fiber sensors are playing an important role in bio-sensing applications. Depending on the applications, suitable types of fiber sensors have been produced timely. The aim of these sensors is to provide better sensitivity and high detection capabilities. Thereafter, these tri-core fibers can be applied to sense various substances like salinity, glucose, stress, and temperature. This kind of sensing PCF was introduced by Philip Russell in 1991. In 1996, his group declared the first example of working PCF. Due to their unique properties, such types of fibers showed more impact in sensing applications and widely used in high compact sensitivity aids like temperature sensors, magnetic field sensors, pressure sensors, stress monitoring devices, and seawater salinity sensors. Apart from these sensing devices, the photonic crystal fibers-based sensors have been widely used in social terms and society-based applications. It includes various designs of single channel, semi channel, and multi channel sensors along with different materials. For example, the detection of glucose in human blood and stress level of human beings can be easily detected with such types of tri-core PCF. It is being found that sensitivity of glucose is observed at 19,007.17 nm/RIU with the use of sensor-based PCF. This sensitivity can be further increased by using Sagnac interferometer along with sensor-based PCF. With this, the sensitivity so observed is 22,120 nm/RIU.

S. Sharma (B) · L. Tharani Department of Electronics & Communication Engineering, Rajasthan Technical University, Kota, India e-mail: [email protected] R. K. Sharma Arya College of Engineering and I.T, Kukas, Jaipur, India e-mail: [email protected] © Crown 2020 G. Mathur et al. (eds.), International Conference on Artificial Intelligence: Advances and Applications 2019, Algorithms for Intelligent Systems, https://doi.org/10.1007/978-981-15-1059-5_3

21

22

S. Sharma et al.

2 Methodology The design process of such sensor-based tri-core photonic crystal fiber is done by opti-FDTD. Number of parameters has been considered while the design process is accomplished. Initially, the diameter range of the air hole is decided. It varies from 0.5 to 1.5 µm range. Thereafter, the distance between two air holes, which is known as PICTH, is considered as 2.0 µm. Three layers design is being proposed for this structure. Optical refraction, dispersion, absorption, scattering, and luminescence are also easily possible by silica glass. So that silica material is used for the preparation of such type of fiber. Constant index profile: n(x) = constant

(1)

n(x) = n(0) + x.{n(w) − n(0)}/w

(2)

n(x) = [n(w) − n(0)].(x/n)2 + n(0)

(3)

Linear index profile:

Parabolic index profile:

Exponential index profile: n(x) = [n(w) − n(0)].e/(e − 1).exp (−x/w) + {e. n(w) − n(0)}/(e − 1) (4) where n(0) and n(w) show the refractive index at x = 0 and x = w, respectively. The total chromatic dispersion which is a summation of material dispersion and waveguide dispersion is obtained for the proposed structure as given below: D=

λ d 2 Re[sn eff ] ps/(nm − km) c dλ2

(5)

Here, we have used finite element method (FEM) to compute any function (temperature, electric potential, pressure, stress, etc.) u. Some basis functions have been shown as follows: u ∼ uh

(6)

and uh =

 i

u i ϕi

(7)

3 Designing a Nonlinear Tri Core Photonic Crystal Fiber …

23

where ψ i represented as the basis functions and ui represented as the coefficients of the functions which approximate u with uh .

3 Proposed Design of Sensor-Based Tri-Core Silica Photonic Crystal Fiber The proposed design of sensor-based tri-core silica photonic crystal fiber is prepared with opti-FDTD. As discussed above, parameters have been selected for this design, and it is shown below in Fig. 1. This design is considered as tri-core PCF having central core as tapered in circular shape and remaining cores in blocked shapes. To obtain the R.I. of the proposed work, the structure is shown in contour form as given below in Fig. 2. Fig. 1 Proposed design of sensor-based tri-core silica PCF

Fig. 2 3D viewer of proposed design

24

S. Sharma et al.

This contour 3D views shown below in Fig. 2 of proposed design show the real and imaginary values of design in 3D. The air holes of this tri-core PCF can be used to evaluate the refractive index of the proposed design. Amplitude and phase variation of proposed design can also be obtained by the 3D viewer. It represents the variation of occurrence of amplitude and phase in terms of pictograph. It has four data sets to show this variation as shown in Fig. 3. Here, electric field distribution (EFD) of a fundamental core mode of tri-core silica PCF is given below in Fig. 4. FEM is used to calculate this distribution. It is used to obtain the outer region of the proposed design of PCF. This can be further used to absorb radiated energy to prevent the reflection of waste energy and provide perfect boundary conditions. Figure 5 presents the refractive index of the proposed sensor-based tri-core silica PCF design. It shows that refractive index of the proposed design is approximately 1.456 in the wavelength range of 0.2 to 0.9 µm.

Fig. 3 Amplitude and phase variation of proposed structure

Fig. 4 Electric field distribution of fundamental glucose core mode of tri-core PCF

3 Designing a Nonlinear Tri Core Photonic Crystal Fiber …

25

Fig. 5 Analysis of refractive index of the proposed structure

Figure 6 is representing the dispersion of the proposed sensor-based tri-core silica PCF. It shows that it is nearly zero in between 0.4 to 1.8 µm wavelength range. Figure 7 shows the confinement loss of the proposed structure. Fig. 6 Dispersion of the proposed structure

26

S. Sharma et al.

Fig. 7 Confinement loss of the proposed structure in fundamental glucose core mode of tri-core PCF

4 Conclusion Author tried to design high sensitivity sensor-based tri-core silica PCF. This design is able to achieve this goal. Finite element method (FEM) is used to accomplish it perfectively along with Maxwell equations. The results so obtained show that a blueshift is observed to analyze the refractive index. The spectral sensitivity so obtained from the results is equal to 12,071.17 nm/RIU at 1.2 µm wavelength. Dispersion profile so obtained is equivalent to zero from 0.4 to 1.8 µm wavelength. With the proposed sensor-based tri-core silica PCF, we are able to achieve high sensitivity toward sensing in various applications and concluded that the proposed structure is applicable to obtain better results in future. The proposed model has been studied for the glucose sensing application as the similar structure can show high resolution and can be recommended for the detection of different bio and organic chemical sensing. Findings With this proposed design, it is observed that such types of PCF designs can be helpful to society in terms of management and detection of stress, pain, and emotions in life. This will be a great field of research in the upcoming years.

References 1. Vigneswaran D, Ayyanar N, Sharma M, Sumathi M, Rajan M, Porsezian K (2018) Salinity sensor using photonic crystal fiber. Sens Actuators A Phys 269:22–28 2. Ayyanar N, Raja RVJ, Vigneswaran D, Lakshmi B, Sumathi M, Porsezian K (2017) Highly efficient compact temperature sensor using liquid infiltrated asymmetric dual elliptical core photonic crystal fiber. Opt Mater 64:574–582

3 Designing a Nonlinear Tri Core Photonic Crystal Fiber …

27

3. Zhao Y, Wu D, Lv RQ (2015) Magnetic field sensor based on photonic crystal fiber taper coated with ferrofluid. IEEE Photonics Technol Lett 27(1):26–29 4. Li W, Cheng H, Xia M, Yang K (2016) An experimental study of pH optical sensor using a section of no-core fiber. Sens Actuators A Phys 199:260–264 5. Otupiri R, Akowuah EK, Haxha S, Ademgil H, AbdelMalek F, Aggoun A (2014) A novel birefrigent photonic crystal fiber surface plasmon resonance biosensor. IEEE Photonics J 6(4):1–11 6. Kumar A, Gupta G, Malik A (2011) Single Mode Optical Fiber based Refractive Index Sensor using Etched Cladding. J Instrum Soc India 41(2):80–83 7. Turner APF, Fragkou V (2009) Commercial biosensors for diabetes. In: Handbook of optical sensing of glucose in biological fluids and tissues, 1st edn. CRC Press 8. Niraj GM, Varshney H, Pandey S, Singh S (2012) Sensors for diabetes: glucose biosensors by using different newer techniques: a review. Int J Ther Appl 6:28–37 9. Rifat A, Mahdiraji G, Chow D, Shee Y, Ahmed R, Adilkhan F (2015) Photonic crystal fiberbased surface plasmon resonance sensor with selective analyte channels and graphene-silver deposited core. Sensors 15(5):11499–11510 10. Baten MA, Seal L, Lisa KS (2015) Salinity intrusion in interior coast of Bangladesh: challenges to agriculture in south-central coastal zone. Am J Clim Change 4:248–262

Chapter 4

Hiding an Image Using Multi-object Vector Steganography and Variable Length Mixed Key Cryptography Manju Kumari Gupta and Vipra Bohara

1 Introduction Steganography is a method, which is used to hide the secret data into the cover medium. This provides a hidden communication channel between the sender party and the receiver party. The primary aim of the steganography method is to provide security for the transmission of hidden data [1]. Embedding efficiency, robustness, and hiding capacity are the three main requirements for a successful steganographic method. Cryptography is a method which is used to achieve security by encrypting messages to make them unreadable form. This means that it is used to protect user data. No one can identify the actual information. Cryptography is basically used to hide the original data into the coded data so as to prevent unauthorized access [2]. The steganography method hides the secret data into the cover medium without changing in the secret data; whereas, the cryptography method hides the secret data by changing the secret data into ciphertext. It encrypts the secret message such that it becomes useless to eavesdroppers. In this paper, we are using video steganography. Video is a collection of frames, and frame is represented as an image. Any video steganography method contains two algorithms, one for embedding and another one for extracting. Video file formats are MPEG, MP4, and AVI. The video contains a set of frames that are played back at a fixed frames rate [3].

M. K. Gupta (B) · V. Bohara Yagyavalkya Institute of Technology, Jaipur, India e-mail: [email protected]

© Springer Nature Singapore Pte Ltd. 2020 G. Mathur et al. (eds.), International Conference on Artificial Intelligence: Advances and Applications 2019, Algorithms for Intelligent Systems, https://doi.org/10.1007/978-981-15-1059-5_4

29

30

M. K. Gupta and V. Bohara

2 Literature Survey Some related methods are mentioned in this section. In [4], the author has described a parallel LSB algorithm for hiding a secret image in a 24-bit color image. Parallelism is achieved using an OpenCL parallel programming technique. The header is used as a key to extract the secret image. In [5], the “Logical Stratified Steganography Technique” is used for hiding a gray image into the cover image. OR operation is performed to combine the cover and hidden image. In this, the 4LSB of each pixel in the cover image is replaced by the 4MSB of each pixel in the secret image. In [6], video steganography is used based on the principle of linear block code. First frames are separated into YUV format. The secret message is encoded by the Hamming code (7, 4) algorithm. By using a private key, the pixel’s positions of both cover video and a secret message are randomly reordered. In [7], author has hidden R, G, and B secret image into other R, G, and B cover image by separating both the images in three planes (red, green, and blue). For hiding, DCT technique is used. Here, alpha factor value is used for encryption. In the extraction phase, the cover image is not required to obtain a secret image back. The alpha value is used to extract the secret image. Here, alpha value is worked as a key. In [8], Daubechies wavelet, LSB, and pseudo random number techniques are used for hiding miscellaneous data (one color image, one text file, and three grayscale images) in a single cover image. In [9], the author has used the DWT algorithm to hide an image in a selected frame at specific locations with the help of the LSB algorithm by replacing the last bit of pixel value of that specific location with the secret data. With the help of high-pass filter and low-pass filter, wavelet transformation is done.

3 Methodology In our work, for hiding an image into a video, we have used two algorithms; one is multi-object vector steganography and the second one is VLMKC. VLMKC algorithm is used to encrypt the object after the steganography operation is performed. Multi-object vector steganography algorithm is used to hide multi images into a video. But in this paper, we have used a single image for hiding. Here, the cover video frame size is greater than the secret image so the secret image takes less space in the cover video frame and the remaining space is free.

3.1 Variable Length Mixed Key Cryptography This cryptography method is a type of mixed key cryptography. This method employs a private key of variable length. Here firstly, the private key is generated by using dual non-divisible seeds (prime numbers). For providing more randomness, some

4 Hiding an Image Using Multi-object Vector Steganography …

31

random mathematical operations have performed on both seeds. The length of the private key is determined by the object to be encrypted. The generated key is bit based. Each bit is repeatedly computed, and eight such bits are combined to form a byte of the private key. Now, byte by byte XORing operation is performed between the corresponded byte of an object to be encrypted and the corresponded byte of the private key. The XORing operation involved is bitwise XOR thus providing eight times randomness in private key.

3.2 Multi-object Vector Steganography In our proposed approach, secret image bits are embedded randomly into the cover video pixels instead of sequentially. By using this algorithm, we can hide multi images in a single video frame. So, this method requires the hidden content to be substantially smaller in the number of bytes (or size) as compared to the cover video frame. In simple LSB algorithm, it is possible for the hacker to retrieve the hidden data due to the simplicity of technique; therefore, unwanted people can easily try to extract the message if they are suspicious that cover data contain some important information. So to overcome the sequence mapping problem generated by LSB, multi-object vector steganography is used. This algorithm employs linearization of the cover frame into a 1-D vector, containing the total number of bytes in the cover frame. The images or content to be hidden is also converted into a 1-D linear vector, which means all bits in R, G, and B plane of cover frame and the secret image are converted into 1-D array. After linearization of both images, the logarithmic calculation is used to generate random pixel positions in the cover image for hiding secret image without any repetition. Two least significant bits of random pixels in cover image are used to hide bits of secret image. The percentage of space contained by a secret image can be displayed. If percentage > 100, this prints “too many images or too big image for hiding.” After hiding, steganography image is converted into its original 3-D matrix size or original size. The performance of the proposed method is calculated by PSNR and MSE. Embedding phase Step1: Take the video as a cover object. Step2: Convert video into frames. Step3: Select a frame randomly. Step4: Hide secret image into the selected frame using multi-object vector steganography. Step5: After steganography, frame is replaced with other frames. Step6: All frames are encrypted using the variable length mixed key cryptography and private key. Step7: Combine all encrypted frames for making encrypted video. This encrypted video is transmitted by the sender.

32

M. K. Gupta and V. Bohara

Extracting phase Step1: Take the encrypted video as an input. Step2: Encrypted video converted into encrypted frames. Step3: All frames are decrypted using variable length mixed key de-cryptography and private key. Step4: Select same steganographed random frame which we have used in the embedding phase. Step5: Using multi-object vector de-steganography algorithm, secret image extracted from steganographed frame. Step6: Selected random frame replaced with the other frames. Step7: Original video generated.

4 Results Tables 1and 2 represent test images specifications and test videos specifications. Table 3 represented experimental results of video steganography. In this, we have operated different types of images like binary logo image, grayscale image, and RGB image with different videos. We get the same reconstructed image as the hidden image Table 1 Test images specifications S. No.

Image Name

Image size (Kb)

Image resolution

1

Img1

17.4

240*210

2

Kid

135

218*314

3

Binary Logo

29.05

400*400

4

Leena Grayscale

13.5

256*256

Image

4 Hiding an Image Using Multi-object Vector Steganography …

33

Table 2 Test videos specifications S. No.

Name of video

Size of video

Resolution

Duration (sec)

Frame rate/sec

1

Rhinos

25 mb

320*240

07

15

2

Viptrain

10 mb

360*240

20

30

3

News

213 kb

352*288

09

29

Table 3 Experimental results S.no

Cover video

PSNR(db)

MSE

1

Rhinos

Hidden image

Reconstructed image

55.9704

0.1645

2

Viptrain

54.8085

0.2149

3

News

52.1465

0.3967

4

News

55.7903

0.1701

by selecting the same frame number at the decoding side as we have selected at the encoding side. We have calculated PSNR and MSE values in each image. PSNR value of 55.9704 and MSE value of.1645 are the best results for the cover video “Rhino” and the hidden image “img1.” By using this algorithm, we get an image without degradation after performing steganography operation. By observing the table, we can also analyze that the binary image has a low value of PSNR and a high value of MSE which is not good. Table 4 presents a comparison between the proposed method and some other video steganographic methods. We can analyze that our proposed method has the best PSNR value, so we can say that our proposed method is more secure than other steganographic methods.

34

M. K. Gupta and V. Bohara

Table 4 Table of comparison with other video steganography papers S. No.

Title

Video used (.avi)

Secret message

PSNR (db)

1

Zhang et al. [10]

News

Leena grayscale

36.743

Proposed method

55.7903

2

Xu et al. [11]

3

Banerjee et al. [12]

News

Binary logo

40.75 37.82

4

Cao et al. [13]

40.68

Proposed method

52.1465

5 Conclusion In this paper, secure video steganography has been proposed based on multi-object vector steganography and VLMKC. Here, video is converted into frames, and the frame is randomly selected by user or system for hiding images. Video is encrypted after hiding image into video thus increasing security, only authorized person having a valid private key and knowledge of random frames used as a cover object, that person can successfully decode the hidden image. The multi-object vector steganography employed embedded to be hidden in LSB to MSB of a cover video frame. This algorithm provides high embedding efficiency. The visual quality is measured by MSE, PSNR and results in having a PSNR value above 51db, which is very good.

References 1. Thakur V, Saikia M (2013) Hiding secret image in video. In: International conference on Intelligent Systems and Signal Processing (ISSP), IEEE 2. Zaheer R, Gaur RS, Dixit V (2017) A literature survey on various approaches of data hiding in images. In: International Conference on Innovations in information Embedded and Communication Systems (ICIIECS), IEEE 3. Divya KP, Mahesh K (2014) Random image embedded in videos using LSB insertion algorithm. Int J Eng Trends Technol (IJETT) 13(8):381–385 4. Kini NG, Kini GV (2019) A parallel algorithm to hide an image in an image for secured steganography. Integrated intelligent computing, communication and security. Springer, Singapore, pp 585–594 5. Mani MR, Eswari GS (2018) A novel image hiding technique based on Stratified steganography. Smart computing and informatics. Springer, Singapore, pp 45–52 6. Mstafa RJ, Elleithy KM (2014) A highly secure video steganography using hamming code (7, 4). In: IEEE LISAT 2014 Long Island Systems, Applications and Technology 7. Ghosh E, Debnath D, Banik BG (2019) Blind RGB image steganography using discrete cosine transformation. Springer, Singapore 8. Viraktamath SV, Kinagi B, Kumar K, Pavan MS, Hunagund V (2018) Performance analysis of steganography for hiding miscellaneous data using Daubechies wavelet, IEEE 9. Reddy MA, Reddy PS (2013) DWT and LSB algorithm based image hiding in a video. Int J Eng Sci Adv Technol 3(4):170–175 10. Zhang Y, Zhang M, Wang XA, Niu K, Liu J (2016) A novel video steganography algorithm based on trailing coefficients for H.264/AVC. Informatica 40:63–70

4 Hiding an Image Using Multi-object Vector Steganography …

35

11. Xu D, Wang R, Shi YQ (2014) Data hiding in encrypted H.264/AVC video streams by codeword substitution. IEEE Trans Inf Forens Secur 9(4):596–606 12. Banerjee A, Jana B (2018) A secure high capacity video steganography using bit-plane slicing through (7, 4) hamming code. Advanced computational and communication paradigms. Springer, Singapore, pp 85–98 13. Cao Z, Yin Z, Hu H, Gao X, Wang L (2016) High capacity data hiding scheme based on (7, 4) hamming code. Springer Plus 5(1):175

Chapter 5

Customer Attrition Estimation Modelling Based on Predominant Attributes Using Multi-layered Feed-Forward Neural Network Vaishnavi Sidhamshettiwar, Yash Gaba, Rutika Jadhav and Kiran Gawande

1 Introduction As businesses are evolving in the digital era, it becomes highly important for an organization to retain its existing customers. Customer churn prediction refers to identifying customers who are at the risk of exiting the services of the business and taking effective measures to get them involved in the business again. However, data available for statistical calculations is unstructured and very large in volume. Hence, sophisticated machine learning techniques have to be deployed for predicting accurate results. Our work has a twofold objective: firstly, to predict the probability that a particular customer will leave the firm and secondly to identify the reason for which the customer will exit the organization. A system with the following features has been implemented: – User-friendly interface: Easily interpretable visual data. – Predict probability of a current customer to leave (ranging from 0 to 1): Where 0 will mean that client will never leave the firm and 1 will indicate that client is unquestionably going to leave the firm.

