high resolution jamming detection in global …
TRANSCRIPT
HIGH RESOLUTION JAMMING DETECTION IN
GLOBAL NAVIGATION SATELLITE SYSTEMS
By
Mohamed Moussa
A thesis submitted to the Department of Electric and Computer Engineering
In conformity with the requirements for the
Degree of Doctor of Philosophy
Queen’s University
Kingston, Ontario, Canada
May, 2015
Copyright ©Mohamed Moussa, 2015
ii
Abstract
Global Positioning System (GPS) is increasingly threatened by interference and
especially jamming. They are substantial threats to the functions that rely on the GPS position
velocity and time solutions. The ultimate objective of this thesis is to enhance GPS receivers’
anti-jamming abilities. Particular focus is given to the detection of Continuous Wave (CW)
jamming signals that are close-by in frequency and in space. To tackle the challenge, two high
resolution signal processing methods are proposed for single antenna receivers and for antenna
array receivers. The first method operates in the frequency domain and targets accurate and
efficient detection and frequency estimation of single and multiple in-band CW jammers that lie
between two FFT frequency lines. This is achieved by utilizing high resolution spectral
estimation that is based on orthogonal search. On the other hand, the second method operates in
the space domain to estimate the Direction of Arrivals (DoA) of the jamming signals. Unlike
conventional DoA estimation methods that have a low resolution and fail to detect multiple
close-by jammers, the proposed method utilizes Fast Orthogonal Search (FOS) for nonlinear
modelling to accurately detect the presence of multiple neighboring jammers and estimate their
DoAs efficiently and accurately.
Investigation of the proposed methods’ performance is conducted on GPS signals
obtained using a Spirent GSS 6700 GPS simulator. The proposed frequency domain method
outperformed FFT in terms of number of jamming signals detected and the accuracy of their
frequencies’ estimation at up to a tenth of the required FFT window size. The proposed spatial
domain method was implemented on antenna array geometries such as Uniform Linear Array
(ULA), Uniform Circular Array (UCA) and optimized UCA. Performance was compared to
predominant DoA estimation methods such as the Classical method and Multiple Signal
iii
Classification (MUSIC). The proposed method yields higher accuracy jammer DoA estimates in
addition to accurately and correctly estimating jammer amplitudes due to the immunity to noise.
The research conducted in this thesis contributes to the enhancement of GPS anti-
jamming systems in both frequency and spatial domains. It increases the robustness of single
antenna receivers and improves theirs anti-jamming ability by providing accurate estimates of
multiple jammers whose frequencies are close-by. Additionally, for multi-antenna systems, the
accurate estimation of the number of multiple close-by jammers and their DoAs significantly
improves the anti-jamming process by limiting erroneous spatial attenuation of GPS signals
arriving from an angle close to the jammer.
iv
Acknowledgments
I would like to offer my sincerest gratitude to my family for their continuous support and
encouragement; for always having faith in me and for pushing me forwards in the hardest of
times throughout this difficult journey.
I would like to express my deepest thanks to my supervisor, Prof. Aboelmagd Noureldin, for his
continuous guidance, scientific, educational and financial support. I would also like to thank my
co-supervisor, Prof. Michael Korenberg for his support, insight and continuous encouragement.
The valuable knowledge and advice that my supervisors constantly provided guided me to reach
where I am today.
An acknowledgment has to be made to Dr. Abdalla Osman who worked closely with me and
offered me his valuable expertise and time in the field of array processing.
I would like to thank my sincerest friend Dr. Mohamed Tamazin to whom I owe an immense
amount of gratitude; he always put my best interests beforehand and was a brother to me.
My colleagues at the NavINST lab are to be thanked for their support, friendship and enjoyable
scientific debates.
From all my heart I thank my wife, who has always been patiently by my side, who was always
pushing me towards success, and offered endless support to our family and our son.
