SScchhoollaarrss'' MMiinnee Doctoral Dissertations Student Theses and Dissertations Spring 2015 DDeetteeccttiioonn aanndd rreeccooggnniittiioonn ooff RR//FF ddeevviicceess bbaasseedd oonn tthheeiirr uunniinntteennddeedd eelleeccttrroommaaggnneettiicc eemmiissssiioonnss uussiinngg ssttoocchhaassttiicc aanndd ccoommppuuttaattiioonnaall iinntteelllliiggeennccee mmeetthhooddss Shikhar Prasad Acharya Follow this and additional works at: https://scholarsmine.mst.edu/doctoral_dissertations Part of the Computer Sciences Commons, Electrical and Computer Engineering Commons, and the Operations Research, Systems Engineering and Industrial Engineering Commons DDeeppaarrttmmeenntt:: EEnnggiinneeeerriinngg MMaannaaggeemmeenntt aanndd SSyysstteemmss EEnnggiinneeeerriinngg RReeccoommmmeennddeedd CCiittaattiioonn Acharya, Shikhar Prasad, "Detection and recognition of R/F devices based on their unintended electromagnetic emissions using stochastic and computational intelligence methods" (2015). Doctoral Dissertations. 2373. https://scholarsmine.mst.edu/doctoral_dissertations/2373 This thesis is brought to you by Scholars' Mine, a service of the Missouri S&T Library and Learning Resources. This work is protected by U. S. Copyright Law. Unauthorized use including reproduction for redistribution requires the permission of the copyright holder. For more information, please contact [email protected]. DETECTION AND RECOGNITION OF R/F DEVICES BASED ON THEIR UNINTENDED ELECTROMAGNETIC EMISSIONS USING STOCHASTIC AND COMPUTATIONAL INTELLIGENCE METHODS by SHIKHAR PRASAD ACHARYA A DISSERTATION Presented to the Faculty of the Graduate School of the MISSOURI UNIVERSITY OF SCIENCE AND TECHNOLOGY In Partial Fulfillment of the Requirements for the Degree DOCTOR OF PHILOSOPHY in SYSTEMS ENGINEERING 2015 Committee Members Dr. Ivan G. Guardiola, Advisor Dr. Akim Adekpedjou Dr. Steven Corns Dr. Cihan H. Dagli Dr. Randy H. Moss iii ABSTRACT Radio Frequency (RF) devices produce some amount of Unintended Electro- magnetic Emissions (UEEs). UEEs are generally unique to a device and can be thought of as a signature of the device. This property of uniqueness of UEEs can be used to detect and identify the device producing the emission. The problem with UEEs is that they are very low in power and are often buried deep inside the noise band which makes them difficult to detect. There are two types of UEE detection methods. The first one is called stimulated detection method where the UEEs of a deviceareenhancedusingexternalstimulationsignalandthedetectionismadebased on the analysis of the enhanced stimulated signal. This method, however, is resource intensive as the generation, transmission, and reception of the stimulation signal re- quires hardware components. The second UEE detection method is called passive detection method where the UEE signals are not tampered with and are analyzed in its original raw form. Since the UEEs are weak in strength, the challenge with passive detection method is to measure and analyze UEEs in a noisy environment. In order to detect and recognize RF devices through the UEE, the first step is to measure the leakage of electric signal that is emitted outside of the RF devices as UEEs. UEE samples are collected from two RF devices at three different distances of 3 feet, 6 feet and 10 feet and also for noise in a similar environment. The three methods explored inthis researchare Principal ComponentsAnalysis (PCA), Hidden Markov Model (HMM), and Support Vector Machine (SVM). This research studies theperformanceofthesethreealgorithmsforpassivedetectionofUEEsandcompares it with the performance of Neural Network (NN). The explored methods gives signif- icant better results than existing methods and can be used as an alternative for the costly and resource intensive stimulated detection methods. One of the major appli- cation of UEE is in the detection of Improvised Explosive Devices (IEDs). Effective IED detection system for military operation should accurately perform the task of detection, localization, and direction of malicious devices. This research contributes to the detection and recognition of IED detection system by proposing models based on stochastic and computational intelligence methods. These methods proved to have promise if it can be implemented in real life with more applied research. iv ACKNOWLEDGMENTS I would like to thank my academic advisor Dr. Ivan G. Guardiola for his continuous support. This work would not have been possible without his guidance and supervision. I would also like to thank my committee members Dr. Akim Adekpedjou, Dr. Steven Corns, Dr. Cihan Dagli, and Dr. Randy Moss for their thoughtful comments to improve the quality of this work. I would like to thank all the members of Smart Engineering Systems Lab, especiallytheonesaheadofmeforcreatingapositiveandfriendlyworkenvironment. My special thanks goes to Ritesh Arora for his significant contribution in collecting the UEE signal used in this work. I would also like to thank Nepali community of Rolla for making me feel at home. I am very thankful to Dr. Bonnie Bachman for accepting me as her research assistant during the summer of 2014. Many many thanks to Nancy and Tiny, my guardians in Rolla. It was my privilege to know Sashi Gurung: friend, family member, and a great human being. Frequent visits of Dilip Yogiprovidedmewiththemuchneededsocialoutingsthatkeptmegoingthroughout the PhD program. My special thanks goes out to my parents Keshav Prasad Acharya and Lalita Acharya. Their love and support throughout my life was vital in one way or the other in completing this program. My younger sister Achala completed PhD ahead of me as if to tell me if I can do it, you can too. Ideally I should not thank my daughter Juneli for this work. Her contribution, if anything, was distraction. But she is the kind of distraction that makes life worth living. My wife Shristy Bashyal Acharya has sacrificed so much for the completion of my PhD degree that her name should be somewhere in the title page. Thank you for your support and encouragement. v TABLE OF CONTENTS Page ABSTRACT .......................................................................... iii ACKNOWLEDGMENTS............................................................ iv LIST OF ILLUSTRATIONS ........................................................ viii LIST OF TABLES ................................................................... x LIST OF NOTATIONS .............................................................. xi SECTION 1. INTRODUCTION.............................................................. 1 2. UNINTENDED ELECTROMAGNETIC EMISSIONS ...................... 8 3. BACKGROUND................................................................ 13 3.1. DETECTION METHODS................................................ 