V. Sidhamshettiwar (B) · Y. Gaba · R. Jadhav · K. Gawande Department of Computer Engineering, Sardar Patel Institute of Technology, Mumbai, India e-mail: [email protected] Y. Gaba e-mail: [email protected] R. Jadhav e-mail: [email protected] K. Gawande e-mail: [email protected] © Springer Nature Singapore Pte Ltd. 2020 G. Mathur et al. (eds.), International Conference on Artificial Intelligence: Advances and Applications 2019, Algorithms for Intelligent Systems, https://doi.org/10.1007/978-981-15-1059-5_5

37

38

V. Sidhamshettiwar et al.

– Provide quick on-demand data of top percentage of people likely to leave: Empowers client to streamline their emphasis on required number of individuals. Anyway, foreseeing for any client from the information is dependably an alternative. – Show specific details: Expectation of likelihood of a client leaving the firm alongside its diagram. Additionally show likelihood of various reasons of leaving for every client and show simple to translate pie outlines exhibiting each explanation behind inside and out examination. This paper is organized as follows. Section 2 summaries the survey on related works on customer churn. Section 3 states the proposed methodology. Section 4 summarizes the data preprocessing and cleaning phase. Section 5 summarizes the architecture of neural networks been implemented. Section 6 specifies the findings of research. Sections 7 and 8 detail the conclusion and future scope of our project. The last section states all references for our research.

2 Survey on Related Works on Customer Churn 2.1 A Comparison of Machine Learning Techniques for Customer Churn Prediction (ScienceDirect, December 2015) This study carried out by Vafeiadis [1] performed 100 Monte Carlo simulations for five state of the art machine learning techniques and accuracy, and F measure for each of the techniques was recorded. The methods presented in the research work are artificial neural networks, support vector machine, decision tree learning, Naive Bayes classifier and logistic regression analysis. Evaluation of these methods was based on four major parameters: accuracy, F measure, precision and recall. In the first stage, all the models were trained and evaluated using a publicly available data set consisting of 18 predictors and 5000 data samples. In the second phase, performance was improved by boosting algorithms. It is observed that boosted versions of models give better performance than non-boosted versions. ANN gives an accuracy of 97% with an F measure of 84%. Performance of ANN can be improved by implementing classical hybrid models.

2.2 Customer Churn Prediction by Hybrid Neural Networks (ScienceDirect, May 2009) This study carried out by Tsai [2] shed light on hybrid models for predicting churn attrition rate. Many supervised and unsupervised machine learning techniques can

5 Customer Attrition Estimation Modelling Based …

39

be combined with ANNs to improve its accuracy. The two methods demonstrated in this research work are as follows: ANN plus ANN: The first neural network model performs data preprocessing, and the second neural network predicts customer churn. Since both are supervised machine learning techniques, correctly predicted data by the first network is used to train the second model. ANN plus Self Organizing Maps (SOMs): SOMs are an unsupervised machine learning technique which uses clustering for data reduction. ANN can then be used to predict churn rate on the reduced data set. Fuzzy test data sets (FTDs) have been used to evaluate the performance of both the hybrid models. It is interesting to note that ANNs combined with SOM (accuracy 80%) do not perform better than classical ANN clubbed with ANN (accuracy 90%).

2.3 Customer Churn Prediction Global Telephony and Service Provider (Deloitte, November 2015) This white paper published by Deloitte [3] studies a case of a leading telecom service provider wanting to identify customers who are at the risk of not renewing their contracts. Deloitte helped the telecom provider to build a predictive analysis model using ANN and also devise strategies to influence clients to renew their contracts. Velocity, variety and complex customer demographics are the major challenges faced by global telecom providers. Deloitte sourced available data of customers like geographic location, usage data, product information and reseller/distributor data. They have identified parameters like contract duration, discount percentage, prior client engagement and helped the client devise a visual model to identify which contracts are likely to be renewed. This saved the client millions of dollars which were to be spent in acquisition of new customer.

3 Proposed Methodology 3.1 Overview The overall process works on records provided by the client using this system. Raw data is initially cleaned and further analysed to normalize the data entries using preprocessing algorithms. Hyper parameters are then selected and further tuned for building the model (Fig. 1). The proposed information is fed into the neural network model. The model is prepared on the preparation informational collection, checked by utilizing approval information and tried on the testing information. This provides the benchmark to evaluate the system. To infer the predominant parameters that mainly cause attrition,

40

V. Sidhamshettiwar et al.

Fig. 1 Suggested procedure flow diagram

a correlation graph will be generated. The whole engineering is demonstrated in six phases every one of which has been examined below: Data preprocessing – User Information acquisition: Client will be providing with the user details about the various services subscribed by the user. Performance metrics are decided here. – Explaining Churn: Specifying parameters which will assist in classifying churners and non-churners. No numeric parameters like happiness, satisfaction of the customer and numeric parameters like number of products, tenure completed are included in this. – Data Labelling: Data items will have to be tagged as exited or not exited based on weightage of various factors involved. Data modelling – Feature Engineering: The raw data will be passed onto algorithms, viz. One Bit Encoding, Label Encoding and Standard Scaling for extracting and standardization important information. – Training: The data is segregated into training and testing data. Classifier is trained using the training data, and performance is evaluated on the testing data. – Prediction: In this stage, we will pass the user data to the model and predict whether the user will be churned or not.

5 Customer Attrition Estimation Modelling Based …

41

3.2 Data Collection Customer churn prediction in banking industry requires lot of information about the past behaviour of the customers so that it can predict their behaviour in near future. The data set used in this research work is provided by DHFL retail banking customer information [4]. The data set consists of 10,000 entries, and each entry is described by 13 attributes. In this data set, Exited attribute is the class label. Other attributes are classified as follows: – – – –

Demographic data: CustomerID, Gender, Age, Geography Subscriptions Availed: Numofprods, hascrCard, IsActiveMember User Profile: EstimatedSalary, Balance, CreditScore Period of customer engagement with the company: Tenure.

One more attribute is described for each entry, i.e. reason for exiting company. This attribute will help us to retain the customers by providing them with the services which were the reason for leaving their company. Target attribute in this case is Exited which will tell that whether customer is going to churn or not. All other values of the attributes will help in predicting the target attribute.

4 Data Preprocessing and Cleaning To use the raw data for any application, we need to clean the data. There are many algorithms to preprocess the data and make it usable for neural network. Initially, we will remove all those attributes which are totally insignificant for the training of the model. In this data set, we have removed RowNumber and CustomerID which are just unique identifiers and Surname attribute as they were of no significance to the model. Another problem with the raw data set was that there are a lot of textual attribute values. Label Encoding [5] is implemented to convert textual fields to numbered entities. This is deployed for a number of columns where fields are of yes or no form. However, the disadvantage of this method is that it presumes that sets of data with higher numeric size are better. This renders the usage of multiclass attributes ineffective. Another major algorithm used was One Hot Categorical Encoding [6] which enables the information to be more detailed by given binary representation to indexed columns. This algorithm also helps to avoid allocating any higher bias to a given attribute (Fig. 2). Since the scope of values of raw data varies widely, in machine learning techniques, objective functions will not work properly without standardization. In this, we are using Standard Scaling method for normalization. Thus, it normalizes data such that its distribution will have a mean value 0 and standard deviation of 1.

42

V. Sidhamshettiwar et al.

Fig. 2 Categorical encoding of data

5 Neural Network Architecture The implemented model is based on artificial feed-forward neural networks. The model is sequential model, i.e. it consists of layers which are seen as a straight pile of neural layers. The model comprises of dense layers which are fully connected layers, so all the neurons in a layer are connected to those in the next layer. The model consists of three layers that are (1). Input layer which takes all the entries of the customers with 11 attributes and passes it on to the next layer, (2). Hidden layer which processes the output of input layer and passes it on to the next layer, (3). Output layer which gives us the final output whether the customer is going to churn or not churn. Input layer comprises of 11 neurons, hidden layer comprises of six neurons, and output layer comprises one neuron (Fig. 3). The activation function used in this model is ReLU and sigmoid. Rectifier is defined as the positive part of its argument: f (x0 ) = x + = max(0, x)

Fig. 3 Neural network architecture

5 Customer Attrition Estimation Modelling Based …

43

where x is the input to a neuron. A unit employing the rectifier is called a rectified linear unit (ReLU). Sigmoid function curve looks like S-shape. As in our application, we are going to predict the customer churn probability whose values exist from 0 to 1. Thus, we are using sigmoid activation function because it is used for models where we have to predict the probability as an output. Sigmoid function is defined as Sθ (x) =

1 1 + e− θ T x

In this model, we are using Adam optimizer which is an adaptive learning rate method, which means, it computes individual learning rates for different parameters. The model was trained for 150 epochs, and the batch size for one epoch is two.

6 Results and Discussion 6.1 Accuracy Result From the above technique, we have processed the data and tested it on the approval data set. An accuracy of 89.5% is attained which is a significant rise compared to our literature survey. The confusion matrix shows the percentage of values which have been predicted correctly (Fig. 4). Accuracy = (TP + TN)/2 = (0.93 + 0.86)/2 = 0.895

6.2 Correlational Analysis To infer predominant parameters having an impact on prediction, we deploy the technique of correlation [7]. Correlation displays the effect of one set of entities in presence of another set of entities, which in this case would represent dependency of attributes with the output labels. The following plot shows the correlation coefficient of all the attributes plotted against the output label (Fig. 5). A positive correlation indicates that the parameter is directly related to churning, and a negative correlation indicates that the parameter is inversely related to churning. From the above graph, it is clear that parameters which affect churn the most are Number of products, Balance, Tenure and EstimatedSalary. Age prevents churn as in the banking industry, people tend to settle in with their regular banks once age increases. Other parameters like HasCreditCard and CreditScore play a negligible role in calculating churn.

44

Fig. 4 Confusion matrix

Fig. 5 Correlation graph

V. Sidhamshettiwar et al.

5 Customer Attrition Estimation Modelling Based …

45

Fig. 6 Graphical representation

6.3 Graphical Analysis To evaluate the model and identify major parameters, various graphical analyses have been performed on the mode. There are four numeric parameters: Tenure, Balance, EstimatedSalary and NumOfProducts. The following graphs show a MATLAB (Fig. 6) plot of these four parameters against churn (1: Exited, 0: Non-Exited). Tenure indicates faithfulness of a customer. As the tenure increases, the number of customers being churned tends to decrease. Number of Products shows the extent to which a user engages with the business. Greater the number of products, higher the probability that a customer will stay loyal to the business. In the banking sector, bank balance and estimated salary show the purchasing power of the customer; hence, for higher values of these attributes, customer churn will tend to decrease.

7 Conclusion The implemented multi-layered neural network model achieved an accuracy of 89.5% for churn classification. The model also identified major parameters that affect churn rate directly or inversely. This model also assists organizations to identify reasons

46

V. Sidhamshettiwar et al.

for exiting of customers and devise strategies to retain them. As it is statistically observed, the cost for acquiring a new customer is five times more than retaining existing customers.

8 Future Scope Retention policies can be designed by extending association between churn rate and lifetime value of a customer. This is necessary for a firm as these policies must be adopted for each customer individually and will increase revenues by manifolds in future.

References 1. Vafeiadis T, Diamantaras KI, Sarigiannidis G, Chatzisavvas KC (2015) A comparison of machine learning techniques for customer churn prediction. Department of Information Technology, TEI of Thessaloniki, GR-57400 Thessaloniki, Greece 2. Tsai CF, Lu YH (2009) Customer churn prediction by hybrid neural networks. Department of Information Management, National Central University, Taiwan 3. Deloitte (2015) Opportunities in telecom sector: arising from big data. Aegis School of Business, Data Science and Telecommunication 4. https://tinyurl.com/y3oadn9v 5. Li C, Kang Q, Ge G, Song Q, Lu Q, Cheng J (2016) Deep BE: learning deep binary encoding for multi-label classification. In: Proceedings of the IEEE conference on computer vision and pattern recognition workshops, pp 39–46 6. Choong ACH, Lee NK (2017) Evaluation of convolutionary neural networks modeling of dna sequences using ordinal versus one hot encoding method. In: International conference on computer and drone applications, pp 60–65 7. Andrew G, Arora R, Bilmes J, Livescu K (2013) Deep canonical correlation analysis. In: International conference on machine learning

Chapter 6

A New Methodology to Implement ASCII-Based Cryptography Shubha Agarwal

1 Introduction In today’s era, it is impossible to think of any work without Internet. In this digital world, it is very much necessary to have confidentiality while interchanging data to and fro. Cryptography is only a practice that provides such a level of security mechanism while any kind of transaction through a world [1]. Cryptography is s technique of science that provides secure communication between two devices. It is a study of making non-readable information with the goal that only intended individual can read that ciphered text (coded text). Cryptography basically includes encryption and decryption techniques in which the change of plain text into non-readable text is depicted as encryption and reverse of this encryption process is termed as decryption [2]. Cryptography is mechanism to secure data with secrete pattern from third party. These third-party people are attackers to steal private data or message. Security is provided by applying some mathematical transformation, some optimal algorithm and also with addition of one public or private key [3]. Basically, the flow of cryptography is like originator will encrypt plain text by apply locking with key and recipient will unlock that cipher text with same key to decrypt the message [2]. This cryptographic key is like a password to encrypt or decrypt any message. There are two types of cryptographic keys: Symmetric key and Asymmetric key. Besides this locking mechanism, some security algorithm is also applied at the duration of encryption and decryption processes. Symmetric Key is termed as secret key also, as only one key is used for coding and decoding the text. The symmetric key encryption algorithm is the simplest technique in which single key is shared among parties who need to receive message. Sender S. Agarwal (B) Department of Computer Science, Jodhpur Institute of Engineering and Technology, NH-65, Pali Road, Mogra, Jodhpur, Rajasthan, India e-mail: [email protected] © Springer Nature Singapore Pte Ltd. 2020 G. Mathur et al. (eds.), International Conference on Artificial Intelligence: Advances and Applications 2019, Algorithms for Intelligent Systems, https://doi.org/10.1007/978-981-15-1059-5_6

47

48

S. Agarwal

uses key before sending the encrypt message and receiver use same key to decrypt that message. Some approaches that are based on symmetric encryption are RC2, RC4, RC5, AES, DES, 3DES and Blowfish [4]. Asymmetric Key is termed as public key also, as one public key is available for everyone who wishes to send message. The asymmetric key encryption algorithm is the complex technique in which pair of public and private keys are used because public key is shared among all senders to encrypt the message and private key is used to decrypt the message. Private Key is kept at secured place by owner of public key. This technique is comparably slow because it uses two keys to encrypt and decrypt data. Some approaches that are based on asymmetric encryption are RSA, DSA, Diffie–Hellman and ECC [4].

2 Related Work The various studies on cryptography techniques including substitution and transposition are done by different researchers. Some research papers had also helped us in study of cryptography with different type of mechanism. Some existing techniques are discussed in this paper with their drawbacks.

2.1 The Substitution Cipher This technique is simplest one in which each letter is replaced with another one. The substitution table is purely random so due to symmetric encryption, this table is distributed to all receivers for deciphering the message. The major drawback of this technique is that highest possibility of mapping of one letter to another one is equal to the maximum number of size of alphabets that is 26. Means attackers need to have only 26 times for deciphering the message. One other trick for attackers to decipher message is by analyzing frequent letters that are substituted with same latter [5]. For example A→K B→G C→E

2.2 The Reverse Cipher This technique uses a pattern of reversing the original string to convert into cipher text. Attackers can easily break cipher text so this technique is very bad option in encryption.

6 A New Methodology to Implement ASCII-Based Cryptography

49

For example Plain Text ABCD Cipher Text DCBA

2.3 The Caesar Cipher It is a special case of above technique in which every letter is shifted with some fixed number of position in alphabet. This shifting is done in loop of 26 alphabets. The drawback is that attackers have to do match only 26 times of every alphabet to match the original text [5]. For example if key is 4 A→E B→F ...... Y→C Z→D

2.4 The ROT13 Cipher It is a special case of Caesar cipher in which every letter of alphabet is shifted by position 13 in round robin manner. This technique can also easily broken by hackers by reverse shifting by same number. For example A → N B→O C→P

2.5 The Affine Cipher This is another technique based on substitution in which original message is encrypted by key with pair of value. One part of this key pair is multiplied with letter and other part is added. This technique works on alphabets only so it is also like a ring of 26 letters. By frequent letters and substitution, attackers can easily find out the deciphered text [5].

50

S. Agarwal For example Plain Text ABC (0, 1, 2 that are position in alphabets) Key (2, 4) Cipher Text CEG (4, 6, 8) (0 ∗ 2 + 4 = 4, 1 ∗ 2 + 4 =, 2 ∗ 2 + 4 = 8)

2.6 The One-Time Pad Cipher This technique is powerful compared to above all techniques. This method uses the same length of key to the original message to encrypt text. Key is a collection of some random letters that are modulo with original message. This technique is secure but so impractical because of key length that must be equal to plain text. For example Plain Text THIS Key XVHE Cipher Text QCPW

3 Proposed Work In this approach, to encrypt message, key is generated by algorithm itself instead of decided by owner like above all approaches. Proposed method uses three levels to encrypt any text which is hard to be breakable by attackers. Working of encryption and decryption algorithms is shown below with one example in step by step.

3.1 Proposed Algorithm for Encryption Step 1 Enter the text in a file Plain Text: E N C R Y P T Step 2 Convert the text into ASCII form ASCII Content: [69, 78, 67, 82, 89, 80, 84] Step 3 Find the maximum ASCII value and take mod of that value with 165 Key generation: mod (Maximum, 126) = 37 Step 4 Substitution Level (Key addition to every ASCII value in a list) New ASCII Content: [106, 115, 104, 119, 126, 117, 121] Step 5 Transposition Level 1 (Reverse the list) Transposition Level 2 (Interchange every two element in a list) Cypher Text: u y w ~ s h j.

6 A New Methodology to Implement ASCII-Based Cryptography

51

Fig. 1 Steps in the encryption algorithm

3.2 Proposed Algorithm for Decryption Step 1 Convert the text into ASCII form ASCII Content: [117, 121, 119, 126, 115, 104, 106] Step 2 Substitution Level (Key subtraction to every ASCII value in a list) New ASCII Content: [80, 84, 82, 89, 78, 67, 69] Step 3 Transposition Level 1 (Interchange every two element in a list) Transposition Level 2 (Reverse the list) Decipher Text: E N C R Y P T (Figs. 1 and 2).

4 Implementation and Results Implementation of proposed technique is done in Python (2.7 version) language. Figures 3 and 4 are the screenshots of ciphered and deciphered text after implementing proposed algorithm. Figure 5 depicting a graph between encryption times that vary with size of plain text. Encryption time is increasing at initial stage but as size of plain text is increasing, time taken in encryption is constant. So if we compare with existing other encryption

52

S. Agarwal

Fig. 2 Steps in the decryption algorithm

Fig. 3 Encryption of plain text using proposed algorithm

Fig. 4 Decryption of cipher text using proposed algorithm

methods, other than this proposed method, in all techniques, encryption time is linearly increases as size of plain text increase.