v
Table of Contents
Abstract ......................................................................................................................................................... ii
Acknowledgments ........................................................................................................................................ iv
Table of Contents .......................................................................................................................................... v
List of Figures ............................................................................................................................................ viii
List of Tables ............................................................................................................................................... xi
List of Abbreviations .................................................................................................................................. xii
List of Symbols ........................................................................................................................................... xv
List of Notations ...................................................................................................................................... xviii
Introduction .................................................................................................................................. 1 Chapter 1
1.1 Motivation ........................................................................................................................................... 1
1.2 Previous Work .................................................................................................................................... 3
1.3 Research Objectives ............................................................................................................................ 6
1.4 Research Contributions ....................................................................................................................... 8
1.5 Thesis Outline ................................................................................................................................... 10
GPS System Overview ............................................................................................................... 12 Chapter 2
2.1 Introduction ....................................................................................................................................... 12
2.2 GPS Segments ................................................................................................................................... 12
2.2.1 Space Segment ........................................................................................................................... 13
2.2.2 Control Segment ........................................................................................................................ 14
2.2.3 User Segment ............................................................................................................................. 15
2.3 GPS Signal ........................................................................................................................................ 15
2.4 GPS Observables .............................................................................................................................. 17
2.4.1 Pseudorange Measurements ....................................................................................................... 17
2.4.2 Carrier Phase and Doppler ......................................................................................................... 18
2.5 GPS Receiver .................................................................................................................................... 19
2.5.1 The GPS Antenna....................................................................................................................... 21
2.5.2 RF Front-End ............................................................................................................................. 22
2.6 GPS Errors ........................................................................................................................................ 24
2.6.1 Hardware Clock Error ................................................................................................................ 24
2.6.2 Unintentional Delays.................................................................................................................. 24
2.6.2.1 Ionospheric and Tropospheric Delays ................................................................................. 25
vi
2.6.3 Receiver Noise ........................................................................................................................... 25
2.6.4 Errors Caused by RF Signals in the Receiver’s Vicinity ........................................................... 26
2.6.4.1 Multipath Delay .................................................................................................................. 28
2.7 Interference and Jamming ................................................................................................................. 28
2.7.1 Jamming Attacks on GPS .......................................................................................................... 31
2.7.2 Jamming Effect on GPS ............................................................................................................. 32
2.8 Jamming Detection ........................................................................................................................... 35
2.9 Anti- Jamming Techniques ............................................................................................................... 36
2.9.1 Single Antenna Receiver-Based Solutions................................................................................. 37
2.9.1.1 Front-End-Based Solution ................................................................................................... 37
2.9.1.2 Adaptive Notch Filter (ANF) .............................................................................................. 38
2.9.1.3 Switching Between GNSS Constellations .......................................................................... 38
2.9.1.4 Integrating GNSS with Inertial Navigation Systems (INS) ................................................ 39
2.9.2 Antenna Array Based Solutions ................................................................................................. 39
2.9.2.1 Null Steering ....................................................................................................................... 40
2.9.2.2 Beamforming ...................................................................................................................... 41