13 3.1.1 Stimulated Detection Methods.................................... 14 3.1.1.1 Modulation method....................................... 14 3.1.1.2 Long PN code method.................................... 14 3.1.1.3 Stagner method ........................................... 16 3.1.2 Passive Detection Method......................................... 16 3.1.2.1 Matched filter method.................................... 16 3.1.2.2 Cascading correlation method ........................... 17 3.1.2.3 Hurst parameter method ................................. 17 3.1.2.4 Granulometric size distribution.......................... 18 3.2. RECOGNITION METHODS ............................................ 19 3.2.1 Neural Networks ................................................... 19 3.3. SUMMARY................................................................ 20 4. METHODOLOGY.............................................................. 22 4.1. PRINCIPAL COMPONENTS ANALYSIS.............................. 23 4.1.1 Definition ........................................................... 28 4.1.2 UEE Detection Using PCA........................................ 30 4.1.3 Feature Extraction................................................. 31 4.2. HIDDEN MARKOV MODELS .......................................... 32 4.2.1 HMMs and UEEs .................................................. 33 vi 4.2.2 Definition ........................................................... 34 4.2.3 Three Problems for HMM......................................... 35 4.2.4 Solution to the Evaluation Problem .............................. 36 4.2.5 Solution to the Decoding Problem................................ 37 4.2.6 Solution to the Learning Problem ................................ 38 4.2.7 Data Collection and Preprocessing ............................... 38 4.2.8 Feature Extraction................................................. 38 4.2.9 Training............................................................. 39 4.2.10Assumptions of HMM ............................................. 40 4.3. SUPPORT VECTOR MACHINE........................................ 42 4.4. NEURAL NETWORKS .................................................. 44 5. RESULTS ....................................................................... 48 5.1. DETECTION.............................................................. 48 5.1.1 Principal Components Analysis................................... 48 5.1.2 Hidden Markov Models............................................ 54 5.1.3 Support Vector Machine........................................... 55 5.1.4 Neural Networks ................................................... 56 5.2. RECOGNITION........................................................... 56 5.2.1 Principal Components Analysis................................... 57 5.2.2 Hidden Markov Models............................................ 57 5.2.3 Support Vector Machine........................................... 58 5.2.4 Neural Networks ................................................... 59 6. CONCLUSION AND FUTURE WORKS .................................... 60 APPENDICES A. Matlab Code for UEE Data Processing.................................... 64 B. Matlab Code for PCA calucation of UEEs ................................ 72 C. R code for Baum Welch algorithm ......................................... 74 D. R code for Support Vector Machine........................................ 76 E. Data segment for D1 at 3 feet .............................................. 79 F. Data segment for D1 at 6 feet .............................................. 81 G. Data segment for D1 at 10 feet............................................. 83 H. Data segment for D2 at 3 feet.............................................. 85 I. Data segment for D2 at 6 feet............................................... 87 vii J. Data segment for D2 at 10 feet ............................................. 89 K. Data segment for Noise ..................................................... 91 BIBLIOGRAPHY .................................................................... 93 VITA .................................................................................. 100 viii LIST OF ILLUSTRATIONS Figure Page 1.1 Mobile Cellular Subscriptions in US (per 100 people)...................... 2 1.2 Improvised Explosive Device (adapted from commons.wikimedia.org)... 5 1.3 Coalition Forces Death in Afghanistan ...................................... 5 2.1 Unintended Electromagnetic Emission from a walkie talkie radio ......... 9 2.2 UEE and Noise Signal ........................................................ 10 2.3 Local Oscillator................................................................ 11 2.4 Block Diagram of Superheterodyne Receiver................................ 11 2.5 Block Diagram of Super Regenerative Receiver............................. 11 4.1 NumberofObservationsRequiredtoEstimatetheStandardMultivariate Normal Density Function such that the Mean Square Error is less than 0.1 23 4.2 PCA of White Gaussian Noise ............................................... 25 4.3 PCA of Square Signals........................................................ 26 4.4 PCA of Triangle Signals ...................................................... 27 4.5 PCA of Sawtooth Signals..................................................... 27 4.6 Percentage of Variation Explained by Each of the Top 10 PCs of Noise, Square Signal, Triangle Signal, and Sawtooth Signal....................... 28 4.7 Experimental Setup for Data Collection..................................... 30 4.8 Application of PCA for UEE detection...................................... 31 4.9 Basic Structure of Hidden Markov Model ................................... 33 4.10 Hidden Markov Model for UEE Identification .............................. 34 4.11 Signal Identification Process.................................................. 35 4.12 Ergodic Markov Chain........................................................ 39 4.13 Support Vector Machine...................................................... 43 4.14 Neural Networks............................................................... 46 5.1 Average Contribution of Top 10 PCs for Devices and Noise ..................... 50 5.2 Principal Components of Noise............................................... 51 5.3 Principal Components of Device 1 at 3 feet................................. 51 ix 5.4 Principal Components of Device 1 at 6 feet................................. 51 5.5 Principal Components of Device 1 at 10 feet................................ 52 5.6 Principal Components of Device 2 at 3 feet................................. 52 5.7 Principal Components of Device 2 at 6 feet................................. 52 5.8 Principal Components of Device 2 at 10 feet................................ 53