6 A New Methodology to Implement ASCII-Based Cryptography

53

Fig. 5 Encryption time of different size of plain text

5 Conclusion In this paper, one new approach for cryptography is described which is little different from comparable approaches like Caesar, reverse, affine ciphers. Proposed technique has major point is its key generation level that is automatically done by algorithm itself. Every technique has some pros and cons, and so, one drawback of this method is that attacker can break security only by frequent pattern analysis

References 1. Patni P (2013) A poly alphabetic approach to caesar cipher algorithm. Int J Comput Sci Inf Technol (IJCSIT) 4(6):954–959 2. Gupta SS, Dhanjal G, Bambardeka S, Vartak P (2018) Data security using compressed classical technique. In: International Conference on Smart City and Emerging Technology (ICSCET). https://doi.org/10.1109/icscet.2018.8537387 3. Raghu ME, Ravishankar KC (2015) Application of classical encryption techniques of securing data, a threaded approach. Int J Cybern Inf (IJCI) 4(2):125–132 4. A fundamental difference between symmetric and asymmetric encryption. https://www. rapidsslonline.com/blog/fundamental-differences-between-symmetric-and-asymmetricencryption. Last accessed 28 March 2019 5. Paar C, Pelz J (2010) Understanding cryptogrpahy. Springer, Berlin

Chapter 7

Sentiment Analysis of Product Reviews of Ecommerce Websites Shubhojit Sarkar and Souparna Palit

1 Introduction The world is ever-evolving and in the era of the Internet, people are changing, their ideas, businesses are evolving. Nearly all Business Houses this day have an online portal. As the digital world grows, almost everybody is now connected to the internet and so are a number of business houses. These e-commerce websites have catered to countless people in the past few years and that number is exponentially increasing. With the increasing number of online e-commerce sites and the products, they are offering, the dilemma of the consumer in which product to choose is increasing as well. On top of that the consumer has to choose the product just by judging its picture. Pictures may be deceiving in many cases, therefore, the customer may require a ‘word of mouth’ from anybody who has purchased and used the product in the past. Some websites like Amazon were the first to introduce the option of rating and reviewing a product by authentic customers. But then a new problem arises, with the increase in number of purchases, the number of product reviews and ratings too increases and as a result of which judging whether a product is useful or not becomes cumbersome for the customer. This led to the advent of Sentiment Analysis. The objective of this research is to provide a solution to the problem of sentiment analysis [1] by preprocessing the reviews and ratings of customers using a scent of Bidirectional Associative Memory and classifying them using Naïve Bayes, Support Vector Machine and Decision Tree.

S. Sarkar (B) · S. Palit Department of Information Technology, St. Thomas’ College of Engineering and Technology, Kolkata, West Bengal 700023, India e-mail: [email protected] © Springer Nature Singapore Pte Ltd. 2020 G. Mathur et al. (eds.), International Conference on Artificial Intelligence: Advances and Applications 2019, Algorithms for Intelligent Systems, https://doi.org/10.1007/978-981-15-1059-5_7

55

56

S. Sarkar and S. Palit

2 Previous Works In this study, we will be processing data using a scent of Bidirectional Associative Memory [2]. The processed data will be classified using different classifiers like Support Vector Machine, Naïve Bayes and Decision tree. Support Vector Machines are used for learning text classifiers from examples. They study the particular characteristics of learning with the text and picks out why SVMs are best suited for this mission. Experimental results support the already known theory. They achieve considerable advancement over the top-performing methods and behave vigorously over a range of diverse learning tasks. Moreover, they are completely automatic thereby eliminating any needs of physical constraint regulation [3]. The Naive Bayes classifier significantly makes it easier to learn by assuming that features are independent in a given class. Naïve Bayes is a type of classifier that uses independence of the features as an assumption. Although this might be considered as a disadvantage, still it can outperform many complex classifiers given the proper dataset [4]. Decision Tree is a non-linear classifier that is widely considered to be an optimum choice for classification problems such as character recognition. It is considered a powerful classifying tool because of its technique of decomposing complex problems in decision-making into small and simple parts. This breaking down ensures that it can be understood easily [5]. We have also gone through some of the works of this field where the preprocessed data was taken and machine learning algorithms were applied to classify the reviews that are good or bad. The study concluded that Machine Learning procedures gave the best results to classify the Products Reviews [6].

3 Proposed Methodology Sentiment Analysis or Opinion Mining is the analytics done on some datasets to derive the sentiments of the people. For an e-commerce website like Amazon, etc., this analysis is very important. One of the ways of doing this analysis is by taking the reviews given by the customers for the products that they buy. The approach given in this paper takes both the ratings as well as reviews of these customers into consideration for generating the sentiments that were evoked for purchasing that product. A review is basically a paragraph typed by some customers. Each of these paragraphs may have multiple sentences. A sentence can be categorized as good, bad or neutral based on the words that are being used in the sentence. The proposed approach preprocesses the reviews obtained from the dataset with the influence of Bidirectional Associative Memory technique. This preprocessed data then goes to the various classifiers being used to classify that review, thereby giving the final verdict

7 Sentiment Analysis of Product Reviews of Ecommerce Websites

57

of whether that review was positive, negative or neutral. During this preprocessing method, we also consider the ratings that were provided alongside the respective reviews [7]. So, the final data fed into the classifier is based on both the reviews as well as ratings. The adjectives are passed through sign function in MATLAB which changes their weights from bigger numbers to 0, 1, etc. For every review, after tokenizing, each adjective is passed through the string matching algorithm and matched with their respective weights from the adjective list. If matched, then the weights are added to the respective review. Finally, it returns to the main function and in the main function, all the reviews are being collectively worked on. Before classification, the weights from the ratings are added as mentioned earlier. The classifiers that are being used for the classification of the sentiments are Naïve Bayes Classifier, Decision Tree and Support Vector Machine (SVM). Each of these classifiers has their own techniques and algorithms that they follow. In our approach, we have used them to draw up a comparative analysis between these classifiers, while being used on unprocessed data and preprocessed data. This shows that the effects and significance of the requirement of preprocessing of any data before it can be used in any sort of experiment. Many researchers have done work in this field like M. Chen, Y. Sun. However, our approach seemed to give better results than the previously mentioned research works. A systematic approach has been taken in order to perform this experiment. Each step that was taken during the experiment led to the occurrence of the next one. Figure 1 demonstrates the suggested approach that was taken.

Fig. 1 Proposed methodology of sentiment analysis

58

S. Sarkar and S. Palit

3.1 Source and Type of Dataset Used The dataset used for this experiment was taken from the Consumer Reviews of Amazon products. The dataset is available in the following link: http://jmcauley.ucsd. edu/data/amazon/. The products considered for this dataset are the various musical instruments and their parts. The dataset consisted of reviewer id, asin, reviewer name, helpful, review text, overall, summary, Unix review and review time. This dataset was in .json file format. So, the file was first converted into an excel file to extract and utilize the data for the experiment. The file consisted of 1000 reviews and ratings given by the customers. All of the data was used in the processing stage. From this excel file, we extracted the reviews and ratings as we required only these two for the project. An example of such data is as follows: Review: ‘This pop filter is great. It looks and performs like a studio filter. If you’re recording vocals this will eliminate the pops that gets recorded when you sing.’ Rate: ‘5’ After extracting the reviews and ratings, we then extracted the set of adjectives dataset. This dataset contains all the known adjectives with 10 features already being provided with it. It contains 7062 adjectives containing the most popular positive, negative and neutral adjectives. After this we took each of the reviews and preprocessed them. A sample of adjective dataset is given below: Adjective: ‘admirable 2 3 3 3 4 3 2 2 2 2’ We have used WEKA tool to apply various Machine Learning tools on the dataset. As far as the supervised machine learning tools are concerned, the dataset can work with any type of classifiers like decision tree, LSTM, SVM, etc. When it comes to unsupervised Machine Learning tools, the data is required to form clusters before applying any clustering tool.

3.2 Data Preprocessing Preprocessing started with tokenizing the reviews by decomposing the entire paragraphs into words. This tokenizing is done by separating the paragraph with respect to the various whitespaces and punctuation marks. The words thus obtained are then used further calculation. Then we started an iteration process that iterated through all the adjectives present in our dataset. For each adjective, we searched its frequency against the set of tokenized words that we obtained from the previous step. For that particular review, we multiplied the obtained frequency of an adjective with that adjective’s weights. In this way, we obtained a 7062 × 10 matrix with the calculated weights. We then applied a part of BAM algorithm to obtain a set of 10 features for each of the reviews. The rates respective to the reviews were also added to this 10-feature dataset to make it a dataset of 11 features. This entire preprocessed was then used for sentiment classification.

7 Sentiment Analysis of Product Reviews of Ecommerce Websites

59

Fig. 2 Basic structure of BAM

3.3 Bidirectional Associative Memory Our preprocessing technique starts with using a type of hetero-associative memory known as Bidirectional Associative Memory (BAM). This is a type of content addressable memory, i.e. it stores some data into a matrix or table and then returns the address of the matching data when an input data is searched for using it. Figure 2 shows the basic structure of BAM. In our approach, we used BAM as a technique for calculating the set of features for each of the reviews. We used sign () function to calculate the weights of the adjectives. This ensures a ternary classification of the sentiments as well as follows the BAM algorithm for the input into the iterative process. After the iterative process, the result was added to get the sum of the features for every review.

3.4 Sentiment Classification After preprocessing stage, the reviews are classified using suitable classifiers. The classifiers used are Naïve Bayes, Support Vector Machine and Decision Tree. These classifiers helped us in classifying the reviews into positive, negative and neutral reviews. The classification was done by using WEKA [8] tool.

3.4.1

Naïve Bayes

Naïve Bayes Classifier is one of the simplest forms of classifier that follows a probabilistic approach. It is based on Bayes’ Theorem [9] that requires no dependence or very strong independence between the features.

60

S. Sarkar and S. Palit

posterior =

prior × likelihood evidence

(1)

3.5 Support Vector Machine Support Vector Machine is a type of supervised learning model that is mainly used for classification and regression problems [10]. This is a linear model (or non-linear model), after being trained with data of two or more classes, starts to create new samples for every category. This is quite in contrast with the abovementioned Naïve Bayes Classifier, thus becoming a non-probabilistic model.

3.5.1

Decision Tree

Decision tree is a non-linear structure that is used for the classification problem. It is a very powerful yet simple technique that is heavily used in data mining [11]. It consists of a root node, internal and external nodes and their connecting branches. It is used to classify the data sets based on the supplied features.

4 Experimental Results 4.1 Binary Classification Figure 3 shows the experimental results for binary classification. As is evident from the chart below, the Naïve Bayes classification gives the maximum error in the above experiment. The Support Vector Machine gives the least percentage of error but the Decision Tree classifier gives the fastest results.

4.2 Ternary Classification In this classification, as we can see from Fig. 4, the Naïve Bayes gives the lowest accuracy. The Support Vector Machine Classifier and the Decision Tree classifier both give the same accuracy. The Support Vector Machine and the Decision Tree both gives almost the same percentage of errors but the Decision Tree is faster.

7 Sentiment Analysis of Product Reviews of Ecommerce Websites

61

Fig. 3 Graphical representation of experimental results for binary classification

Fig. 4 Graphical representation of experimental results for ternary classification

4.3 Ternary Classification (Using Unprocessed Data) After preprocessing the data, both the Support Vector Machine and Decision Tree gives better accuracy and are faster as well. Among the Support Vector Machine and Decision Tree, the Decision tree gives the best result, as shown in Fig. 5.

62

S. Sarkar and S. Palit

Fig. 5 Graphical representation of experimental results for ternary classification (using preprocessed data)

5 Conclusion and Future Work Sentiment Analysis for the reviews of e-commerce websites is now becoming more and more popular due to the increasing demand. Almost everyone who has proper access to internet uses these websites for shopping or at least viewing the different varieties of products. To ease the work of both the customers as well as the Business House, Sentiment Analysis plays a big role. With our experiment, we provided a novel approach for sentiment classification in a way that it is influenced from BAM. We further proved that Decision Tree outperforms Naïve Bayes and SVM for ternary classification. For binary classification, Decision Tree stands out as the fastest classifier while SVM being the model with least error. For future work, further modifications can be done in the field of feature extraction by completely implementing BAM as well as other associative memory methods like Hopfield.

References 1. Liu B (2012) Sentiment analysis and opinion mining, synthesis lectures on human language technologies. Morgan & Claypool Publishers, San Rafael 2. Kosko B (1988) Bidirectional associative memories. IEEE Trans Syst Man Cybern 18(1):49–60 3. Joachims T (1998) Text categorization with support vector machines: learning with many relevant features. Lec Notes Comput Sci 137–142 4. Rish I (2001) An empirical study of the naive Bayes classifier. In: IJCAI workshop on empirical methods in AI 5. Safavian SR, Landgrebe D (1991) A survey of decision tree classifier methodology. IEEE Trans Syst Man Cybern 21(3):660–674

7 Sentiment Analysis of Product Reviews of Ecommerce Websites

63

6. Jagdale RS, Shirsat VS, Deshmukh SN (2018) Sentiment analysis on product reviews using machine learning techniques. Adv Intell Syst Comput 639–647 7. Anshuman, Rao S, Kakkar M (2017) A rating approach based on sentiment analysis. In: 7th international conference on cloud computing, data science & engineering—confluence, pp 557–562 8. Umadevi V (2014) Sentiment analysis using weka. Int J Eng Trends Technol (IJETT) 18(4):181–183 9. Pawlak Z (2003) A rough set view on Bayes’ theorem. Int J Intell Syst 18(5):487–498 10. Garcia-Gutierrez J, Martínez-Álvarez F, Troncoso A, Riquelme JC (2014) A comparative study of machine learning regression methods on LiDAR data: a case study. In: International joint conference SOCO’13-CISIS’13-ICEUTE’13. Advances in intelligent systems and computing, vol 239. Springer, Cham 11. van der Aalst W (2016) Data mining. In: Process mining. Springer, Heidelberg

Chapter 8

Novel FPGA-Based Hardware Design of Canonical Signed Digit Matrix Multiplier and Its Comparative Analysis with Other Multipliers Ritik Koul, Mukul Yadav and Kriti Suneja

1 Introduction DSP applications make use of a large number of multiplication operations. In these applications, we require techniques for faster multiplication for increasing the speed of the system. This goal can be achieved by encoding a number in such a manner that the number possesses minimum nonzero digits so that upon multiplication with some other number, the partial products generated are automatically reduced which also results in minimum number of additions required, and hence, reduces the total time associated with the multiplication operation. This technique of encoding a number is referred to as CSD representation [1]. In this paper, first of all, the CSD converter has been implemented which can convert signed as well unsigned binary numbers into its equivalent CSD form directly. Most of the papers presented earlier in this area have neglected the conversion of signed numbers into its CSD equivalent. Multiplication of two signed numbers is quite common in various signal processing applications, for example, discrete Fourier transform (DFT), inverse discrete Fourier transform (IDFT) and in calculating system response through convolution. All these operations can be viewed as linear operations and thus can be implemented using matrix multiplication, and the time required for these applications may involve a great amount of overhead in real-time systems.

R. Koul (B) · M. Yadav · K. Suneja Department of Electronics and Communication, Delhi Technological University, Delhi 110042, India K. Suneja e-mail: [email protected] © Springer Nature Singapore Pte Ltd. 2020 G. Mathur et al. (eds.), International Conference on Artificial Intelligence: Advances and Applications 2019, Algorithms for Intelligent Systems, https://doi.org/10.1007/978-981-15-1059-5_8

65

66

R. Koul et al.

2 Various Multiplication Techniques Conventional multiplication using array multiplier performs multiplication using add and shift operations of the partial products. Several iterations are carried out, and, in each iteration, a partial product is generated by the multiplication of each bit of multiplier by multiplicand. They are then shifted as per their bit orders and are added in the final stage. The number of partial products generated is equal to the number of multiplier bits as shown in Fig. 1. We can reduce the number of partial products significantly by using compressors as done in the Wallace multiplier [2, 3]. It consists of three steps; 3:2 compressors are used to reduce the number of partial product rows iteratively till the number of partial product rows is limited to two only. In the final step, these two rows are added simply to obtain the result as shown in Fig. 2. We can further make the necessary reductions at each level resulting in a design with lesser number of half adders and full adders which is exploited in Dadda multiplication technique [4] as shown in Fig. 3. Involvement of ripple carry addition in conventional array multiplication causes very high amount of delay which can be reduced with the use of carry save addition as shown in Fig. 4 where each box represents an adder. Booth’s algorithm is a signed multiplication algorithm that can be used to halve the number of partial products which in turn increases the speed of multiplication by reducing the number of partial product additions that must take place. The flowchart for this algorithm is shown in Fig. 5 [5]. Ancient mathematical techniques may further lead us in our quest to reduce the delay involved. Vedic multiplication technique is illustrated by the Urdhvatiryakbhyam Sutra of Vedic arithmetic which means vertical and crosswire. This technique is illustrated in Fig. 6 [6]. Fig. 1 Array multiplier structure

8 Novel FPGA-Based Hardware Design of Canonical Signed …

67

Fig. 2 Wallace multiplier structure

3 Multiplication Using CSD Conversion The canonical signed digit (CSD) is a type of number representation. There are various characteristics that favor the use of this kind of representation of numbers for multiplication purposes [7]. Some of these are as follows: • • • • •

In this representation, only three numbers are used which are 0, 1 and −1. Each number has unique CSD representation. The number of nonzero digits in the CSD representation of a number is minimal. There cannot be two consecutive nonzero digits. All these factors lend themselves to the fact that minimum number of partial products will be generated when we multiply the two numbers by representing one of them in CSD form.

68

R. Koul et al.

Fig. 3 Dadda multiplier structure

In our implementation, representation of negative numbers in CSD form has been taken care of, which allows to even use this multiplication algorithm in nearly all signal processing applications without any hindrance. The algorithm for conversion of an n bit binary number to CSD representation is shown in Fig. 7. The algorithm to convert a binary number into its CSD equivalent is as follows: Let the number of bits present in the binary number which is to be converted into CSD equivalent be n; then, the CSD representation of such a number consists of n + 1 number of bits.

8 Novel FPGA-Based Hardware Design of Canonical Signed …

69

Fig. 4 CSA multiplier structure

The algorithm involves the use of two variables which are ‘carr’ and ‘a.’ Initially, the value of ‘carr’ and ‘a’ is set to zero, MSB is set as the MSB of the given binary number, depending upon the value of the ‘carr’ generated in the previous stage successive two bits of the n bit number are compared starting from the LSB, and values of ‘cgen’ and ‘carr’ are set depending upon the result of comparison according to the flowchart. These sequences of steps are iterated for n number of times. Equivalent CSD representation is obtained in ‘cgen.’ However, conversion of a negative number imposes further constraints on this algorithm, in order to be able to convert signed number we have used the techniques as used in the 2’s complement representation of a signed binary number. If we algorithmically just change −1 by 1 and 1 is replaced by −1, the CSD equivalent for positive number becomes CSD equivalent of negative number. Further, the problem arises that representation of −1, 1 and 0 cannot be achieved using a single bit. Thus, for representing these numbers, they have to be encoded into 2 bits and then proceeded with.