2.10 DoA Estimation of Jamming Signals .............................................................................................. 43
2.10.1 Array Data Model .................................................................................................................... 44
2.10.2 Conventional Beamforming ..................................................................................................... 46
2.10.3 Capon’s Beamformer ............................................................................................................... 47
2.10.4 DoA Estimation Using Subspace Techniques .......................................................................... 48
2.10.5 Multiple Signal Classification (MUSIC) ................................................................................. 49
2.10.5.1 Spatial Smoothing ............................................................................................................. 50
2.10.6 Search-Free DoA Estimation Techniques ................................................................................ 51
2.10.7 Root MUSIC ............................................................................................................................ 51
2.10.8 Estimation of Signal Parameters via Rotational Invariance Technique (ESPRIT) .................. 52
Enhanced GPS Narrowband Jamming Detection Using High Resolution Spectral Estimation 56 Chapter 3
3.1 Introduction ....................................................................................................................................... 56
3.2 Continuous Wave Interference: ........................................................................................................ 57
3.2.1 Out -of -Band Signals ................................................................................................................ 57
3.2.2 Impacts of CW Interference ....................................................................................................... 59
3.2.3 Related Work ............................................................................................................................. 61
3.3 Proposed Method .............................................................................................................................. 63
3.4 Jammer Detection ............................................................................................................................. 64
vii
3.4.2 FOS-based Jamming Detection .................................................................................................. 70
3.4.3 Experimental Work .................................................................................................................... 71
3.4.4 Simulation Scenarios.................................................................................................................. 73
3.4.5 Results and Discussions ............................................................................................................. 74
3.5 Conclusion ........................................................................................................................................ 84
DoA Estimation of GPS Narrowband Interference .................................................................... 85 Chapter 4
4.1 Introduction ....................................................................................................................................... 85
4.2 DoA Estimation Using Nonlinear Signal Modeling ......................................................................... 87
4.2.1 Stopping Conditions for FOS Algorithm ................................................................................... 92
4.2.2 FOS complexity: ........................................................................................................................ 93
4.2.3 Candidate Function Selection for FOS Algorithm ..................................................................... 93
4.3 Antenna Array Design ...................................................................................................................... 94
4.3.1 Uniform Linear Array ................................................................................................................ 94
4.3.2 Two-Dimensional Arrays ........................................................................................................... 97
4.3.2.1 Rectangular Arrays ............................................................................................................. 98
4.3.2.2 Circular Array ................................................................................................................... 102
4.4 Array Geometry Optimization ........................................................................................................ 106
4.5 High resolution FOS-based DoA Estimation Using a 2D Circular Array with Central Element ... 112
4.6 Experimental Setup ......................................................................................................................... 114
4.7 Simulation and Results ................................................................................................................... 120
4.7.1 Results for ULA ....................................................................................................................... 121
4.7.1.1 Single Jammer ................................................................................................................... 121
4.7.1.2 Multiple Jamming Signals ................................................................................................ 123
4.7.2 Results for UCA ....................................................................................................................... 126
4.7.2.1 Single Jammer ................................................................................................................... 126
4.7.2.2 Multiple Jamming Signals ................................................................................................ 128
4.7.3 Results of Optimum Array ....................................................................................................... 135
4.7.3.1 Single Jammer ................................................................................................................... 136
4.7.3.2 Multiple Jamming Signals ................................................................................................ 138
4.8 Conclusion ...................................................................................................................................... 146
Conclusions and Recommendations ......................................................................................... 148 Chapter 5
5.1 Conclusions ..................................................................................................................................... 148