4 Matrix Multiplication Since most of the signal processing applications may be viewed as linear operations on input samples, they may be implemented with the help of matrix multiplication. For example, frequency domain analysis of digital systems involves a lot of operations involving matrix multiplications like DFT. Though the use of FFT algorithm has

70

Fig. 5 Booth’s algorithm

Fig. 6 Vedic multiplication algorithm

R. Koul et al.

8 Novel FPGA-Based Hardware Design of Canonical Signed …

71

Fig. 7 Flowchart for CSD conversion

reduced the time required for DFT computation by a significant amount by decreasing the number of computations required, the performance may still be enhanced with the use of proposed multiplier since it can reduce the delay by a significant amount. We can use the proposed method of multiplication for implementing matrix multiplication. By providing the required shifts to the multiplicand, according to the corresponding bit of the multiplier to form the partial products and then adding all the partial products will automatically decrease the number of partial products than, in other multipliers.

72

R. Koul et al.

5 Simulation Results Following are the simulation results for signed and unsigned multiplication performed on ModelSim (shown in Figs. 8 and 9). The comparison of the propagation delay for various multipliers (array multiplier, Wallace tree multiplier, Dadda multiplier, dual carry save multiplier, Booth’s multiplier, Vedic multiplier, CSD multiplier) is shown in Fig. 10. The delay analysis clearly shows that multiplying the two numbers using CSD representation technique gives the least amount of time delay among all the multiplication techniques and that too by a significant amount. If N number of multiplications is to be performed, the resulting improvement is also increased by a factor of N. The logic levels used in various multiplication techniques are shown in Fig. 11. Though the number of logic levels in CSD multiplier is a little more than some multipliers but being a signed multiplier, this may be accepted because of the fact that the only signed multiplier, Booth’s multiplier uses very large number of logic levels for computation of the result.

Fig. 8 2 × 2 Unsigned matrix multiplication

Fig. 9 2 × 2 Signed matrix multiplication

8 Novel FPGA-Based Hardware Design of Canonical Signed …

73

Fig. 10 Propagation delay for various multipliers

Fig. 11 Levels of logic

The comparison of delay involved in matrix multiplication using the CSD conversion technique, and without using it, is shown in Fig. 12. The comparison of the numbers of LUT slices for various multipliers is shown in Fig. 13. It is to be noted that these results for CSD multiplier also include the number of LUT slices that have been used for conversion of binary number in CSD form.

6 Conclusion and Future Work A high-speed matrix multiplier, using CSD conversion methodology, was designed for faster digital signal processing systems. Multiplication requires a large number of hardware components due to large number of partial product generation. In the proposed methodology, reduction in number of partial products was achieved by

74

R. Koul et al.

Fig. 12 Comparison between time delay of matrix multiplier

Fig. 13 Number of slice LUTs used

using CSD conversion of the number resulting in the improvement in speed. This technique can be used for the implementation of dynamic time warping (DTW) algorithm since the algorithm requires a large number of multiplications [8]. Further computation time required in various algorithms like FFT and IFFT may also be reduced if the multiplications required are performed using the proposed multiplier.

8 Novel FPGA-Based Hardware Design of Canonical Signed …

75

References 1. Sharma DH, Ramesh AP (2016) Floating point multiplier using canonical signed digit. Int J Adv Res Electron Commun Eng (IJARECE) 2(11) (F: Article title. J 2(5):99–110) 2. Sharma B, Bakshi A Comparison of 24 × 24 bit multipliers for various performance parameters. Int J Res Advent Technol (E-ISSN: 2321–9637) 3. Momeni A, Montuschi P (2015) Design and analysis of approximate compressors for multiplication. IEEE Trans Comput 64(4):984–994 4. Bhattacharjee A, Sen A (2017) Compare efficiency of different multipliers using Verilog simulation & modify an efficient multiplier. Int J Latest Technol in Eng Manag Appl Sci (IJLTEMAS) 6(3), ISSN 2278–2540 5. Goyal N, Gupta K, Singla R (2014) Study of combinational and booth multiplier. Int J Sci Res Publ 4(5):1 ISSN 2250–3153 6. Yogendri, Gupta AK (2016) Design of high performance 8-bit Vedic multiplier. In: 2016 international conference on advances in computing, communication, & automation (ICACCA) (Spring), Dehradun, pp 1–6 7. Vishwanath BR, Theerthesha TS (2015) Multiplier using canonical signed digit code. Int J Res Appl Sci Eng Technol (IJRASET) 3(5) 8. Suneja K, Bansal M (2015) Hardware design of dynamic time warping algorithm based on FPGA in Verilog. Int J Adv Res Electron Commun Eng (IJARECE) 4(2):165–168

Chapter 9

Design of CMOS Instrumentation Amplifier Using Three-Stage Operational Amplifier for Low Power Signal Processing Shubham Saurabh, Mujahid Saifi, Shylaja V. Karatangi and Amrita Rai

1 Introduction The instrumentation amplifiers are used in amplifying the small differential voltage signal which is produced at the output of sensor transducers. Signals from sensor output become more reliable and easier for processing when signal magnitude is amplified. In recent times, the instrumentation amplifiers are widely employed in power management of battery-operated portable devices such as mobile phones, etc. This amplified voltage signals that occur at the output of instrumentation amplifier are converted by ADC into digital signal [1]. Further, this digital signal is fed to the microcontroller for digital signal processing. Its major applications include Navigation, Radar systems or Biomedical physiological signal processing like ECG, EMG and EEG that helps in diagnosis of diseases present in human body [2]. The Instrumentation amplifier is also used in devices that require high precision likely electronic measuring devices such as multi-meter. The standard Instrumentation amplifier consists of three Op-Amp two-stage structure: Gain stage includes two Op-Amps at input in saturation region and one Op-Amp at differential stage at output in Linear region shown in Fig. 2. The Instrumentation amplifier must have high power gain, high CMRR and with low power consuming behaviour. This paper is organised

S. Saurabh (B) · M. Saifi · S. V. Karatangi · A. Rai G.L Bajaj Institute of Technology and Management, Greater Noida 201306, India e-mail: [email protected] M. Saifi e-mail: [email protected] S. V. Karatangi e-mail: [email protected] A. Rai e-mail: [email protected] © Springer Nature Singapore Pte Ltd. 2020 G. Mathur et al. (eds.), International Conference on Artificial Intelligence: Advances and Applications 2019, Algorithms for Intelligent Systems, https://doi.org/10.1007/978-981-15-1059-5_9

77

78

S. Saurabh et al.

Fig. 1 Block diagram of input/output system of instrumentation amplifier

Fig. 2 Standard Op-Amp based instrumentation amplifier

as follows: Sect. 1—Introduction, Sect. 2—Standard Op-Amp based Instrumentation amplifier is illustrated, Sect. 3—describes the proposed three-stage Op-Amp in CMOS Instrumentation amplifier, Sect. 4—Simulation result, Sect. 5—Conclusion (Fig. 1).

2 Standard Op-Amp Based IA Amplifier The Standard Op-Amp based Instrumentation amplifier consists of three stages: the first stage includes Op-Amp A1 and A2 with their identical matched resistor configuration. Second stage is difference amplifier designed using Op-Amp A3 and four resistors. Op-Amp A1 and A2 are connected in non-inverting configuration. Due to virtual ground concept, the voltage V1 and V2 appear across the resistor 2R1 producing the differential voltage (Vid = V2 − V1) across this resistor [3]. Thus, a current of Ia = (Vid/2R1) flows through the resistance 2R1 and R2. This current further produces the voltage difference of Vo2 − Vo1=[(1+2R2/2R1) * Vid] at the

9 Design of CMOS Instrumentation Amplifier Using Three-Stage …

79

input of Op-Amp A3 at second stage which is difference amplifier and operates on this difference signals [(1 + R2/R1) * Vid], where Vid=Vo2 − Vo1 and gives the output as:  Vo =

R4 R3

   R2 1+ ∗ Vid R1

Thus the differential gain is  Ad =

R4 R3



 1+

R2 R1



This circuit has the advantage of very high input resistance, high differential gain and capability of rejecting common-mode signal when both inputs are connected to a common signal Vcm, then equal potential appears across resistance 2R1 which in turn results into a zero potential difference with Ia = 0. Thus, no common-mode signal is amplified. Hence, high CMRR is achieved (Ideally infinite). 

Ad CMRR(dB) = 20 log Ac



But, in above Op-amp based analysis we have assumed the Op-Amp to be Ideal and high common-mode rejection ratio is achieved by matching the values of resistances, which is difficult task to maintain due to its dependency on physical parameters such as temperature, etc. Thus, there is the need for CMOS Instrumentation Amplifier which is designed using Integrated technology, where resistance values can be designed accurately using VLSI layout design technique on a single monolithic chip. CMOS Integration technique has various benefits such as it reduces power consumption, requires lesser chip area comparison to Op-Amp based circuit, faster operation with high accuracy with improved CMRR performance and higher gain. The previous works [1] on two-stage Operational amplifier have some disadvantages if operated at high speed by reducing channel length then gain is reduced. The threestage direct-coupled capacitor compensation linear CMOS operational amplifier is proposed in the further section of this paper which can be used in obtaining better performance of Instrumentation Amplifier.

3 Proposed Instrumentation Amplifier In the proposed Operational Amplifier used in Instrumentation Amplifier, the transistor level diagram of Op-Amp is shown in Fig. 3. The conventional two-stage Operational amplifier has disadvantages of low DC gain and higher power consumption [4]. Since the Operational amplifiers have a significant role in the performance

80

S. Saurabh et al.

Fig. 3 Three-stage instrumentation amplifier

Table 1 Aspect ratio of CMOS instrumentation amplifier

Transistor

Block

Aspect ratio (W/L)

MB1, MB2

Op-Amp

6 µ/1 µ

M1, M2, M5, M6, M9, M10

Op-Amp

10 µ/1 µ

M3, M4, M7, M8, M11, M12

Op-Amp

12 µ/1 µ

characteristics of the Instrumentation Amplifiers. Also the transistors in current mirror have same channel length(L) [5], thus only the channel width(W) in proposed Operational amplifier is varied (aspect ratios are shown in Table 1) in order to achieve low power consumption, better noise immunity, low offset voltage, high CMRR, high power gain. Gain in the first two stages is increased to a much extent that now even if the resistive load is applied at the output stage the overall gain remains higher than two-stage Op-Amp. Other Design parameters, Capacitance C1, C2 = 240 pF and CL = 0.1 Pf, Ibias = 20 µA.

4 Simulation Result AC analysis of proposed Operational Amplifier circuit in H-spice is done at 180 nm whose bode plot is shown below (Figs. 4 and 5). Phase margin can be increased by increasing the value of compensation capacitor.

9 Design of CMOS Instrumentation Amplifier Using Three-Stage …

81

Fig. 4 Magnitude plot of the output voltage

Fig. 5 Phase plot of the output voltage

5 Conclusion The output parameters achieved from the three-stage Operational amplifier in instrumentation amplifier is better than two-stage operational amplifier in terms of power gain, common-mode rejection ratio, noise immunity and power consumption.

82

S. Saurabh et al.

References 1. Sharma BP, Mehra R (2016) Design of CMOS instrumentation amplifier with improved gain & CMRR for low power sensor applications. In: 2nd international conference on next generation computing technologies (NGCT-2016) Dehradun, India 14–16 October 2016 2. Bansal M, Ranjan RD (2017) CMOS instrumentation amplifier design. Int J Electron Electr Comput Sys IJEECS 6(11), ISSN 2348-117X 3. Sedra AS, Smith KC (2017) Microelectronic circuits, 6th edn. Oxford university press 4. Ren MY, Zhang CX, Sun DS (2012) Design of CMOS instrumentation amplifier. In: International workshop on information and electronics engineering (IWIEE). Elsevier 5. Baker RJ, CMOS circuit design, layout, simulation, 3rd edn. In: IEEE series of microelectronic system. Wiley Publication, IEEE Press 6. Bandyopadhyay S, Mukherjee D, Chatterjee R (2014) Design of two stage CMOS operational amplifier in 180 nm technology with low power and high CMRR. ACEEE, 11 7. Saranya RV, Sureshkumar R (2017) CMOS Instrumentation amplifier for biomedical applications. Int J Eng Res Mod Educ (IJERME) 2(1) 8. Yadav AS, Mishra DK (2016) Int J Adv Res Electr Electron Instrum Eng 5(7) 9. Shojaei-Baghini M, Lal RK, Sharma DK (2004) An ultra low-power CMOS instrumentation amplifier for biomedical applications. IEEE 10. Worapishet A, Demosthenous A, Liu X (2011) A CMOS Instrumentation Amplifier With 90-dB CMRR at 2-MHz using capacitive neutralization: analysis, design considerations, and implementation. IEEE (4) 11. Sharma A (2015) Design and analysis of CMOS instrumentation amplifier. Int J Electr Electron Eng 2(1) 12. Karnik S, Jain PK, Ajnar DS (2012) Design of CMOS instrumentation amplifier for ECG monitoring system using 0.18 µm technology. Int J Eng Res Appl (IJERA) 2(3) 13. Gupta G, Tripathy MR (2014) CMOS instrumentation amplifier design with 180 nm technology. In: 2014 international conference on circuit, power and computing technologies [ICCPCT]. IEEE 14. Schaffer V, Snoeij MF, Ivanov MV, Trifono DT (2009) A 36 V programmable instrumentation amplifier with sub-20 V offset and a CMRR in excess of 120 dB at all gain settings. IEEE J Solid-State Circuits 44(7) 15. Xiu L, Li Z (2012) Low-power instrumentation amplifier IC design for ECG system applications. In: International workshop on information and electronics engineering (IWIEE). Elsevier

Chapter 10

A Novel Image Encryption Technique Using Arnold Transform and Asymmetric RSA Algorithm Gaurav Kumar Soni, Himanshu Arora and Bhavesh Jain

1 Introduction Images are the most commonly used communication modes in various fields such as medical, research, industrial and military fields. Important image transfers are done on an insecure Internet network. Therefore, it is necessary to set appropriate security so that the image prevents unauthorized persons from accessing important information. The advantage of the image is that it needs to cover and protect more multimedia data. Cryptography is a form of image security. It provides a secure way to send and save images on the Internet. Security is the primary concern of the system to preserve image integrity, confidentiality and authenticity. Encryption is the most efficient method, but security problems will also arise if more gray-level data is used [1]. Encryption is the method that protects confidential information by encrypting the information in an unreadable format, and decryption is the process which is reverse to the encryption in which get again in format information that is only readable by who has access rights of it [2].

G. K. Soni Department of ECE, Arya College of Engineering and Research Centre, Jaipur, Rajasthan, India e-mail: [email protected] H. Arora (B) · B. Jain Department of CSE, Arya College of Engineering and Research Centre, Jaipur, Rajasthan, India e-mail: [email protected] B. Jain e-mail: [email protected] © Springer Nature Singapore Pte Ltd. 2020 G. Mathur et al. (eds.), International Conference on Artificial Intelligence: Advances and Applications 2019, Algorithms for Intelligent Systems, https://doi.org/10.1007/978-981-15-1059-5_10

83

84

G. K. Soni et al.

2 Related Work Mathur et al. [2] conferred the modified approach includes a changed approach of RSA that has exponential type of RSA with four prime numbers and plenty of public keys with k-neighbor rule. The proposed modified approach introduces a further level of safety using the closest k-neighborhood rule. It improves the randomness of the calculated value of encoded text as a result of it removes the matter of redundancy in ciphertext same as plain text. In the modified method, the encrypted text does not repeat in the same way as plain text. By using this, any person non grata finds it difficult to hack the info being transmitted. Panda et al. [3] proposed a hybrid security algorithmic program for RSA referred to as HRSA. The planned theme system maintains the safety of the knowledge in encoding and decipherment methodology. They have also described that cryptography is one of the most common techniques which is used in data security. In general, cryptography has two main categories, such as symmetric key and asymmetric key cryptography. In the symmetric key cryptography, used the same key to encrypt and decrypt the information or important data, and in asymmetric key cryptography used the different key of pair in which one key is used to encrypt the information, and another key is used to decrypt the information. In asymmetric key cryptography, the pair of key is called public key and private key. Encryption security also depends principally on the length of the key. Santhosh Kumar et al. [4] discussed encryption is a domain that provides many privacy policies for the image. Confidentiality, authentication, security and safety are necessary to protect the data sent over the Internet or by any means. Data may be attacked by third parties. To ensure its security, we have many techniques such as RSA and AES.

3 Arnold Transformation The transformation of Arnold is the most used in techniques for mixing digital images. Because of this cyclical change, the safety problem, frequent repetition, higher scrambling and therefore the speed or the noise, efficiency is less. In addition, the degree of randomization in the different frequency step is unstable, and sometimes, things get worse, it is also a hidden security problem [5]. We can outline Arnold’s transformation as follows. Let (x, y) inform within the unit sq. It moves to the (x  , y ) by the subsequent equation [1 1]. 

x y



 =

11 12

  x mod 1 y

(1)

10 A Novel Image Encryption Technique Using Arnold Transform …

85

L is the length of a square unit. This conversion called Arnold’s 2D conversion. For digital images, Arnold’s transformation can be defined as follows. Using the next transformation [6], move the pixels of the square digital image I = [I i, j]NXN to (i, j).      i 11 i = mod N j 12 j

(2)

Arnold’s transformation is cyclical and reversible. Moreover, Arnold’s transformation is only valid for square images. The Arnold’s transformation is used to modify the parameters of digital images, especially for digital watermarks. Many articles believe Arnold’s conversion period is 40 dB in rejection band Bandwidth: 2.320–2.345 GHz (rejection band) Return Loss in notch band: 0 db (S11) In this paper different simulated circuits are available with a center frequency of 2.3324 GHz.

3 Findings of Literature Survey Srikanta Pal, Roger D. Norrod, Michael J. Lancaster introduced a novel scaled-down six-pole high-temperature superconducting microstrip bandstop channel working at S-band. Pseudo curved conduct in the recurrence trademark is acquainted by structuring the resonators with reverberate at offbeat frequencies and by changing the entomb resonator stage distinction [1]. G. Zhang, M. J. Lancaster, F. Huang, N. Roddis presented a seven-pole HighTemperature Superconducting (HTS) microstrip bandstop channel at L band. The channel has application in a radio stargazing collector. The crisscross circle resonator and the crisscross stage line are created to decrease the parasitic impact of the immediate resonator-to-resonator coupling [2]. K. Dustakar and S. Berkowitz exhibit the deliberate aftereffects of an ultraprofound (90 dB) HTS lumped-component step channel coordinated over an octavewide recurrence extend, which speaks to a formerly hidden bit of step channel parameter space. The channel was acknowledged utilizing a seven-shaft semi-elliptic plan and was manufactured in a HTS microstrip design. The created step channel has a stop transmission capacity of 3 MHz, a 3 dB pass data transfer capacity of 8 MHz, and an inside recurrence of 1.03 GHz. The channel is all around coordinated from 900 to 1500 MHz with a 16 dB return misfortune. The passband clamor figure was 0.3 dB. The deliberate information demonstrates that the manufactured channel intently adjusts to the reproduction [3].

4 Problem Formulation Using Lumped elements Inductor (L) and Capacitor (C) in the network improves the return loss and insertion loss. Using a Simple microwave office 2002 simulator an idealized bandstop filter can be simulated to help determine realistic values for Return Loss and Insertion Loss.

12 Designing of Band Reject Filter for Radio Astrophysics

93

5 Theoretical Modelling Six-pole bandstop filter is based on Butterworth prototype is simulated in this paper. By taking the different number of elements we observe the results based on return and insertion loss. First of all we have calculated the values of inductors and capacitors by using the formulas based on the bandstop filter. Then draw the circuit in the microwave office simulator 2002. Then analyze the results and compare the return and insertion loss.