5.2 Recommendations and Future Work ............................................................................................... 151
viii
List of Figures
Figure 2-1: Different GPS segments ........................................................................................................... 13
Figure 2-2: GPS satellites in orbit [22] ....................................................................................................... 14
Figure 2-3: Locations of the different sections of the control segment around the world [25]................... 15
Figure 2-4 : GPS L1 signal structure .......................................................................................................... 16
Figure 2-5: Pseudorange measurements [21] .............................................................................................. 18
Figure 2-6: Block diagram of a GPS receiver ............................................................................................. 20
Figure 2-7: Receiver channels in the signal processing phase [29] ............................................................ 20
Figure 2-8: Typical RHCP and LHCP normalized radiation pattern of NovAtel antenna GPS-702 GG for
GPS L1 and L2 Signals [34] ....................................................................................................................... 21
Figure 2-9: Building blocks of the AGC signal conditioning stage ............................................................ 23
Figure 2-10: Distribution of 1 sec of real I and Q data sampled at 2.5 MHz .............................................. 23
Figure 2-11: LoS signal (red) and its reflection (green) in an urban canyon .............................................. 28
Figure 2-12: Absence of LoS component in an urban canyon .................................................................... 28
Figure 2-13: Common interference and jamming sources .......................................................................... 31
Figure 2-14: Different types of PPDs [47] .................................................................................................. 32
Figure 2-15: Tracking results of an experiment after applying swept CW and Gaussian noise jamming. . 34
Figure 2-16: Tracking results of an experiment without applying jamming............................................... 35
Figure 2-17: Classification of anti-jamming techniques ............................................................................. 37
Figure 2-18: A generic nulling antenna [56] ............................................................................................... 42
Figure 2-19: Multiple output beamformer [56] ........................................................................................... 43
Figure 2-20: Different arrival times of satellite signals impinging on ULA ............................................... 44
Figure 2-21: Comparison of Classical and MVDR Beamformers with an 8 degree separation ................. 48
Figure 2-22 : Eigen decomposition of antenna array signals for jamming detection and DoA estimation
[60] .............................................................................................................................................................. 49
Figure 2-23: Comparison between MUSIC, MVDR and the classical beamformer with closely spaced
sources ........................................................................................................................................................ 50
Figure 2-24: The doublets between the two subarrays X and Y, non-overlapping subarrays .................... 53
Figure 2-25: The doublets between the two subarrays X and Y, overlapping subarrays ............................ 53
Figure 3-1: Spectrum describing CW interference [36] .............................................................................. 57
Figure 3-2: Out of band signal being falsely considered as an in band signal [36]. ................................... 58
Figure 3-3: Effective range of a 4 Watt GPS jamming device [66] ............................................................ 59
ix
Figure 3-4: Distribution of a CW signal ..................................................................................................... 60
Figure 3-5 Histogram of a GPS L1 signal before and after being subjected to moderate and high power
CW jamming. .............................................................................................................................................. 60
Figure 3-6: Proposed jamming detection method applied in the pre-correlation stage of the receiver ...... 63
Figure 3-7 Estimated jammer frequency due to FFT (window size = 1 ms) .............................................. 64
Figure 3-8 FFT detection of multiple simultaneous CW jammers ............................................................. 65
Figure 3-9: The Novatel FireHose Frontend [83] ....................................................................................... 72
Figure 3-10: Simulation setup using the Spirent GSS 6700 in the NAVINST lab. .................................... 72
Figure 3-11: Demonstration of the procedure for conducting scenarios ..................................................... 73
Figure 3-12: Detection error of FOS vs FFT at JSR = 0 dB ....................................................................... 75
Figure 3-13: Detection error of the second jammer component. ................................................................ 76
Figure 3-14: Effect of jammer power on FOS detection of three jammers. ................................................ 78
Figure 3-15: Effect of JSR on correct detection ......................................................................................... 79
Figure 3-16: FOS vs FFT jammer detection at different JSR and a 2 ms window size .............................. 81
Figure 3-17: FOS vs FFT jammer detection at different window sizes for 3 jammers at different JSR (A)
.................................................................................................................................................................... 82
Figure 3-18: FOS vs FFT jammer detection at different window sizes for 3 jammers at different JSR (B)
.................................................................................................................................................................... 83
Figure 4-1:Uniform linear array .................................................................................................................. 95
Figure 4-2 Array factor for ULA ................................................................................................................ 97
Figure 4-3 Rectangular array geometry ...................................................................................................... 98
Figure 4-4 Array factor magnitude for rectangular array .......................................................................... 100