6 Design Specification Software used in this project is Circuit Simulator Microwave Office2002. Desired center frequency = 2.3324 GHz Source Impedance = 75  Load Impedance = 75  No of element = 6 The shunt capacitor of the low-pass prototype is converted to series LC circuits having element values given by 1 , ω0 Ck Ck . Ck = ω0 L k =

Using the table we have calculated all the required elements (Table 1; Fig. 1). Table 1 Element calculated for six-pole BSF

Element no.

Inductor (nH)

Capacitor (pF)

1.

0.02834

164.347

2.

338.378

0.01377

3.

0.10582

44.0382

4.

247.714

0.01889

5.

0.07746

60.1560

6.

924.528

0.005041

94

M. Vats et al.

Fig. 1 Schematic of six-pole bandstop filter

7 Simulation Setup The software used to demonstrate and reenact bandstop filter is Microwave office2002 programming. Microwave Office is a full-wave electromagnetic test system. This is the AWR (Applied Wave Research) programming for planning Microwave Integrated Circuits It investigates 3D and multilayer structures of general states of Microstrip Line. It has been broadly utilized in the plan of MICs, RFICs, fix reception apparatuses, wire radio wires and other RF/remote receiving wires. It tends to be utilized to compute and plot the S11 parameters, VSWR, transfer speed, return misfortune just as the radiation designs. An assessment form of the product was utilized to acquire the outcomes for this paper.

8 Results and Discussion We have seen the schematic circuits of bandstop filter. Their analysis is given underneath. Last and the best outcome are given in the last. From the diagram, Return Loss, Isolation Loss just as Bandwidth have been determined.

8.1 Analysis of Three-Pole Bandstop Filter See Table 2 and Figs. 2, 3. Zo = 75  is used in the designing.

12 Designing of Band Reject Filter for Radio Astrophysics

95

Table 2 Element calculated for three-pole BSF Element no.

Inductor (nH)

Capacitor (pF)

1.

L1 = 0.05498

C1 = 85.38

2.

L2 = 240.144

C2 = 0.01955

3.

L3 = 0.05498

C3 = 85.38

Fig. 2 Schematic of three-pole bandstop filter

Fig. 3 Simulated Graph showing Return Loss S11 and Isolation Loss S21

96

M. Vats et al.

8.2 Analysis of FourPole Bandstop Filter See Table 3 and Figs. 4, 5.

8.3 Analysis of Five-Pole Bandstop Filter See Table 4; Figs. 6, 7.

8.4 Final Result (Analysis of Six-Pole Bandstop Filter) See Tables 5, 6 and Figs. 8, 9. Table 3 Element calculated for four-pole BSF Element no.

Inductor (nH)

Capacitor (pF)

1.

0.04193

111.14

2.

258.97

0.01799

3.

0.10288

46.03

4.

625.19

0.0074548

Fig. 4 Schematic of four-pole bandstop filter

12 Designing of Band Reject Filter for Radio Astrophysics

97

Fig. 5 Simulated graph of S11 and S21 Table 4 Element calculated for five-pole BSF Element no.

Inductor (nH)

Capacitor (pF)

1.

0.03398

138.15

2.

296.827

1581.63

3.

0.10997

42.69

4.

296.827

1581.63

5.

0.03398

138.15

Fig. 6 Schematic of five-pole bandstop filter

98

M. Vats et al.

Fig. 7 Simulated graph of S11 and S21

Table 5 Element calculated for six-pole BSF Element no.

Inductor (nH)

Capacitor (pF)

1.

0.02834

164.347

2.

338.78

0.01377

3.

0.10582

44.0382

4.

247.714

0.018890

5.

0.07746

60.156

6.

924.528

0.005041

Table 6 Calculation of insertion loss and return loss S parameter

Center frequency

Insertion loss

S21

2.3324 GHz

−122.15 dB

S parameter

Center frequency

Return loss

S11

2.3324 GHz

−0.5487 dB

9 Conclusion In this paper, we improved the insertion loss and return loss of the base paper. We improved the result and design the bandstop filter for radio astrophysics. By using the Butterworth technique we designed the desired filter for astronomy. In previous work, the value of S21 was about −78 db and we improved it up to −122 db so it has more power rejection in the stopband. According to Table 6 S21—at the center frequency 2.3324 GHz value of S21 parameter is −122 db (approx.), it means at the center frequency only

12 Designing of Band Reject Filter for Radio Astrophysics

99

Fig. 8 Schematic of six-pole bandstop filter

Fig. 9 Simulated graph of S11 and S21

6.09*10ˆ(−13)watt power will be transmitted, which is very low in comparison of 1 W power as an input. So it is very good rejection band. S11—at the center frequency 2.3324 GHz value of S11 parameter is −0.5487 db (approx.) it means at the center frequency 0.8813 W power will be reflected back to the input port, which is very high in comparison of 1 W power as an input. So it is very good rejection band. So we can say that by using this technique we can improve the rejection band at the center frequency 2.3324 GHz.

100

M. Vats et al.

References 1. Pal S, Lancaster MJ, Norrod RD (2012) HTS bandstop filter for radio astronomy, ieee microwave and wireless components letters, vol 22, no 5, May 2012 2. Zhang G, Lancaster MJ, Huang F, Roddis N (2005) A superconducting microstrip bandstop filter for an L-band radio telescope receiver. In: Proceedings of the 35th European microwave conference, pp. 697–700 3. Dustakar K, Berkowitz S (2004) A HTS lumped-element notch filter. In: 2004 IEEE MTT-S international microwave symposium digest, pp 127–130, June 2004 4. Li Y, Lancaster MJ, Huang F (2003) Superconducting microstrip wide band filter for radio astronomy. IEEE MTT-S international microwave symposium digest, pp 551–554, June 2003 5. Huang F (2003) Ultra-compact superconducting narrow-band filters using single- and twinspiral resonators. IEEE Trans Microwave Theor Tech 51(2):487–491 6. Zhou J, Lancaster MJ, Huang F (2003) Superconducting microstrip filters using compact resonators with double-spiral inductors and interdigital capacitors. In IEEE MTT-S international microwave symposium digest, vol 3, pp 1889–1892, IEEE, June 2003 7. Soares ER (1999) Design and construction of high performance HTS pseudo-elliptic band-stop filters. In 1999 IEEE MTT-S international microwave symposium digest (Cat. No. 99CH36282), vol 4. IEEE, pp 1555–1558 8. Hammond RB, Soares ER, Willemsen BA, Dahm T, Scalapino DJ, Schrieffer JR (2002) Intrinsic limits on the Q and intermodulation of low power high temperature superconducting microstrip resonators. In: Selected Papers of J Robert Schrieffer: In Celebration of His 70th Birthday, pp 127–132 9. Hammond RB et al (1998) Intrinsic limits on the Q and intermodulation of the low power high temperature superconducting microstrip resonators. J Appl Phys 84(10) 15 Nov 1998 10. Matsumoto A (ed) (2015) Microwave filters and circuits: advances in microwaves, vol 1. Academic Press 11. Wallage S, Tauritz JL, Tan GH, Hadley P, Mooij JE (1997) High Tc superconducting CPW bandstop filters for radio astronomy front ends [YBa2Cu3O7-LaAlO 3]. IEEE Trans Appl Supercond 7(2)

Chapter 13

Novel Hardware Design of Correlation Function and Its Application on Binary Matrix Factorization Based Features Mayank Jain, Rahul Saini, Manish and Kriti Suneja

1 Introduction In data analysis and machine learning applications, features are extracted from raw data before classification. If this step is accelerated using parallel computing feature of FPGAs, the overall implementation of the application can become fast. Binary matrix factorization provides an efficient way to extract important features from that huge pool of data set. It is an effective method to extract implicit features from a given set of raw data without having expert knowledge of a particular domain. We have used an algorithm by which a large matrix can be decomposed into two small-sized matrices, namely feature matrix and weight matrix. Feature matrices of incoming data (in binary), which may be of various kinds like envelope information, RMS values, etc. provided by software, are checked sequentially, how they are correlated to each other using another algorithm. This hardware method provides a dynamic system in which it calculates correlation in real-time. So our method combines BMF calculation [1] and correlation algorithm [1] to produce correlation which is independent of features need to be compared. Our method not only reduces hardware because of sequential execution but also speeds up signal processing. We have implemented this algorithm in Xilinx based on FPGA using Verilog with an emphasis on optimizing the speed M. Jain (B) · R. Saini · Manish · K. Suneja Department of Electronics and Communication, Delhi Technological University, Delhi 110042, India e-mail: [email protected] R. Saini e-mail: [email protected] Manish e-mail: [email protected] K. Suneja e-mail: [email protected] © Springer Nature Singapore Pte Ltd. 2020 G. Mathur et al. (eds.), International Conference on Artificial Intelligence: Advances and Applications 2019, Algorithms for Intelligent Systems, https://doi.org/10.1007/978-981-15-1059-5_13

101

102

M. Jain et al.

of calculating correlation of raw data. FPGAs provide reconfigurable facility for the hardware designs in initial phases of being manufactured. The possibility of parallel computing can help in accelerating the iterations of the software algorithm. With the continuous increase in algorithm complexity, hardware acceleration based on reconfigurable FPGAs is necessary to maintain compatibility with real-time system. We have selected Verilog because it is the most commonly used hardware description language in the design and verification of digital circuits. We have used the Xilinx tool for final implementation and MATLAB for testing. This paper is organized as follows: Sect. 2 represents the related work with the hardware implementation of dimensionality reducing methods. Section 3 describes the algorithm and MATLAB implementation of BMF. Feature matrices provided by BMF is then used to calculate the correlation. Correlation implementation is given in Sect. 4. Then we have described the approach of its implementation in VERILOG, after that we have given the simulation results. At last, we have given conclusions and further possibilities of improvement in its speed.

2 Related Work BMF breaks a matrix into two low-rank matrices such that the difference between their product and the original matrix is minimum [2]. BMF can also be defined in the same way as Non-negative matrix factorization (NNMF) [3]. It provides ease of data manipulation, which can be used in many applications including information filtering system, correlation analysis, artificial intelligence, analysis of gene expression [4], protein-protein complex interaction network [5], etc. It can also be used in image processing applications like various zeros in binary images can be simplified by using BMF [6, 7]. NNMF has been used for the decomposition of binary data [8]. Miettinen and Vreeken in [9] have proposed a method for order selection of a model of BMF. Battenberg and Wassel in [10] have introduced the idea of using multicore processors for parallelizing the code implementation of Non-Negative Matrix Factorization with an emphasis on fast implementation. In [11], Platos et al. suggested GPU based implementation of NMF using CUDA technology and claimed to achieve a speedup of 200 times. Machine learning application is growing rapidly which requires tremendous data manipulation. Novel techniques are required for reducing the processing time and hardware required. Thus, this is helpful in generating the domain-specific applications and to automate the feature extraction process from a large set of data. Binary matrix factorization technique can make it happen. Bogush et al. [1] have introduced an effective method to implement this concept.

13 Novel Hardware Design of Correlation Function and Its …

103

3 Implementation of BMF 3.1 Algorithm At first input is taken as a matrix A which is divided into two matrices D and B in which B represents a feature matrix while matrix D represents weight matrix for the given matrix A. Initially D and B are null matrices then unique features are extracted from matrix A (using algorithm given below) for matrix B. Matrix D is calculated using Eq. (2) [1]. [A] = [D][B]  [D]i j =

1 if i − th row of A = j − th row of B 0 if i − th row of A = j − th row of B

(1) (2)

If the size of matrix A is n * m then D should have the size of n * k and B should have the size of k * m here k represents a number of hidden features extracted and maximum value of k can be n.

3.2 Flow Chart and Result See Fig. 1. For n = 3, MATLAB execution time is 0.003050 s and for n = 4 execution time is 0.004034 s.

4 Implementation of Correlation 4.1 Algorithm First, specific features that are to be compared like an envelope, RMS value, etc., are selected. Then using MATLAB software features are extracted and converted into binary files F1 and F2. Once these files are obtained, next step is to read them in XILINX during initialization time into two memory blocks in the sequential order such that each row containing a whole block, having size n2 which is to be compared. Then one block, i.e. the entire row of each memory block are read and converted into matrices M1 and M2 of size n × n. Using BMF, reduced feature matrices are extracted from M1 to M2 as B1 and B2 which are less than or equal to the size of M1 and M2. Using the correlation algorithm, their correlation is calculated. If correlated value belongs to very first rows of memory block then it is saved into a variable

104

M. Jain et al.

Fig. 1 Flowchart for BMF

otherwise its value is added to the previously saved value. This process continues until whole memory blocks are read out in sequential order giving a net correlation value, using this value and maximum possible value net percentage is calculated and compared with a threshold. If it is above the decided threshold, the threshold to be decided based on application requirements, then the result would be one otherwise considered as zero.

13 Novel Hardware Design of Correlation Function and Its …

4.2 Flow Chart See Fig. 2.

Fig. 2 Flow chart for correlation

105

106

M. Jain et al.

Fig. 3 XILINX Simulation Results (for n = 4)

5 Implementation of BMF Using Xilinx The proposed hardware is implemented (synthesized and simulated) using Xilinx Integrated Synthesis Environment (ISE), a tool used to synthesize, simulate and validate various HDL algorithms on FPGA platform. For the simulation as mentioned in Sect. 2, we have used xa6slx100-2-fgg484 as the target device having 126,576 number of Slices Registers, 63,288 number of slices LUTs and having 326 number of IOBs. For this hardware implementation, we have taken binary numbers as input.

5.1 Simulation and Synthesis Results of XILINX ISE For a square matrix of size n, we have the following results: For n = 3, XILINX ISE combinational path delay is 8.322 ns, number of slices LUTs used is 22, number of LUTs flip flop used is 6 and number of IOBs used is 32. For n = 4, XILINX ISE combinational path delay is 16.026 ns, number of slices LUTs used is 80, number of LUTs flip flop used is 12 and number of IOBs used is 53. For n = 5, XILINX ISE combinational path delay is 25.982 ns, number of slices LUTs used is 260, number of LUTs flip flop used is 20 and number of IOBs used is 80. For n = 7, XILINX ISE combinational path delay is 42.376 ns, number of slices LUTs used is 1046, number of LUTs flip flop used is 42 and number of IOBs used is 152 (Fig. 3). Figure 4 shows the results of the propagation time delay as we increase the size of the matrices (n), it shows that delay is increasing almost linearly with n. Synthesis results (Figs. 6 and 7) of XILINX shows that Slice LUTs and IOBs usage increases exponentially with n for example, for n = 3 LUTs and IOBs usage is 22 out of 63,288 and 32 out of 326 while for n = 6 LUTs and IOBs usage is 611 out of 63,288 and 113 out of 326.

5.2 Graphs See Figs. 4, 5, 6 and 7.

13 Novel Hardware Design of Correlation Function and Its … Fig. 4 Combinational path delay in ns

Fig. 5 Number of slices

Fig. 6 Number of LUTs

Fig. 7 Number of IOBs

107

108

M. Jain et al.

6 Implementation of Correlation in Xilinx The proposed hardware architecture is implemented (synthesized and simulated) using Xilinx Integrated Synthesis Environment (ISE) tool. For synthesis, xc6slx1003-fgg484 has been used as the target device having 126,576 number of Slices Registers, 63,288 number of slices LUTs and having 326 number of IOBs. For this hardware implementation, inputs have been taken in binary format.

6.1 Simulation and Synthesis Results of XILINX ISE For a square matrix of size n, the following results have been obtained: For n = 3, XILINX ISE combinational path delay is 46.5 ns, number of slices used is 12, number of LUTs used is 240 and number of IOBs used is 23. For n = 4, XILINX ISE combinational path delay is 42.073 ns, number of slices used is 24, number of LUTs used is 279 and number of IOBs used is 38. For n = 5, XILINX ISE combinational path delay is 81.67 ns, number of slices used is 40, number of LUTs used is 1072 and number of IOBs used is 57. For n = 6, XILINX ISE combinational path delay is 101 ns, number of slices used is 60, number of LUTs used is 1611 and number of IOBs used is 80. For n = 7, XILINX ISE combinational path delay is 127 ns, number of slices used is 84, number of LUTs used is 2612 and number of IOBs used is 107. For n = 8, XILINX ISE combinational path delay is 139 ns, number of slices used is 112, number of LUTs used is 4090 and number of IOBs used is 138. Figure 10 shows the results of the propagation time delay as we increase the size of the matrices (n), it shows that delay is increasing almost linearly with n. Simulation results (Figs. 11 and 13) of XILINX show that Slice LUTs and IOBs usage increases exponentially with the size of the matrix.

Fig. 8 Simulation result for different data inputs a and b

13 Novel Hardware Design of Correlation Function and Its …

109

Fig. 9 Simulation result for the same data inputs a and b

Fig. 10 Combinational path delay

Fig. 11 Number of LUTs

Figures 8 and 9 shows simulation results for different data inputs and same data inputs, respectively, where data is read from files and stored into the memory block a and b, for different data correlation is less, 45%, while for similar data correlation is more, 100% for same data.

110

M. Jain et al.

Fig. 12 Number of slices

Fig. 13 Number of IOBs

6.2 Graphs See Figs. 10, 11, 12 and 13.

7 Conclusion FPGAs have been instrumental in accelerating the computationally intensive algorithms. BMF is very efficient in extracting hidden features from a huge data set. Thus, in this paper, we have proposed an architecture of correlation using BMF on FPGA so that it can be incorporated in end-user applications and achieve a significant speed gain. The synthesis results show that hardware utilization increases exponentially with the increase in matrix order. However, the delay is increasing linearly for both correlation and BMF. Also, a significant speedup has been achieved as for n = 3, the execution time in MATLAB is 0.003050 s, for BMF, while the delay in XILINX ISE is around 7 ns. The future work includes the comparative analysis of all feature extraction and dimensionality reduction algorithms on the basis of hardware utilization and delay and suggest the best one for the real-time applications.