Figure 4-5 Array factor pattern for rectangular array ............................................................................... 101
Figure 4-6 XY-Plane view of array factor pattern for rectangular array .................................................. 101
Figure 4-7 Circular array geometry .......................................................................................................... 102
Figure 4-8 Array factor magnitude for circular array ............................................................................... 104
Figure 4-9 Array factor pattern for circular array ..................................................................................... 104
Figure 4-10 XY-plane view of array factor pattern for circular array ...................................................... 105
Figure 4-11 Circular array with central element ....................................................................................... 109
Figure 4-12 Array factor magnitude for circular array with central element ............................................ 110
Figure 4-13: Array factor pattern for circular array with central element ................................................ 111
Figure 4-14 XY-Plane view of array factor pattern for circular array with central element ..................... 112
Figure 4-15: Simulation sequence of an M-element array of GPS antennas ............................................ 115
Figure 4-16: Experimental setup and Skyplot of the experiment. ............................................................ 118
x
Figure 4-17: Acquisition results confirming the utilized satellites after the rest of the available satellites
have been switched off using SimReplay Plus ......................................................................................... 119
Figure 4-18 DoA estimation of one jammer at JSR=15 dB ...................................................................... 122
Figure 4-19: DoA estimation of one jammer at JSR=45 dB ..................................................................... 123
Figure 4-20 DoA estimation of 3 jammers at JSR=15 dB and frequencies of 100Hz, 200Hz and 300Hz 124
Figure 4-21 DoA estimation of 3 jammers arriving at JSR=45 dB and frequencies of 100Hz, 200Hz and
300Hz ........................................................................................................................................................ 125
Figure 4-22 DoA estimation of one jammer using Classical DoA at JSR=45 dB .................................... 126
Figure 4-23: DoA estimation of a single jammer using MUSIC at JSR = 45 dB ..................................... 127
Figure 4-24 DoA estimation of one jammer using FOS at JSR = 45 dB .................................................. 127
Figure 4-25: DoA estimation of 3 closely spaced jammers using MUSIC at JSR = 15 dB ...................... 128
Figure 4-26: DoA estimation of 3 closely spaced jammers using FOS at JSR = 15 dB ........................... 129
Figure 4-27: DoA Estimation of 3 closely spaced jammers using MUSIC at JSR = 45 dB ..................... 130
Figure 4-28: DoA estimation of 3 closely spaced jammers using FOS at JSR = 45 dB ........................... 130
Figure 4-29: DoA estimation of 2 closely spaced jammers using MUSIC at JSR = 15 dB ...................... 132
Figure 4-30: DoA estimation of 2 closely spaced jammers using FOS at JSR = 15 dB ........................... 133
Figure 4-31: DoA estimation of 2 closely spaced jammers using MUSIC at JSR = 45 dB ...................... 134
Figure 4-32: DoA estimation of 2 closely spaced jammers using FOS at JSR = 45 dB ........................... 134
Figure 4-33: DoA estimation of one jammer using MUSIC at JSR=15 dB (optimum array) .................. 136
Figure 4-34: DoA estimation of one jammer using FOS at JSR=15 dB (optimum array) ........................ 137
Figure 4-35 DoA estimation of one jammer using MUSIC at JSR=45 dB (optimum array) .................. 137
Figure 4-36: DoA estimation of one jammer using FOS JSR=45 dB (optimum array) ........................... 138
Figure 4-37: DoA estimation of 3 jammers using MUSIC at JSR=15 dB (optimum array) ..................... 139
Figure 4-38: DoA estimation of 3 jammers using FOS at JSR=15 dB (optimum array) .......................... 139
Figure 4-39: DoA estimation of 3 jammers using MUSIC at JSR=45 dB (optimum array) ..................... 140
Figure 4-40: DoA estimation of 3 jammers using FOS at JSR=45 dB (optimum array) .......................... 141
Figure 4-41: DoA estimation of 2 closely spaced jammers using MUSIC at JSR=15 dB (optimum array)
.................................................................................................................................................................. 142
Figure 4-42 DoA estimation of 2 closely spaced jammers using FOS at JSR=15 dB (optimum array) ... 143
Figure 4-43: DoA estimation of 2 closely spaced jammers using MUSIC at JSR=45 dB (optimum array)
.................................................................................................................................................................. 144
Figure 4-44: DoA estimation of 2 closely spaced jammers using FOS at JSR=45 dB (optimum array) .. 144
xi
List of Tables
Table 2-1: Typical pseudo-range measurement errors for a single –frequency L1 receiver [20] ............... 26
Table 2-2: Classification of interferences based on spectral characteristics ............................................... 29
Table 2-3: Types of RF interference and their probable sources ................................................................ 30
Table 4-1: Settings adopted for data collection using FireHose frontend ................................................. 116
Table 4-2: Location of the array elements used in the experiment ........................................................... 117
Table 4-3: Experimental results for DoA estimation of 5 satellites. ......................................................... 120
Table 4-4: Simulation results of 3 closely spaced jammers at JSR=15dB ................................................ 129
Table 4-5: Simulation results of 3 closely spaced jammers at JSR=45dB ................................................ 131
Table 4-6: Simulation results of 2 closely spaced jammers at JSR=15dB ................................................ 133
Table 4-7: Simulation results of 2 closely spaced jammers at JSR=45dB ................................................ 135
Table 4-8: Simulation results of 3 closely spaced jammers using optimum array at JSR=15dB ............. 140
Table 4-9: Simulation results of 3 closely spaced jammers using optimum array at JSR=45dB ............. 141
Table 4-10: Simulation results of 2 closely spaced jammers using optimum array at JSR=15dB............ 143