13 Novel Hardware Design of Correlation Function and Its …

111

References 1. Bogush R, Maltsev S, Ablameyko S, Uchida S, Kamata S (2001) An efficient correlation computation method for binary images based on matrix factorisation, 2001 IEEE, pp 312–316 2. Jiang P, Heath MT (2013) Mining discrete patterns via binary matrix factorization. In: 2013 IEEE 13th international conference on data mining workshops, pp 1129–1136 3. Husek D, Moravec P, Snasel V, Frolov AA, Rezankova H, Polyakov P (2007) Comparison of neural network boolean factor analysis method with some other dimension reduction methods on bars problem. PReMI 4. Zhang Z-Y, Li T, Ding C, Ren X-W, Zhang X-S (2009) Binary matrix factorization for analyzing gene expression data. Data mining and knowledge discovery 5. Zhang Z, Li T, Ding C, Zhang X (2007) Binary matrix factorization with applications. In: Seventh IEEE international conference on data mining (ICDM 2007), pp 391–400 6. Varga A (1998) Computation of inner-outer factorization of rational matrices. IEEE Trans Autom Control 43(5):684–688 7. Verwoerd WS, Nolting V (1998) Angle decomposition of matrices. Comput Phys Commun 108(2–3):218–239 8. Larsen JS, Clemmensen LKH (2015) Non-negative matrix factorization for binary data. In: 2015 7th international joint conference on knowledge discovery, knowledge engineering and knowledge management (IC3K), Lisbon, pp 555–563 9. Miettinen P, Vreeken J (2011) Model order selection for boolean matrix factorization. In: Proceedings of the 17th ACM SIGKDD international conference on knowledge discovery and data mining, pp 51–59 10. Battenberg E, Wessel D (2009) Accelerating non-negative matrix factorization for audio source separation on multi-core and many-core architectures. In: 10th international society for music information retrieval conference, ISMIR 2009 11. Platos J, Gajdos P, Kromer P Snasel V (2010) Non-negative matrix factorization on GPU. In: International conference on networked digital technologies NDT 2010, CCIS, vol 87, pp 21–30

Chapter 14

Design and Implementation of Biometrically Activated Self-Defence Device for Women’s Safety Sumiran Mehra, Shaurya Deep Singh, Subhangini Kumari, Shylaja Vinaykumar Karatangi, Reshu Agarwal and Amrita Rai

1 Introduction Women are worshiped in India, on the other hand, is being tortured, assaulted, molested, raped and teased either openly or inside their houses. Crime against women is increasing at an alarming rate. The public places, tourist attractions, transports and streets have become the territory of the hunters. There is a silent terror in the street. Women wear full covered clothes to avoid the glance of the roving gaze in the streets. With limited time in our hands and time flying fast, it is our duty to wake up and do something for the improvement of current scenario. The report of WHO states that. “A violence act against female gender disturbed the public health life of society and also it violates the human rights of women.” Protection of women in our society has become a compulsion, not an attribute. To ensure full-fledged protection for women from being harassed, here an equipped electrical device is proposed which can be S. Mehra (B) · S. D. Singh · S. Kumari · S. V. Karatangi · R. Agarwal · A. Rai G.L. Bajaj Institute of Technology and Management, Greater Noida 201306, India e-mail: [email protected] S. D. Singh e-mail: [email protected] S. Kumari e-mail: [email protected] S. V. Karatangi e-mail: [email protected] R. Agarwal e-mail: [email protected] A. Rai e-mail: [email protected]

© Springer Nature Singapore Pte Ltd. 2020 G. Mathur et al. (eds.), International Conference on Artificial Intelligence: Advances and Applications 2019, Algorithms for Intelligent Systems, https://doi.org/10.1007/978-981-15-1059-5_14

113

114

S. Mehra et al.

used by women in adverse cases. From past so many years women safety is one of the prior concern of Indian government and so many measures has already been taken to ensure their safety, it includes pepper sprays, mobile apps, chemical-based devices, electric shock based devices and weapons, firearms, one of the major drawback that is being noticed in all these products is “it can be used against the victim also”. This device eradicates this issue, by using a biometric sensor to get started, hence the device can only be used by the registered user. The main idea behind this product is to empower women and make them stand for themselves and hence a handheld electric shock weapon is being made which ensures no permanent damage [2]. It discharges a high voltage low current electric pulse for a very short duration, which disallows muscles self-triggering mechanism. Many of these devices still exist in the market like stun guns and tasers. But this is the advanced version of these products, aimed at overcoming its drawbacks. This device consists of Bluetooth module by which it is connected to a mobile phone in order to get the geographical location of the device wherever it is being operated. The location is sent to five registered contact numbers, at the same time, when the device is activated a security alarm gets activated using piezo siren. The device consists of two metal electrodes attached from the device which discharges the voltage to the body of the victim. The paper proceeds as follows, in Sect. 1 presents the block diagram model of the device (Fig. 1). Section 2 describes the functionality of each block. Section 3 explains the working of prototype developed. Section 4 presents the conclusion of the paper.

2 Block Diagram Microcontroller: We are using an Arduino Uno microcontroller board. It consists of 14 input–output digital pin and 6analog pin. It uses atmega328p microcontroller chip. This board controls the Bluetooth, sensors and also generates the required PWM for switching. Voltage Booster: A voltage booster is here a combination of boost converter or step-up convertor and voltage multiplier, it is a DC-to-DC power converter that boosts up voltage from input to output while stepping down the current. In order to avoid voltage ripple at the output filters made up of capacitors are used at the converter output. Power for the boost converter is provided by a dc source here: battery. As in the output voltage increases and overall power must be conserved, the output current is lower than the input current [3]. Expression: P = V I

(1)

14 Design and Implementation of Biometrically Activated Self- …

115

Fig. 1 Device model

Model: The boost converter is made on a standard breadboard. A 9 V DC input voltage is provided by a battery. All DC measurements were taken using multimeter and all waveforms were noted in DSO. The main motive behind this boost converter is to increase the voltage from 9 to 425 V, lowering the current. In order to obtain this, a PWM switching circuit is required [4]. Here PWM up to 20 kHz is generated from the microcontroller itself. To raise the voltage from 425 V to 2 kV a voltage multiplier block is used.

116

S. Mehra et al.

The circuit diagram of boost converter using Matlab is shown below (Fig. 2): Working Principle: The booster circuit takes input from dc source and boosts this voltage. PWM input from Arduino is used in switching the booster circuit. High voltage output is received at the capacitor [5]. The voltage characteristics across the resistor at the output of the boost converter is shown below (Fig. 3): Bluetooth Module: Bluetooth is a wireless technology to share data within a short-range. We use a HC05 BT Module, which is operated in the range of 2.4 GHz frequency range [6]. This connects the device to the smartphones and one-way communication is established. Android App: An app is used to process the data received from the Bluetooth. It accesses the GPS location of the phone and sends an emergency message to five registered contacts using radio networks. Biometric Touch Sensor: We are using capacitive touch sensor in this device. It is a transducer that changes the biometric print of a person into an electrical signal. These electrical signals are matched with the prestored print of the finger. If this matching is successful, device gets activated and ready to use.

3 Prototype Developed The operation of this device is divided into three sections. In first section the device can only be activated using biometric sensor, hence only registered users can activate

Fig. 2 Circuit diagram

14 Design and Implementation of Biometrically Activated Self- …

117

Fig. 3 Analysis waveform

the device. Once the device gets activated, the Bluetooth sends signal to the mobile and the location of the device is sent along with an emergency message to five registered contacts (Fig. 4). In the third section, simultaneously the device produces a high voltage pulse of 2 K volts and a low current of 8 mA, across the electrodes for a duration of 20 ms [4, 7]. When these electrodes come in contact with human body having skin resistance between 6  and 50 k. And the person gets numb for few minutes. The current characteristics, when the electrodes come in contact with human skin (Fig. 5).

4 Conclusion The idea behind this proposed paper is to help women to take control of the critical situations. According to the latest revision of NCRB (National Crime Record Bureau) rape is the fourth most common crime against women in India. Many devices have already been developed for women safety, but they can easily be used against the victim. In order to remove this issue, we come up with a device that can be accessed biometrically by the victim. It is portable, doesn’t require proper skill to operate, and at the same time is used for location tracking. Whenever the device is used, the GPS tracks the location and sends the tracked location to the registered contacts, at the same time a high voltage and low current spark is generated for protection against the threat. Thus, this device will add to

118

S. Mehra et al.

Fig. 4 Arrangement of prototype developed

the safety of women and help them to go outside with confidence and catch the fast-moving lifestyle of this era. The compact size of the device makes it a highly preferred choice.

14 Design and Implementation of Biometrically Activated Self- …

Fig. 5 Final prototype developed

119

120

S. Mehra et al.

References 1. Abdallah Dafallah HA (2014) Design and implementation of accurate real time GPS tracking system. In: 2014 third international conference on e-technologies and networks for development (ICEND) 2. Bhilare P, Mohite A, Kamble D, Makode S, Kahane R (2015) Women employee security system using GPS and GSM based vehicle tracking. Int J Res Emerg Sci Technol 2(1) 3. Chougula B, Naik A, Monu M, Patil P, Das P (2014) Smart girls security system. Int J Appl Innov Eng Manage (IJAIEM) 3(4) 4. Thooyavan V (2014) Advanced security system for women, Department of ECE Vidyaa Vikas College of Engineering and Technology Vasai Thane India. In: IEEE 2014 project list under real time target surveillance system, 24 Jun 2014 5. Liu J, Hu X (2013) The research of PWM DC-DC converter based on TMS320F28335, College of Mechatronics and Control Engineering, Hubei Normal University, Huangshi Hubei, China, May 2013 6. Mandapati S, Pamidi S, Ambati S (2015) A mobile based women safety application (I Safe Apps), Department of Computer Applications R.V.R & J.C College of Engineering Guntur India. IOSR J Comput Eng (IOSR-JCE) (Version I) 17(1):29–34. www.iosrjournals.org 7. Baishya BK (2014) Mobile phone embedded with medical and security applications, Department of Computer Science North Eastern Regional Institute of Science and Technology Nirjuli Arunachal Pradesh India. IOSR J Comput Eng (Version IX) 16(3):30–33. www.iosrjournals.org 8. Bhardwaj N, Aggarwal N (2014) Design and development of “Suraksha”—a women safety device. Int J Inf Comput Technol 4 9. Dwivedi CK, Daigavane MB (2010) Multi-purpose low cost DC high voltage generator (60 kV output), using Cockcroft-Walton voltage multiplier circuit

Chapter 15

Facial Expression Recognition Using Random Forest Classifier Kamlesh Tiwari and Mayank Patel

1 Introduction The biometric recognition of people consists of measuring, storing, and comparing specific characteristics of individuals. Among these characteristics, we can highlight the fingerprints, the facial images, the geometry of the hands and fingers, the iris, the retina, the signature, and the voice. For a biometric feature to be effective in the recognition of people, the following properties should be fulfilled to the greatest possible extent: universality (for all users), singularity (different for each user), invariant (with respect to capture conditions), and resistant (to fraud attempts). The biometric method selected will depend, to a large extent, on the type of application in which it will be used and the possible acceptance it generates among its users. Thus, for example, the signature of an individual is a very socially accepted method, but the techniques based on the exploration of the retina by means of a low-intensity infrared beam provoke a strong distrust on the part of the users [1, 2]. In previous research work [3], the authors used PCA and Euclidean distance for face recognition which shows lower accuracy (93.57%). The present research work uses support vector machine classifier-based approach for calculating the similarity score for performance evaluation which provides improved results in terms of recognition accuracy.

K. Tiwari (B) · M. Patel Computer Science and Engineering Department, Geetanjali Institute of Technical Studies, Udaipur 313001, India e-mail: [email protected] M. Patel e-mail: [email protected] © Springer Nature Singapore Pte Ltd. 2020 G. Mathur et al. (eds.), International Conference on Artificial Intelligence: Advances and Applications 2019, Algorithms for Intelligent Systems, https://doi.org/10.1007/978-981-15-1059-5_15

121

122

K. Tiwari and M. Patel

2 Proposed Method Face recognition using extraction of Harris corner and Gabor wavelet features is proposed in this heading. Figure 1 shows the flow diagram of the proposed research work. The process of face recognition is a sequential task. The methods of face recognition are generally studied in three domains that are classified based on their approach. The template matching methods identify the group of pixels in test image that resembles the template image. The rest of methodology is explained as follows:

2.1 Preprocessing of Face Algorithm-1:

Fig. 1 Flow diagram for the proposed approach

15 Facial Expression Recognition Using Random Forest Classifier

123

• Resize function of MATLAB is used to resize the input image into 250 × 250 pixels. This is saved in JPEG data folder. • Image resizing is followed by the RGB to gray conversion using rgb2gray function.

2.2 The Viola–Jones Method for Face Detection Viola–Jones is a real-time face detection algorithm, developed by Paul Jones in 2001 [4]. Primarily, this algorithm was developed for face detection using Haar features in a classifier. The algorithm has four basic steps: Haar Features: These are the features that are used to compute a single value. All the pixels in the black rectangle are added, and all the pixels in the white rectangle are added. Now, the two values are subtracted. Integral Image: In an input image with a resolution of 224 × 224 pixels, this will amount to a lot of computation; thus instead of using an input image, an integral image is used. An integral image is formed by adding all the pixels on the left and top of the pixel under test. This integral image is used for calculating the sum of pixels inside any rectangle using four values. This value is calculated using the four corner values. Integral Sum at D = d − (b + c) + a

(1)

AdaBoost: It is referred as adaptive boost. It is used to select only those features which are likely to give higher values when they match the features. This is done by linearly combining the weight of weak classifiers to form a strong classifier. The equation is given as: F(x) = ∝1 f 1 (x)+ ∝2 f 2 (x)+ ∝3 f 3 (x) + . . .

(2)

A weak classifier is formed by determining the value of each feature on an image and then putting an appropriate weight to it. The mathematical description of a weak classifier is given below: h(x, f, p, θ ) = {1 if p f (x) > pθ

(3)

This explains that an image, v, is classified as positive or negative by the factors as f the feature applied, p the polarity, and θ the threshold. Cascading: Cascading is accomplished using cascade classifier which contains several phases, where each phase is a group of weak learners. These weak learners are trained by using boosting. Boosting provides the ability to train a highly accurate classifier by taking a weighted average of the decisions made by the weak learners.

124

K. Tiwari and M. Patel

2.3 Morphological Operations Dilation: Dilation, also called expansion, filling, or growth, produces a thickening effect on the edges of the object. This algorithm is used to increase the contour of the objects and to join the discontinuous lines of these, produced by some filtration; mathematically, the binary dilation is defined as:   A ⊕ B = c ∈ E N |C=a+b for all a ∈ A and b ∈ B

(4)

Erosion: Erosion is the dual function of expansion, but it is not the reverse; i.e., if erosion is done, then in a dilation the resulting image will not be equal to the actual image; mathematically, erosion is defined as:  AB = x ∈ E N ∨ x + b ∈ A for all b ∈ B

(5)

The median filtering first proceeds by sorting the gray-level values of the neighborhood followed by a selection of the middle element of the sorting. Sorting is done in ascending order generally. It leads to the ordered set of gray values of the neighborhood of f (x0 , y0 ). Since the ordered elements are denoted by f i , the ascending sort is characterized by: f q < f 2 < · · · < f N 2+1 < · · · < f N −1 < f N

(6)

The middle element of the neighborhood is f N 2+1 . Its property is to be preceded by N 2−1 lower values and followed by as many higher values. The filtering consists of replacing f (x0 , y0 ) by the median value of the neighborhood f N 2+1 [5].

2.4 Median Filter Consider a discrete image F characterized by a gray level f (x, y). Let V (x0 , y0 ) be the neighborhood associated with the coordinate point (x0 , y0 ); it is assumed that this neighborhood has N coordinate pixels (x0 − u, y0 − v) with odd N. Let { f 1 , f 2 , . . . , f i , . . . , f N −1 , f N } the gray levels associated with the N pixels of V (x0 , y0 ).

15 Facial Expression Recognition Using Random Forest Classifier

125

2.5 Feature Extraction The steps involved in feature extraction are presented in Fig. 1. We use two feature extraction methods: (a) Gabor Wavelet The feature extraction is based on the Gabor wavelet transform. The Gabor transform performs an analysis of the signal with which it is possible to represent the signal components at each instant of time. It is therefore a time–frequency domain. Gabor’s wavelets are formed by multiplying a sine wave with a Gaussian function. The Gaussian function has a limiting effect; here, the values of each pixel close to a characteristic point of the stone contribute to the convolution. The set of conversion coefficients for the kernels of a different orientation and scale on each characteristic point forms the local feature vector. Each wavelet component describes a section of grayscale values in an image I (x) around a pixel x = (x, y). This is based on the Gabor transform, defined as a convolution.      (7) x ) I x ψ j x − x d 2 x G j ( (b) Harris Corner Method Harris corner detection is a post-processing technique used for smile recognition by the lip corner data, eye detection, and many more [6]. Harris corner detector has to detect the corners in the face image I I . Going stepwise, firstly I I is used to create gradient image through filtering it with Gaussian Mask. And then, the Harris corner method is imposed [6]. 1. Find the vertical gradient and horizontal gradient represent as follows:  M=

P R R Q



 =

Ix2 Ix y Ix y I y2

 (8)

2. Compute x and y derivatives of image: Ix = G σx ∗ I, I y = G σy ∗ I

(9)

3. Compute product of derivatives at every pixel: Ix2 = Ix ∗ Ix , I y2 = I y ∗ I y , Ix y = Ix ∗ I y

(10)

4. Apply Gaussian filter to the image:  wμ,v = exp

 −1  2 μ + v2 /δ 2 2

 (11)

126

K. Tiwari and M. Patel

where wμ,v is a Gaussian window [6] 5. Calculate the r-value of the pixel:  2

2  − k Ix2 + I y2 Rr = Ix2 + I y2 − Ix I y

(12)

6. Select the local extreme points. 7. Threshold determination and corner point selection.

2.6 Similarity Measure Using Random Forest Classifier Decision forest methods have been particularly effective and competitive with the most successful learning methods. In particular, methods that use randomness to generate diversity in tree sets have multiplied. Breiman introduced a formal framework for this category of methods that he named random forest (RF) [7]. This formalism, if it is not at the origin of the use of these methods, allows anyway to draw up a generic theoretical framework and gives some important elements to understand and manipulate these principles of randomization. The common element in all these procedures is that for the kth tree, a random vector θk is generated, independent of the last random vectors θ1 , . . . , θk−1 but with the same distribution; a tree is developed using the training set and θk , resulting in a classifier where h(x, θk ) is an input vector. As indicated above, the random forest method is based on a set of decision trees; i.e., a sample enters the tree and is subjected to a series of binary tests on each node, called split, until it reaches a leaf in which the answer is found. This technique can be used to divide a complex problem into a set of simple problems. In the training stage, the algorithm tries to optimize the parameters of the split functions from the training samples. θk∗ = argmaxθ j ∈ τ I j j

(13)

For this, the following information gain function is used:  S ij   H S ij I j = H ( j) − S j

(14)

i ∈ 1,2

where S represents the set of samples in the node to be divided, and S i are the two sets that are created from the split. The function measures the entropy of the whole and depends on the type of problem we are dealing with [7].

15 Facial Expression Recognition Using Random Forest Classifier

127

3 Simulation and Results The simulation is carried out by using image processing toolbox of MATLAB software. DATABASE: The Japanese Female Facial Expression (JAFFE) Database: The database comprises 213 images of 10 Japanese female models with 7 facial expressions, of which 6 are basic facial expressions and 1 is neutral. Based on 60 Japanese subjects, each of the images is being rated on six emotion adjectives. The database photographs have been captured at the Psychology Department in Kyushu University [8]. Figure 2 shows test images for JAFFE (Figs. 3, 4, 5, 6, 7, 8, and 9). In the proposed support vector machine classifier-based approach, there is not any threshold value for face recognition. Random forest classifier itself does the similarity measure and recognizes test image. Finally, confusion matrix plot shows the performance of Gabor wavelet and Harris corner-based method (Fig. 10).

Fig. 2 Test images for Japanese female facial expression (JAFFE) [8]

Fig. 3 Input image

Fig. 4 Resized to 224 × 224

Fig. 5 Face detection using Viola–Jones method

Fig. 6 Cropped image

128

K. Tiwari and M. Patel

Fig. 7 Eye and mouth detection

Fig. 8 Cropped image

Output Class

Fig. 9 Resized to 110 × 110

an

12 11.2%

0 0.0%

0 0.0%

0 0.0%

0 0.0%

0 0.0%

0 0.0%

100% 0.0%

di

1 0.9%

15 14.0%

0 0.0%

0 0.0%

0 0.0%

0 0.0%

0 0.0%

93.8% 6.3%

fe

0 0.0%

0 0.0%

17 15.9%

0 0.0%

0 0.0%

0 0.0%

0 0.0%

100% 0.0%

ha

0 0.0%

0 0.0%

0 0.0%

14 13.1%

0 0.0%

0 0.0%

0 0.0%

100% 0.0%

ne

0 0.0%

0 0.0%

0 0.0%

0 0.0%

17 15.9%

0 0.0%

0 0.0%

100% 0.0%

sa

0 0.0%

0 0.0%

0 0.0%

0 0.0%

0 0.0%

13 12.1%

0 0.0%

100% 0.0%

su

0 0.0%

0 0.0%

0 0.0%

0 0.0%

0 0.0%

0 0.0%

18 16.8%

100% 0.0%

92.3% 7.7%

100% 0.0%

100% 0.0%

100% 0.0%

100% 0.0%

100% 0.0%

100% 0.0%

99.1% 0.9%

an

di

fe

ha

ne

sa

su

Target Class Fig. 10 Confusion matrix plot for the proposed approach

15 Facial Expression Recognition Using Random Forest Classifier

129

Table 1 Comparison of result with the previous research work Experiments

Recognition rate (accuracy in %) An

Di

Fe

Ha

Ne

[9]

23.1

66.7

58.3

63.5

–

36.7

66.7

[10]

96.7

82.8

84.4

83.9

93.3

83.9

80

[11]

88

73

73

78

–

81

85

JAFFE Database stage-I—7 emotions

Method 1 [3]

100

90

100

95

100

75

95

Method 2 [3]

100

85

95

90

85

80

90

Method 3 [3]

100

80

90

95

85

70

85

JAFFE Database stage-II—5 emotions

Method 1 [3]

90.47

61.9

–

61.9

80.9

–

57.1

Proposed approach: 7 emotions

Using support vector machine classifier

92.3

100

100

100

100

100

100

Previous research

Sa

Su

The row and column are the classes of facial expression database. There are 7 sets of classes and each class having a different set of expressions. The confusion matrix plot indicates the accuracy, i.e., 99.1% for the proposed algorithm. Notations for expression: An—Angry, Di—Disgust, Fe—Fear, Ha—Happy, Ne— Neutral, Sa—Sad, Su—Surprise (Table 1).

4 Conclusion The foremost use of facial expression recognition is in security purposes, yet the technical advancements integrated this technology in the present technical applications such as smile detection in camera. Face recognition is useful in cases when a person adopts disguise looks and makes hard for human eyes to recognize. The applications of face recognition are crucial in security aspects; hence, the need for this research is justified. However, the face recognition is not easy in artificial intelligence and suffers numerous challenges and the process has a specific model to follow. This paper presents facial expression recognition system using the extraction of Gabor wavelet and Harris corner features classified by random forest classifier. Confusion matrix demonstrates that the proposed random forest-based approach gives more accuracy than the previous research works.

130

K. Tiwari and M. Patel

References 1. Meher SS, Maben P (2014) Face recognition and facial expression identification using PCA. In: IEEE international conference in advance computing (IACC), pp 1093–1098 2. Wang G, Ou Z (2006) Face recognition based on image enhancement and Gabor features. In: Proceedings 6th world congress on intelligent control and automation, pp 9761–9764 3. Mehta N, Jadhav S (2016) Facial emotion recognition using log Gabor filter and PCA. In: 2016 international conference on computing communication control and automation (ICCUBEA), IEEE, pp 1–5 4. Viola P, Jones M (2001) Rapid object detection using a boosted cascade of simple features. In: Proceedings of the 2001 IEEE computer society conference on computer vision and pattern recognition, CVPR 2001, IEEE, vol 1, pp. I–I 5. Ibrahim SM, Lagendijk RL (eds) (2012) Motion analysis and image sequence processing, vol 220. Springer Science & Business Media 6. Li Y, Shi W, Liu A (2015) A Harris corner detection algorithm for multispectral images based on the correlation, pp 5–5 7. Breiman Leo (2001) Random forests. Mach Learn 45(1):5–32 8. Lyons MJ, Akamatsu S, Kamachi M, Gyoba J, Budynek J (1998) The Japanese female facial expression (JAFFE) database. In: Proceedings of third international conference on automatic face and gesture recognition, pp 14–16 9. Lajevardi SM, Lech M (2008) Facial expression recognition using neural networks and loggabor filters. In: DICTA’08 Digital image computing: techniques and applications, IEEE, pp 77–83 10. Rose N (2006) Facial expression classification using Gabor and Log-Gabor filters. In: IEEE 7th international conference on automatic face and gesture recognition 11. Sanchez-Mendoza David, Masip David, Lapedriza Agata (2015) Emotion recognition from mid-level features. Pattern Recogn Lett 67:66–74

Chapter 16

Design and Optimization of PV–Wind–DG and Grid-Based Hybrid Energy System for an Educational Institute in India Narendra Gothwal, Tanuj Manglani, Devendra Kumar Doda, Devendra Somwanshi and Mahesh Bundele

1 Introduction Nowadays, the accessibility of electricity is closely associated with human growth [1]. Electricity is one of the fundamental requirements for the economic growth of any nation too. The challenge of meeting the increasing demand of electricity is intimidating for most of the countries, exerting drastic pressure on the energy infrastructure across the globe [2]. In the same way, in India, the electricity demand is growing day by day at a startling pace [3] to support India’s growing manufacturing sector [4]. Today, fossil fuel is the main source of electrical power. This includes coal, oil, and natural gases. These sources will eventually be finished as they are finite, and also these sources are associated with pollution [5]. The problem of pollution is increasing every day with the rising production of electrical power [6]. The rising civilization leads to increased demand of energy [7]. With the increased demand due to urbanization and industries, the cost of energy is also increasing [8]. The conventional energy sources increase the emission of greenhouse gases [9] with increase in power generation due to increased demand. This leads the popularity of renewable energy system worldwide [10]. The requirement for electricity across globe is expected to boost by additional 70%, with 16% rise in energy-related CO2 secretion by the year 2040 [11]. N. Gothwal (B) · T. Manglani Department of Electrical Engineering, Yagyavalkya Institute of Technology, Jaipur, India e-mail: [email protected] D. K. Doda Department of Electrical Engineering, Poornima University, Jaipur, India D. Somwanshi · M. Bundele Poornima College of Engineering, Jaipur, Rajasthan 302022, India

© Springer Nature Singapore Pte Ltd. 2020 G. Mathur et al. (eds.), International Conference on Artificial Intelligence: Advances and Applications 2019, Algorithms for Intelligent Systems, https://doi.org/10.1007/978-981-15-1059-5_16

131

132

N. Gothwal et al.

Due to the above-mentioned problem associated with fossil fuels, we have to move to the alternative sources of energy which do not have these disadvantages. At this point, the renewable energy sources (RESs) like wind and solar are capable to share the demand of energy significantly. Now, these are considered as feasible selection for addition with power plants that run on fossil fuel. At present, RES is satisfying more than 24% of energy demand worldwide. RES is clean and is kind of never-ending sources of power. In 2016, at least one or more types of RES were adopted by more than 170 countries, which were only 43 countries in 2005 [12, 13]. Due to environmental consideration and increased power demand, renewable energy resources have measured as superior choices. Solar and wind energy are accepted as most reasonable sources. It is broadly identified that we can utilize solar and wind energy relentlessly and their conversion into electrical power is not damaging environment, and their ease of access is additional benefit [14, 15]. Renewable energy has a considerable problem, fickleness. One possible solution of such problem could be the hybrid system, which can ensure reliable operation. This system has a renewable energy source and a conventional energy source. Also, a power storage system increases the reliability of the above-said system considerably [16]. However, the availability of useful energy from RES throughout a day is an uncontrollable parameter; that is why a system for energy storage is quite essential to boost the consistency of this system [17]. In addition with the storage system, diesel generator (DG) also improves the reliability of system when used as backup power source [18] and the storage banks make this system highly reliable [19], which can satisfy electrical loads which need more reliability [20]. In the last two decades, the renewable energy generation capacity across the globe has increased significantly [21]. The International Energy Agency analyses that the use of RES could be 50% in the European Union, almost 30% in China and Japan, and will cross mark of 25% in the USA along with India by 2040 [22]. Figure 1 shows the projection for the energy consumption across the globe by 2040. The paper discusses the design along with optimization of grid-connected solar– wind–DG hybrid energy generation system at the educational institute, which is situated at Sitapura, Jaipur, in the state of Rajasthan, where latitude and longitude coordinates are 26.769128O N, 75.841216O E. This system was designed and analyzed with optimization with detailed resource analysis of the site. The different optimized system configurations capable of least net present cost (NPC), cost of energy (COE), and the maximum annual capacity shortage (MACS) have been recognized. The paper is organized in five sections: Sect. 2 shows the input data such as geographical, average solar radiation, average wind speed, and load data. Section 3 represents the design of main system with different input parameters along with objectives of the study. Section 4 shows various results according to the defined objectives. In Sect. 5, the conclusion and future recommendations are discussed.

16 Design and Optimization of PV–Wind–DG and Grid-Based …

133

Fig. 1 World energy consumption projection by 2040

2 Site Selection and Resources’ Feasibility The selected site, Yagyavalkya Institute of Technology (YIT), is situated in state of Rajasthan which is also known as the state of maximum solar radiation intensity in India [23]. Table 1 shows the geographical data of the selected site.

2.1 Wind Data Rajasthan comes under low-to-moderate wind region from the wind power generation viewpoint [24]. The table shows the monthly average wind flow at the site [25]. The wind speed data at 50 m above the ground is taken from the Web site of the NASA [26]. Table 2 shows the monthly average wind speed in a calendar year is in the range of 3–6 m/s. Table 1 Geographical data of selected load site Selected Site

Latitude

Longitude

Altitude above sea level (m)

Time zone

Yagyavalkya Institute of Technology, Jaipur, Rajasthan

26.76913°

75.84122°

360

+5:30 GMT

134

N. Gothwal et al.

Table 2 Average monthly wind speed and solar radiation per day S. no.

Month

Daily radiation (kWh/m2 /d)

Average wind speed (m/s)

1

January

4.177

5.93

2

February

4.929

5.3

3

March

5.75

5.06

4

April

6.437

5.25

5

May

6.834

4.05

6

June

6.902

3.93

7

July

5.68

3.198

8

August

5.395

3.79

9

September

5.87

3.58

10

October

5.352

4.31

11

November

4.423

4.91

12

December

3.897

5.18

2.2 Solar Radiation Rajasthan has more than 325 sunny days in a year which makes it one of the best locations for power generation from solar radiation [27]. Table 2 shows the average monthly solar radiation per day at YIT, Jaipur [28].

2.3 Load Data The connected load is observed by the electricity billing status of selected site and after that same is cross verified by individual metering process and several times survey conducted to observe the change in load due to change in weather at the selected locations, classes, examinations, and vacations at the institute. It is observed that the total connected load is 339 kW, with daily average consumption of 780 kWh and peak load of 177 kW. The monthly average and peak load of weekdays and weekends at the selected site are given in Table 3.

3 Objectives and System Design Figure 1 shows the proposed SPV–wind–DG hybrid system designed in HOMER. This system is designed to satisfy the load demand with least use of fossil fuels and also by keeping low COE and NPC of the system. Three different sizes of PV arrays are considered for optimization. The PV panels are installed at 26.77° inclination, and the ground reflectance is considered as 20% (Fig. 2).

16 Design and Optimization of PV–Wind–DG and Grid-Based …

135

Table 3 Monthly average and peak load of weekdays and weekends S. no.

Month

Year

Average load (kW)

Maximum/peak load (kW)

Weekdays

Weekend

Weekdays

Weekend 7.835

1

May

2017

42.65

21.32

15.671

2

June

2017

39.81

19.9

12.876

6.438

3

July

2017

40.11

20.06

12.876

6.438

4

August

2017

43.02

21.51

12.876

6.438

5

September

2017

61.16

30.58

13.476

6.738

6

October

2017

45.06

22.53

13.476

6.738

7

November

2017

28.1

14.05

9.3

4.65

8

December

2017

26.05

13.03

9.3

4.65

9

January

2018

27.83

13.92

9.3

4.65

10

February

2018

24.95

12.47

9.3

4.65

11

March

2018

29.13

14.57

9.3

4.65

12

April

2018

36.78

18.39

9.3

4.65

Fig. 2 Design of solar–wind–DG–hybrid system

Seven numbers of DC wind turbine model SW whisper 500 having rotor diameter 4.5 m and hub height 22 m are considered. For uninterrupted power supply, batteries are required; therefore, a battery bank is used which is having 100 batteries, each of 150 Ah at 12 V. DG of 100 kW capacity is also connected. A converter is required to convert the DC supply in AC to supply the load and to convert the AC supply in DC to charge the battery bank. Three converters of different sizes (120, 140, and 160 kW) are considered for optimization. Table 4 shows the different components considered with their technical and economical specifications.

136

N. Gothwal et al.

Table 4 Technical and economical specifications of different components System components Solar PV [29, 30]

Technical input

Economical inputs

Parameters

Values

Parameters

Values (|)

Sizes considered

1–200 kW

Capital cost

39,900/kW

Efficiency

25%

Replacement cost

34,900/kW

Derating factor

80%

O&M cost

798/kW

Lifetime

25 years

Wind turbine (SW whisper 500) [31–33]

Numbers considered

0–7

Capital cost

295,000/unit

Rating

3 kW

Replacement cost

231,500/unit

Lifetime

18 years

O&M cost

5900/unit

Bidirectional converter [34–36]

Sizes considered

0–140 kW

Capital cost

24,000/kW

Efficiency

85%

Replacement cost

23,000/kW

Battery [37]

Diesel generator [38–40]

Grid

Lifetime

18 years

O&M cost

600/kW

Numbers considered

0–100

Capital cost

11,000/unit

Nominal capacity

150 Ah

Replacement cost

11,000/unit

Nominal voltage

12 V

Round trip efficiency

80%

Lifetime

5 years

Size

100 kW

Capital cost

670,000/unit

Lifetime

5000 h

Replacement cost

670,000/unit

O&M cost

17/h

Diesel price

75/l

Buy capacity

100 kW

Price

8.35/kWh

Sale capacity

100 kW

Sellback

8.35/kWh

4 Results and Discussion The designed hybrid system has solar PV–wind–DG–battery bank. HOMER optimizes the system for least COE and NPC of the system. The optimization for the system is done considering MACS as 0, 5, and 10%. Figures 3, 4, and 5 show the optimized results for 0%, 5%, and 10% MACS, respectively. It can easily be seen that the COE and NPC for 0, 5, and 10% MACS are different. The COE and NPC for 0% MACS are |4.301 and |3,06,15,150, respectively, while for 10% MACS, |2.511 and |1,78,52,052. This is because, while we consider the 0% MACS, all changes in load have to be satisfied. If the load is very high for a very short time, the requirement of resources also increases. If we can avoid these overload spikes by demand side management, then this creates a very large difference in overall cost of the system as well as the running cost, which can be seen in Fig. 5. The best economical system is obtained while we consider 10% MACS. From Fig. 5, it can be observed that with least COE (|2.511) and least NPC (|1,78,52,052),

16 Design and Optimization of PV–Wind–DG and Grid-Based …

137

Fig. 3 Optimization results considering 0% maximum annual capacity shortage

Fig. 4 Optimization results considering 5% maximum annual capacity shortage

Fig. 5 Optimization results considering 10% maximum annual capacity shortage

HOMER considers only solar–grid–battery for the system, and if it utilizes all the resources (solar–wind–DG–grid–battery), the COE increases to |7.545 and NPC to |5,36,99,628 which is also the case of most reliable system. While the reliability of the system increases, the MACS decreases to zero. Also, the total capital cost is highest for this system as well.

138

N. Gothwal et al.

5 Conclusion and Further Recommendation The increased demand of electrical energy is decreasing the quantity of fossil fuels as the conventional sources of energy rely on fossil fuels only. Also, excessive use of fossil fuel is contaminating the environment as well. Today, only renewable energy is the only hope to overcome these problems with satisfying the energy demand. If the each and every change in the load has to be satisfied, then the COE and NPC are |4.301 and |3,06,15,150, respectively. By proper load management, the COE can be decreased to |2.511 which much less than the cost of energy from the main grid (|8.35). The most reliable combination has all the resources (PV–wind–DG–grid) in the system, and the COE an NPC for this combination are |7.545 and |5,36,99,628, respectively. This study has considered a typical educational institute in India to analyze its load to optimize grid-connected hybrid renewable energy system. However, it should be implemented to understand the realistic challenges and resolutions.

References 1. Valer LR, Manito ARA, Ribeiro TBS, Zilles R, Pinho JT (2017) Issues in PV systems applied to rural electrification in Brazil. Renew Sustain Energy Rev 78(April):1033–1043 2. Kaundinya DP, Balachandra P, Ravindranath NH (2009) Grid-connected versus stand-alone energy systems for decentralized power-a review of literature. Renew Sustain Energy Rev 13(8):2041–2050 3. Bhardwaj H, Manglani T, Kumawat N (2017) A review on grid connected 100 kW roof top solar plant. X(3):36–40 4. Palchak D et al (2017) Greening the grid: pathways to integrate 175 Gigawatts of renewable energy into India’s electric grid, vol I—national study 5. Adaramola MS, Paul SS, Oyewola OM (2014) Assessment of decentralized hybrid PV solardiesel power system for applications in Northern part of Nigeria. Energy Sustain Dev 19(1):72– 82 6. Singh A, Baredar P, Gupta B (2015) Computational simulation and optimization of a solar, fuel cell and biomass hybrid energy system using HOMER pro software. Procedia Eng 127:743–750 7. Gopalasamy K (2016) Power quality analysis and enhancement of grid connected solar energy system. Circuits Syst 7:1954–1961 8. Eroglu M, Dursun E, Sevencan S, Song J, Yazici S, Kilic O (2011) A mobile renewable house using PV/wind/fuel cell hybrid power system. Int J Hydrogen Energy 36(13):7985–7992 9. Hafez O, Bhattacharya K (2012) Optimal planning and design of a renewable energy based supply system for microgrids. Renew Energy 45:7–15 10. Sadeghi S (2018) Study using the flow battery in combination with solar panels and solid oxide fuel cell for power generation. Sol Energy 170(May):732–740 11. International Solar Alliance (2016) Compendium of global success stories in solar energy. vol 2(Sept):1–256 12. Al Garni HZ, Awasthi A, Ramli MAM (2018) Optimal design and analysis of grid-connected photovoltaic under different tracking systems using HOMER. Energy Convers Manag 155(Oct 2017):42–57 13. REN21, Renewable Energy Police Network (REN) (2017) Renewables 2017: global status report, vol 72, no. Oct 2016