Table 4-11: Simulation results of 2 closely spaced jammers using optimum array at JSR=45 dB........... 145
xii
List of Abbreviations
ADC Analog to Digital Converter
AF Array Factor
AGC Automatic Gain Control
AJ Anti-Jamming
AM Amplitude Modulation
ANF Adaptive Notch Filter
CBF Conventional Beamforming
CRPA Controlled Reception Pattern Antenna
CW Continuous Wave
CWI Continuous Wave Interference
DFT Discrete Fourier Transform
DGA Digital GNSS Antenna
DoA Direction Of Arrival
ESPRIT Estimation of Signal Parameters Via Rotational Invariance Technique
FFT Fast Fourier Transform
FIR Finite Impulse Response
FM Frequency Modulation
FOS Fast Orthogonal Search
GAJT GPS Anti-Jamming Technology
GNSS Global Navigation Satellite Systems
xiii
GPS Global Positioning System
GS Gram Schmidt
IF Intermediate Frequency
INS Inertial Navigation Systems
I & Q In-Phase and Quadrature
JSR Jamming to Signal Ratio
LO Local Oscillator
LoS Line of Sight
MCS Master Control Station
MMSE Minimum Mean Square Error
MSE Mean Squared Error
MVDR Minimum Variance Distortionless Response
MUSIC Multiple Signal Classification
NF Notch Filter
NGA National Geospatial-Intelligence Agency
NLoS Non Line of Sight
PLL Phase Locked Loop
PPD Personal Privacy Device
PPS Precise Positioning Service
PRN Pseudo-Random Noise
PVT Position, Velocity and Time
RFI Radio Frequency Interference
RHCP Right Hand Circularly Polarized
xiv
SLC Side Lobe Canceller
SNR Signal to Noise Ratio
SPS Standard Positioning Service
STFT Short Time Fourier Transform
SVN Space Vehicle Number
UCA Uniform Circular Array
ULA Uniform Linear Array
UWB Ultra-Wideband
VGA Variable Gain Amplifier
WGN White Gaussian Noise
xv
List of Symbols
𝑎 Element separation in rectangular array along the x-axis
𝑎𝑚 Weights of the functional expansion of FOS
𝐴𝑖 Amplitude of the 𝑖𝑡ℎ Useful signal
𝑨(𝜣) Steering matrix
𝑏 Element separation in rectangular array along the y-axis
𝐵𝐼𝐹 GPS signal’s bandwidth after the IF stage
𝐵𝑖𝑛𝑡 Bandwidth of interference signal
𝑐𝑖 Spreading code of the 𝑖𝑡ℎ Signal
𝑑 Inter-element spacing in the antenna array
𝑑𝑖 Navigation message
𝑒(𝑛) Modelling error
𝒆𝒏 Vector of spatially uncorrelated noise at antenna elements
𝜀(𝑛) Error term of orthogonal functional expansion of FOS
𝑓𝑐 GPS carrier frequency
𝑓𝑑𝑖 Doppler frequency affecting the 𝑖𝑡ℎ GPS signal
𝑓𝐼𝐹 GPS signal’s center frequency after the IF stage
𝑓𝑖𝑛𝑡 Carrier frequency of interference signal
f0 Fundamental GPS frequency
fL1 Frequency of GPS L1 signal
fL2 Frequency of GPS L2 signal
xvi
fL5 Frequency of GPS L5 signal
f𝑠 Sampling frequency
𝑔𝑚 Set of orthogonal weights of functional expansion of FOS
𝑀 Number of antenna elements in an antenna array
𝑁𝑠 Number of samples in the record
𝑝𝑚(𝑛) Arbitrary set of non-orthogonal candidate functions
𝑃 Number of candidate freqency pairs
𝑃𝑘 Complex amplitude of the 𝑘𝑡ℎ source
𝑃𝑀𝑈𝑆𝐼𝐶(𝜃) Spatial spectrum produced by MUSIC
𝑄𝑚 MSE reduction given by the 𝑚𝑡ℎ Candidate function
𝑸𝑛 Noise subspace
𝑹𝒚𝒚 Spatial covariance matrix of 𝒚
𝑠𝑘(𝑡) GPS signal transmitted from satellite 𝑘
𝐬𝒏 Transmitted GPS signals vector
𝑡 Time
𝑇 Sampling period
𝑤𝑚(𝑛) Set of orthogonal function
𝑦 Received signal
𝒚(𝑛) Functional expansion of the input signal constructed by FOS
𝒚𝒏 Received signal vector at sample 𝑛
xvii
𝜀2̅̅ ̅ Total MSE
𝛼𝑚𝑟 GS coefficients
λ Wavelength of the received signal
∆0 Magnitude of the displacement vector in ESPRIT
𝛥𝑡 The difference between the satellite and receiver times
𝜃𝑘 Direction of arrival of the 𝑘𝑡ℎ Source (elevation)
𝜏𝑙(𝜃𝑘) Propagation delay across the array associated with the 𝑘𝑡ℎ Source
and the 𝑙𝑡ℎ Antenna
𝜏𝑖 Delay introduced by the communication channel for the 𝑖𝑡ℎ gps
signal
𝑛 Noise encountered at the receiver forthe 𝑖𝑡ℎ gps signal
𝜑𝑖 Random phase of the 𝑖𝑡ℎ gps signal
𝜙𝑘 Direction of arrival of the 𝑘𝑡ℎ Source (azimuth)
𝜔𝑚 Digital frequency of the candidate pair
xviii
List of Notations
(. )𝑇 Transpose
(. )𝐻 Complex conjugate transpose
(. )−1 Inverse
(. )̅̅ ̅̅ Average
{. } Set of a certain quantity
(. )∗ Complex-conjugate transpose of quantity in the parenthesis
‖ . ‖ Norm of the quantity
𝑞1∗𝑞2 Correlation between quantity 𝑞1 and quantity 𝑞2
𝐸[ ] Statistical expectation
𝐶𝑚 Correlation between the input and a candidate function 𝑚
𝐶𝑛𝑘 Correlation between the candidate 𝑛 and the observations vector 𝑘
𝐷𝛽1𝛽2 Correlation between two candidate functions 𝛽1 And 𝛽2
1
Chapter 1
Introduction
1.1 Motivation
Nowadays, the world’s increasing dependence on the Global Positioning System (GPS) is
prevalent. GPS, which was once considered a military asset, is now an indispensable tool
upon which governments, industries and consumers depend; and will increasingly
continue to do so, as GPS integration deepens. An example of integration is in consumer
industries such as the mobile industry, where the majority of phones produced, house a
GPS chip. Smart watches are being introduced; they too have built in GPS chips.
GPS provides positioning, navigation and timing solutions for a multitude of applications
in a vast and growing variety of industries. Therefore, a GPS disruption would have a
stronger effect, given the increased dependability on GPS. Due to the distance GPS
signals travel, they are considered weak signals [1]. In addition to the uncontrollable
factors that deteriorate the GPS signal such as attenuation, fading, multipath and Earth’s
atmosphere, GPS is becoming more vulnerable to sabotage or disruption than ever. GPS
receivers are now confronted with jamming and several other interference sources. There
is a substantial outlawed market for personal privacy and GPS jamming devices, which
make them easily accessible. Recently, offenders have been using jammers to prevent
stolen vehicles from being tracked or simply to remain undetected. Equally, personal
privacy devices (PPDs) are becoming cheaper and easier to acquire leading to the
increased threat of GPS outages for the everyday consumer [2,3] thus posing a larger
threat [4,5].
2
As such, there is a crushing need for new and advanced methods to help preserve GPS
service availability and protect GPS signals from being overwhelmed by powerful
jamming signals that deteriorate accuracy and may lead to complete service denial. In
addition, since the majority of GPS receivers are equipped with one antenna, unlike smart
antenna anti-jamming receivers, they are deprived of the benefits of exploiting the spatial
domain in mitigating a jammer regardless of its type. Therefore, an anti-jamming
enhancement that targets this vast majority, and that operates in domains other than the
spatial domain is necessary. Furthermore, given that smart antenna anti-jamming
systems, that spatially mitigate jammers by electronically steering nulls towards them,
must have knowledge the DoAs of the jammers, the accuracy of those DoA estimation
methods is of paramount importance. Thus, erroneous or inaccurate null steering would
result in the attenuation of the GPS signals that are closest to the jammer and
consequently degrade the GPS solution.