16 Design and Optimization of PV–Wind–DG and Grid-Based …

139

14. Khan FA, Pal N, Saeed SH (2018) Review of solar photovoltaic and wind hybrid energy systems for sizing strategies optimization techniques and cost analysis methodologies. Renew Sustain Energy Rev, 92(Dec 2017):937–947 15. Deshmukh MK, Deshmukh SS (2008) Modeling of hybrid renewable energy systems. Renew Sustain Energy Rev 12(1):235–249 16. Akikur RK, Saidur R, Ping HW, Ullah KR (2014) Performance analysis of a co-generation system using solar energy and SOFC technology 17. Maleki A (2018) Design and optimization of autonomous solar-wind-reverse osmosis desalination systems coupling battery and hydrogen energy storage by an improved bee algorithm. Desalination 435(May 2017):221–234 18. Rajbongshi R, Borgohain D, Mahapatra S (2017) Optimization of PV-biomass-diesel and grid base hybrid energy systems for rural electrification by using HOMER. Energy 126:461–474 19. Zhou W, Lou C, Li Z, Lu L, Yang H (2010) Current status of research on optimum sizing of stand-alone hybrid solar-wind power generation systems. Appl Energy 87(2):380–389 20. Yang H, Lu L, Zhou W (2007) A novel optimization sizing model for hybrid solar-wind power generation system. Sol Energy 81(1):76–84 21. Khare V, Nema S, Baredar P (2016) Solar-wind hybrid renewable energy system: a review. Renew Sustain Energy Rev 58:23–33 22. U. Energy Information Administration (2016) International energy outlook 2016, 2040 23. Rajasthan-Solar-Energy-Policy-2014. https://mnre.gov.in/file-manager/UserFiles/GridConnected-Solar-Rooftop-policy/Rajasthan-Solar-Energy-Policy-2014.pdf 24. Wind Energy Development in Rajasthan. http://www.indiaenvironmentportal.org.in/files/file/ wind%20energy%20Rajasthan.pdf 25. https://power.larc.nasa.gov/downloads/POWER_SinglePoint_Daily_20170501_20180430_ 026d77N_075d84E_dc2f8fc0.txt 26. https://power.larc.nasa.gov/data-access-viewer/ 27. Makhija BK (2012) Solar power in Rajasthan challenges and issues in solar RPO compliance/RECs New Delhi 28. https://www.homerenergy.com/products/pro/docs/3.11/finding_data_to_run_homer.html 29. https://www.greentechindia.co.in/solar-panel-power-plant-for-factory-industrial-building/ 30. https://dir.indiamart.com/impcat/solar-panel-installation.html 31. https://relab.co.in/wind-turbine-hawt 32. https://www.indiamart.com/rightrenewabletech/ 33. https://www.freecleansolar.com/Whisper-500-wind-generator-Southwest-Windpower-p/1wh500-10-24.htm 34. Fadaeenejad M, Radzi MAM, Abkadir MZA, Hizam H (2014) Assessment of hybrid renewable power sources for rural electrification in Malaysia. Renew Sustain Energy Rev 30:299–305 35. Vishnupriyan J, Manoharan PS (2018) Prospects of hybrid photovoltaic–diesel standalone system for six different climate locations in Indian state of Tamil Nadu. J Clean Prod 185:309– 321 36. Tribioli L, Cozzolino R (2019) Techno-economic analysis of a stand-alone microgrid for a commercial building in eight different climate zones. Energy Convers Manag 179(Sept 2018):58–71 37. https://dir.indiamart.com/impcat/tubular-batteries.html 38. http://www.jakpower.com/diesel-genset.html 39. http://www.prashagensets.com/product/15/ashok_leyland_range_diesel_gensets_10_250_ kva_.html 40. http://www.radiantpower.in/portfolio-item/100-to-125-kva-diesel-generator/

Chapter 17

Fast Walsh–Hadamard Transform-Based Artificial Intelligent Technique for Transmission Line Fault Detection and Faulty Phase Recognition Gaurav Kapoor, Vikas Soni and Jitendra Yadvendra

1 Introduction Three-phase transmission line faults have to be detected quickly and perfectly in order to repair the faulted phase, restore electricity supply, and decrease the outage time as soon as possible. The encroachment of a robust and perfect fault detection and faulty phase recognition technique under various fault situations has been exclusively studied over the years. The literature assessment of various techniques which are reported till date is introduced hereafter. Reference [1] used space-time approach (STA). Authors in [2] presented a digital impedance pilot relaying (DIPR) technique for a 230-kV, 300-km-long STATCOM compensated transmission line (SCTL). In [3], WT and di/dt methods have been applied for fault recognition in a 33-kV, two-terminal VSC-based DC grid system. In [4], the continuous wavelet transform (CWT) has been employed for detecting the cross-country faults in the double-circuit transmission lines (DCTLs) under CT saturation condition. In [5], authors employed WT as a fault classifier for the classification of faults in a 400kV, 300-km-long two-terminal transmission line. In [6, 8, 9], authors applied WT for detecting faults on TLs. In [7], WHT has been introduced for the protection of SCCTL. In [10], mathematical morphology has been used for SPTL protection. In this work, the fast Walsh–Hadamard transform (FWHT) is used and it is implemented for the detection and categorization of faults in the TPTL. No such type of work has been reported yet to the best of the knowledge of the author. The G. Kapoor (B) · V. Soni · J. Yadvendra Department of Electrical Engineering, Modi Institute of Technology, Kota, India e-mail: [email protected] V. Soni e-mail: [email protected] J. Yadvendra e-mail: [email protected] © Springer Nature Singapore Pte Ltd. 2020 G. Mathur et al. (eds.), International Conference on Artificial Intelligence: Advances and Applications 2019, Algorithms for Intelligent Systems, https://doi.org/10.1007/978-981-15-1059-5_17

141

142

G. Kapoor et al.

400 kV Source-2

400 kV Source-1

Relay

Bus-1

Fault

Bus-2

Fig. 1 Single-line diagram of TPTL

results exemplify that the FWHT efficiently detects and categorizes the faults, and the consistency of the projected technique is not perceptive to varying fault factors. This article is structured as follows: The specifications of the test system of the TPTL are reported in Sect. 2. Section 3 presents the flowchart of FWHT. Section 4 is devoted to the analysis of the reaction of the FWHT for different fault cases. Section 5 completes the article.

2 Specifications of the Simulink Model of TPTL The simulation test model of a 400 kV TPTL is designed using MATLAB. Figure 1 depicts the single-line illustration of the TPTL. The TPTL has a rating of 400 kV, 50 Hz, and has a total length of 200 km. The TPTL is alienated into two zones of length 100 km each. The transducers are connected at bus-1 for the relaying of the TPTL.

3 Fast Walsh–Hadamard Transform (FWHT) The various stages of the proposed technique as shown in Fig. 2 are depicted in detail hereafter. Step 1 Simulate the system, and generate post fault three-phase current signals. Step 2 Analyze the three-phase currents using FWHT for the removal of their characteristics. Step 3 Estimate the amplitude of FWHT coefficients for each fault current signal. Step 4 The phase will be declared as the faulted phase if its coefficient has the larger amplitude in comparison to the healthy phase.

17 Fast Walsh–Hadamard Transform-Based Artificial Intelligent …

143

Record three phase current signals

Fast Walsh Hadamard transform based current signals decomposition

Characteristics removal in terms of Walsh Hadamard coefficients

No Is |FWHT coefficient| faulted phase > |FWHT coefficient| healthy phase

No fault

Yes

Simultaneous fault detection and faulty phase recognition

Fig. 2 Schematic of the proposed technique

4 Performance Appraisal To confirm the competence of the FWHT, the simulation work has been carried out for numerous faults. The results of the work are examined in the consecutive subsections.

4.1 Performance of FWHT During Fault Type (FT ) Variation The consequence of F T variation on the execution of the FWHT is estimated in this part. Like an illustration, an AG fault at 50% at 0.05 s is analyzed among RF = 5  and RG = 7 . Figure 3 exemplifies the AG fault current waveform. Figure 4 illustrates the FWHT coefficients of AG fault currents. For other faults, the fault

144

G. Kapoor et al.

Fig. 3 AG fault current at 50% at 0.05 s among RF = 5  and RG = 7 

Fig. 4 FWHT coefficients for AG fault current at 50% at 0.05 s

factors are set as T = 0.05 s, F L at 50%, RF = 5 , and RG = 7 . Table 1 reports the results for F T variation. It is inspected that FWHT performs well during F T variation.

4.2 Performance of FWHT During Fault Resistance (RF ) Variation The FWHT is tested for fault resistances from 0.5 to 100 . Figure 5 depicts the ABCG fault current waveform at 57.5% at 0.1 s among RF = 50  and RG = 5 . For other faults, the fault factors are set as T = 0.1 s, F L = 57.5%, and RG = 5 . Table 2 reports the results for RF variation. From the results, it can be concluded that the working of FWHT is not affected during RF variation.

17 Fast Walsh–Hadamard Transform-Based Artificial Intelligent …

145

Table 1 Results of FWHT for different faults at 50% at 0.05 s Fault type

RF ()

RG ()

T (s)

F L (%)

FWHT coefficients Phase A

Phase B

Phase C

AG

5

7

0.05

50

1.0010 × 103

31.5122

21.3192

BG

5

7

0.05

50

15.0563

536.3960

18.5680

CG

5

7

0.05

50

20.0450

24.3578

504.9155

ABG

5

7

0.05

50

1.2483 × 103

757.7414

15.0759

BCG

5

7

0.05

50

32.8774

2.1325 × 103

1.3781 × 103

ACG

5

7

0.05

50

853.6091

18.6097

971.9445

ABCG

5

7

0.05

50

1.5281 × 103

1.0866 × 103

1.0605 × 103

Fig. 5 ABCG fault current waveform at 57.5% at 0.1 s, RF = 50  and RG = 5  Table 2 Results of FWHT for different fault resistances at 0.1 s and at F L = 57.5% Fault type

RF ()

RG ()

FIT (s)

F L (%)

FWHT coefficients

AG

0.5

5

0.1

57.5

552.9276

10.9444

ABG

25

5

0.1

57.5

711.9553

777.4815

19.9078

ABCG

50

5

0.1

57.5

261.1570

317.3046

211.3040

BCG

75

5

0.1

57.5

12.8172

117.1722

203.2980

CG

100

5

0.1

57.5

21.0942

24.3150

183.1514

Phase A

Phase B

Phase C 12.2839

146

G. Kapoor et al.

Fig. 6 Waveform of ACG fault at 47.5% and CG fault at 52.5% at 0.08 s

Table 3 Results of FWHT for two different position faults among RF = 1.5  and RG = 1.5  Fault 1 with location (%)

Fault 2 with location (%)

T (s)

RF ()

AG (35%)

BG (65%)

0.165

1.5

ACG (47.5%)

CG (52.5%)

0.08

BCG (62.5%)

BG (37.5%)

AG (37.5%) BG (67.5%)

RG ()

FWHT coefficients Phase A

Phase B

Phase C

1.5

978.1641

569.6173

188.2872

1.5

1.5

2.5501 × 103

177.3851

2.3548 × 103

0.0555

1.5

1.5

176.3107

1.7037 × 103

2.1233 × 103

ABG (62.5%)

0.1255

1.5

1.5

1.7622 × 103

706.3612

223.9586

BCG (32.5%)

0.1315

1.5

1.5

117.4915

627.5332

372.5664

4.3 Performance of FWHT for Faults at Two Different Positions The two faults can occur at two diverse positions. The FWHT is tested for two different position faults. The waveform of ACG fault at 47.5% and CG fault at 52.5% at 0.08 s among RF = 1.5  and RG = 1.5  is exemplified in Fig. 6. Table 3 reports the results for two different position faults. It is understandable that the FWHT works well for detecting two different position faults.

4.4 Performance of FWHT During Converting Faults The FWHT is tested for the converting faults. Figure 7 illustrates the plot when AG fault at 57.5% at 0.05 s is converted into ABG fault at 57.5% at 0.25 s among RF =

17 Fast Walsh–Hadamard Transform-Based Artificial Intelligent …

147

Fig. 7 Plot when the AG fault at 0.05 s is converted into ABG fault at 0.25 s

Table 4 Results of FWHT for different converting faults with RF = RG = 0.001  and F L = 57.5% Fault type with time (s)

FL (%)

RF ()

RG ()

Transformed fault type with time (s)

FWHT coefficients Phase A

Phase B

Phase C

AG (0.05)

57.5

0.001

0.001

ABG (0.25)

1.0756 × 103

1.6386 × 103

18.7607

BCG (0.1)

57.5

0.001

0.001

BG (0.2)

20.1360

864.6791

706.0797

CG (0.1255)

57.5

0.001

0.001

ABCG (0.25)

676.4762

507.9762

741.0188

ABG (0.072)

57.5

0.001

0.001

ABCG (0.235)

1.4172 × 103

948.9295

341.8116

ABCG (0.116)

57.5

0.001

0.001

BG (0.25)

644.6452

1.4042 × 103

1.1508 × 103

RG = 0.001 . Table 4 reports the results for the converting faults. It is detected that the FWHT performs perfectly for detecting the converting faults.

4.5 Performance of FWHT During Fault Triggering Time (FTT) Variation The FWHT is examined for different values of fault triggering time. Figure 8 exemplifies the result of ABG fault at 42.5% at 0.158 s among RF = 0.25  and RG = 0.25 . Table 5 reports the results for FTT variation. It is examined that the FWHT performs well during FTT variation.

148

G. Kapoor et al.

Fig. 8 Waveform of ABG fault at 42.5% at 0.158 s, RF = 0.25 , and RG = 0.25  Table 5 Results for different fault triggering time at F L = 42.5%, RF = RG = 0.25  Fault type

RF ()

RG ()

FIT (s)

F L (%)

FWHT coefficients Phase A

Phase B

Phase C

BG

0.25

0.25

0.0934

42.5

27.9907

958.3651

34.1232

BCG

0.25

0.25

0.116

42.5

19.9336

1.4459 × 103

2.6087 × 103

ABCG

0.25

0.25

0.08

42.5

3.3339 × 103

1.4088 × 103

1.9547 × 103

ABG

0.25

0.25

0.158

42.5

2.8937 × 103

2.1461 × 103

27.4224

AG

0.25

0.25

0.1412

42.5

852.8849

15.7693

17.6395

5 Conclusion The fast Walsh–Hadamard transform (FWHT) is perceived to be very effectual under diverse fault categories for the TPTL. The FWHT coefficients of the three-phase fault currents are estimated. The fault factors of the simulation model are varied, and it is revealed that the fault factors do not manipulate the fidelity of the FWHT. The results confirm that the FWHT has the competence to protect the TPTL beside various fault categories.

References 1. Gharavi H, Hu B (2018) Space-time approach for disturbance detection and classification. IEEE Trans Smart Grid 9(5):5132–5140 2. Singh AR, Patne NR, Kale VS, Khadke P (2017) Digital impedance pilot relaying scheme for STATCOM compensated TL for fault phase classification with fault location. IET Gener Trans Distrib 11(10):2586–2598

17 Fast Walsh–Hadamard Transform-Based Artificial Intelligent …

149

3. Geddada N, Yeap YM, Ukil A (2018) Experimental validation of fault recognition in VSC-based DC grid system. IEEE Trans Ind Electron 65(6):4799–4809 4. Govar SA, Seyedi H (2016) Adaptive CWT-based transmission line differential protection scheme considering cross-country faults and CT saturation. IET Gener Trans Distrib 10(9):2035–2041 5. Rajaraman P, Sundaravaradan NA, Meyur R, Reddy MJB, Mohanta DK (2016) Fault classification in transmission lines using wavelet multi-resolution analysis. IEEE Potentials 35(1):38–44 6. Kapoor G (2018) Six-phase transmission line boundary protection using wavelet transform. In: Proceedings of the 8th IEEE India International Conference on Power Electronics (IICPE), Jaipur, India 7. Sharma P, Kapoor G, Ali S (2018) Fault detection on series capacitor compensated transmission line using Walsh hadamard transform. In: Proceedings of the IEEE international conference computing, power and communication technologies (GUCON), Greater Noida, India, pp 763– 768 8. Kapoor G (2018) Wavelet transform based detection and classification of multi-location three phase to ground faults in twelve phase transmission line. Majlesi J Mech Syst 7(4):47–60 9. Gautam N, Ali S, Kapoor G (2018) Detection of fault in series capacitor compensated double circuit transmission line using wavelet transform. In: Proceedings on IEEE international conference on computing, power and communication technologies (GUCON), IEEE, Greater Noida, India, pp. 769–773 10. Kapoor G (2018) Six-phase transmission line boundary protection using mathematical morphology. In: Proceedings of the IEEE international conference on computing, power and communication technologies (GUCON), Greater Noida, India, pp. 857–861

Chapter 18

New Designs and Analysis of Multi-Core Photonic Crystal Fiber Using Ellipse with Different Radiuses and Angles Trimesh Kumar, Mayank Sharma and Brijraj Singh Solanki

1 Introduction Photonic crystal fiber has exposed this firmly established area indicates a vast range of novel optical properties to achieve by using conventional fiber technology [1]. PCF is regularly known as microstructure optical fiber. A systematic air hole in the triangular lattice with the axis of the core is utilized along with the solid core [2]. PCF is characterized by geometric constraints such as holes with hole spacing, pitch, air hole diameter, and air filling portions [3]. PCF’s unique manufacturing technology allows accurate modulation of optical properties by changing these geometric parameters. By applying these constraints [4], PCF can be designed to be utilized in a wide range of applications in the field of nonlinear optics [5], optical communication, telecommunications, biomedical sensing, high power technology, and military system. Apart from this, fiber tilt in some applications and properties can be an important issue [6]. Nevertheless, compared to conventional fibers, many air holes in the cladding are much stronger in bending to PCF bending [7]. Based on design details, PCF can be firmly designed for small or large effective mode areas, where small scale or large scale modes area are very powerful or delicate leading to optical nonlinearities [8]. Great mode reduces the loss caused by various nonlinear effects in the area, which limits the performance of upper power systems [9]. In addition, the removal of air holes nearby the core is an alternative technique to increase the core size [10]. The proposed PCF for our knowledge displays the largest effective mode field with some missing air holes to date [11]. It is shown that down doping silica in PCF core can be significantly improved in the effective mode T. Kumar (B) · M. Sharma · B. S. Solanki Department of Electronics and Communication Engineering, Poornima College of Engineering, Jaipur, Rajasthan, India e-mail: [email protected] © Springer Nature Singapore Pte Ltd. 2020 G. Mathur et al. (eds.), International Conference on Artificial Intelligence: Advances and Applications 2019, Algorithms for Intelligent Systems, https://doi.org/10.1007/978-981-15-1059-5_18

151

152

T. Kumar et al.

Fig. 1 PCF structure radius of pitch 1.3 and circular a ellipse R = 0.26 and 1, b ellipse R = 0.3 and 1.4 and PCF structure radius of pitch 1.6 c ellipse R = 0.26 and 1, d ellipse R = 0.3 and 1.4

area [12]. Mostly, studies have shown that while the wide range of wavelengths can be dramatically different in the PCF [13], the reasonable comparison on the specific wavelength shows a light difference in the direction between the effective mode area and the inclination sensitivity [14]. Presently, research for the PCF has been widely searched in an attempt to see and use their broad potential. Mode area is an important section in fiber laser to grow and feel great power systems [15]. In wavelength and structure systems, nonlinear effects are unavoidable and can not be ignored to be systematic. Therefore, it is required to use PCF with a great mode area. For many applications, it is necessary to turn insensitive PCF design which simultaneously displays big mode areas and low confinement loss [16]. In this study, several hexagonal PCF structures are proposed as shown in Fig. 1. This paper is divided into five sections. Section 1 presents a brief introduction. Section 2 discusses the proposed structure. Section 3 is a simulation and the outcome is discussed in Sect. 4. Comparative analysis is presented in Sect. 4 and paper is concluded in Sect. 5.

2 Proposed Structure The main objective of the proposed PCF structure is to achieve a larger effective mode area along with grate birefringence and confinement loss. In this regard, various doping levels have been used in Fig. 1a–d to improve the effective mode area and reduce the loss of the prison. We have designed four hexagonal structures; different ellipses radiuses are used name wise R1 , R2 , R3 , and R4 . They are given the different values R1 = 0.26 µm, R2 = 1 µm, R3 = 0.30 µm, and R4 = 1.4 µm. These are structure design by two pitch values 2 = 1.3 and 2 = 1.6. Wavelength is 0.7–1.8 µm. These three different angles are using