The GPS signal is a direct sequence spread spectrum signal, which entails that the
received GPS signal’s power level is below noise levels and that a despreading process in
the receiver’s signal processing unit restores it to pre-spreading levels. Nonetheless, the
jamming signal can effortlessly overwhelm the GPS receiver’s front-end and prevent the
actual signal from being processed. Unfortunately, this is possible due to the fact that the
spreading gain (approximately 30 dB) is not yet applied to the signal. At that time, even
after the spreading gain is applied, the jamming signal is also amplified therefore
preventing further processing of GPS signals. The gain of the amplifier adjusts itself to
the jamming signal, which is more powerful. Consequently, the Analog to Digital
Converter (ADC) samples the jamming signal instead of the much weaker GPS signal
3
[6]. There are several types of interference and jamming signals that vary depending on
the spectral characteristics (whether the signal is narrowband or wideband), modulation
technique used, or if the signal is high power matched spectrum noise.
Jamming deteriorates the positioning solution accuracy, leads to the total loss of lock on
the satellite signals and therefore impairs the positioning availability. It affects different
parts of the receiver in different manners. Initially, the received GPS signal’s signal to
noise ratio (SNR) suffers severe degradation as the Jamming to Signal Ratio (JSR)
increases. The receiver’s front-end erroneously samples the jamming signals, eventually,
code and carrier phase are lost along with frequency lock and finally the navigation
solution. It is difficult for acquisition and tracking algorithms to recover lost GPS data
when jamming occurs [1].
1.2 Previous Work
The focus of this thesis is directed towards CW signals. CW Interference can obstruct and
hinder the operation of several blocks and parameters of a GPS receiver. This effect is
witnessed on synchronization and acquisition steps, pseudorange measurements and
positioning solutions [1,7,8]. The spectral location of the CW interference compared to
the GPS L1 center frequency is of importance; the closeness of the jammer’s carrier
frequency to the GPS L1 center frequency contributes directly to faster degradation in the
carrier to noise ratio (C/N0), largely because of the correlation conducted in the
acquisition module. It was found that a CW interferer whose center frequency was closest
to the L1 frequency recurrently caused false lock [9]. The notch filter (NF) is a frequently
used tool for the mitigation of CW signals in single antenna receivers. Unfortunately, the
excision of the undesired frequencies by the notch filter will also lead to the removal of
4
the GPS signal spectrum found in the excised bandwidth of the notch. The Automatic
Gain Control (AGC) input histogram was used to develop a notch frequency estimation
and interference signals discrimination detection algorithm [8]. Detection metrics that
were established when CW jamming was applied were acquired in the post-processing
stage of the receiver, for example correlator output power, carrier phase indecision and
the variation in correlator output power. Test statistics that observe the operational
reliability of the GPS receiver were recommended based on these metrics [8]. Moreover,
the least mean squares algorithm was used as a base for an adaptive Finite Impulse
Response (FIR) notch filter which was introduced and implemented to adaptively detect,
locate and mitigate narrowband interferences. Z. Jiang proposed an algorithm that
depends on spectral analysis to excise high power interference [10].
Fast Fourier Transform (FFT) is a predominant spectral analysis method that is widely
used in the detection of jammers in the spectral domain. FFT resolution depends on the
selected window size, which has an inversely proportional relation to processing time. If
a jammer’s frequency is located between frequency bins, it will be approximated to the
nearest bin thus generating an estimation error, and if multiple jammers are closely
spaced between two bins, FFT will neither be able to detect their number nor their
frequencies, as they will be detected as one jammer.
Spatial domain detection and mitigation is a mature field, therefore, many DoA
estimation methods have been proposed. Classical beamforming, also known as
conventional beamforming and the Bartlett method [11] is an earlier approach to DoA
estimation, it scans the spatial spectrum using predetermined angular steps in search of
the direction angle that corresponds to the highest power in the spectrum. Regrettably, the
5
spatial spectrum has a resolution threshold, the major limitation attributed to this method
is it inability to form neither narrow nor sharp peaks unless a large number of elements is
used in the array. Consequently, the method suffers from a limited ability to resolve
closely spaced sources [12].
Capon’s beamformer, also known as the Minimum Variance Distortionless Response
(MVDR), attempts to improve the drawbacks associated with the Classical method. It
results in a significant improvement [13]. The output power is minimized while
constraining the gain in the required direction to unity. The MVDR method demands an
additional matrix inversion compared to the classical method; fortunately, it outperforms
the classical method in the majority of the cases. The disadvantages of the MVDR
method are the necessary additional computations and its failure to estimate the DoAs of
highly correlated (coherent) signals [14].
Advances in the field of direction finding lead to the development of high resolution DoA
estimation methods that utilize the signal subspace. By performing Eigen analysis on the
spatial covariance matrix, the signal and noise subspaces are generated. Multiple Signal
Classification (MUSIC) is among the subspace techniques [15]. MUSIC showed that the
steering vectors associated to the received signals are found in the signal subspace. A
search throughout the possible steering vectors is conducted; the ones that are orthogonal
to the noise subspace are designated as desired signals. However, errors arise in real
scenarios since full orthogonality is difficult to achieve due to errors in the estimation of
the noise subspace. The MUSIC spectrum generates a large value when a match between
the generated steering vector and the actual DoA occurs. The disadvantage of MUSIC is
that it is not able to identify DOAs of correlated signals on its own; the received signal
6
must undergo a pre-processing technique called spatial smoothing [16]. In addition, it is
computationally expensive, sensitive to noise and must have prior knowledge of the
signals it is looking for. Several variants of the method have been proposed to improve its
performance [17].
The Estimation of Signal Parameters via Rotational Invariance Technique (ESPRIT)
algorithm was proposed by R. Roy et al. [18]. It is computationally efficient and more
robust if compared to MUSIC as it does not search the entire spectrum but a significant
drawback is the incompatibility with all array geometries, as it was designed for Uniform
Linear Arrays (ULA). The method has been extended to include multi-dimensional arrays
and many versions have been produced [18,19].
1.3 Research Objectives
The ultimate aim of the research is providing an uninterrupted service for GPS users by
limiting the effect that interference and jamming may have on a GPS receiver. Mitigating
these effects is obtained by improving GPS anti-jamming. By significantly improving
the jamming detection capabilities of current GPS receivers, enhanced jamming
mitigation performance is reached. To attain the desired goals, methods that provide high
resolution jamming detection to a multitude of platforms such as single antenna receivers
and multiple antenna array receivers are proposed. They employ different jamming
detection methods that operate in different domains depending on the structure of the
receiver. The proposed methods are pre-correlation methods; which means they detect the
presence of jamming signals prior to the GPS receiver’s acquisition and tracking
modules.
7
The proposed high resolution jamming detection method for single-antenna receivers is
chosen to operate in the frequency domain, where FFT is a predominant detection
method. On the other hand, the methods proposed for multi-antenna systems exploit the
spatial domain processing which provides, in addition to the frequency domain
processing, DoA estimation of jammers; therefore, benefiting the anti-jamming process
and increasing the resilience of the GPS receiver to jamming. High resolution spectral
estimation that is based on the Gramm-Schmidt orthogonalization procedure is utilized
for single antenna systems. As for multi-antenna systems, a high resolution DoA
estimation method is developed for both ULAs as well as planar arrays, (specifically
uniform circular arrays). The proposed methods address the drawbacks of conventional
DoA algorithms, such as the Classical method which is also known as the Bartlett
method, which cannot provide high resolution estimates.
Considering the above, the outlined research objectives are defined:
1. Developing a high-resolution frequency domain jamming detection method to
isolate jammer frequencies in the received GPS signal. The method aims to
achieve the accurate estimation of the frequencies of single in-band jammer
signals with a resolution higher than the predominant FFT, additionally, the
methods aspires to accurately detect the exact number of multiple in-band
jammers surrounding the receiver and accurately estimate their frequencies.
Exploring the benefits of FOS spectral estimation method will be adopted in this
research due to its very high-resolution capabilities.
2. Developing a high resolution DoA estimation method that is capable of accurately
detecting and differentiating between the DoAs of multiple jammers multiple
8
jammers that are located at close-by distances from each other (i.e. spatially close-
by). This method should be implemented to serve both one-dimensional arrays
such as ULA and two-dimensional planar arrays. The benefit of expanding and
applying the method to two-dimensional arrays is obtaining DoA estimates in
both azimuth and elevation. In addition, a study of array geometries is required so
as to choose the most suitable geometry for the application. In this thesis, UCA
and an optimized version of the UCA which is constructed by adding a center
element are studied.
3. This research will conduct a comparative study between the developed method
and other predominant DoA techniques (conventional and subspace based).
Conventional DoA estimation approaches are still being used in DoA estimation.
As of recently, researchers began to apply MUSIC and other subspace based
techniques. This research will develop these predominant methods and compare
their performance to FOS-based DoA.
1.4 Research Contributions
The aforementioned objectives necessitated extensive research, which when completed,
paved the way for several contributions. They are classified according to the above
mentioned objectives. Initially, the problem of jamming detection for the enhancement of
single antenna receivers was approached. The research contributes to the development of
a high resolution jamming detection method that is based on spectral estimation using a
nonlinear modelling method called Fast Orthogonal Search. The method accomplishes
the following:
9
a. accurate estimation of the frequencies of in-band jammer signals with a
resolution that is finer than that of the predominant FFT
b. accurate and efficient estimation of the number of multiple in-band CW
jammers whose frequency separation is lower than the FFT frequency
resolution
c. accurate estimation of the frequencies of multiple in-band CW jammers.
The next phase of the research required the simulation of multi-antenna systems. A
significant hurdle that prevented the commencement of testing and experimentation was
the lack of suitable hardware for this type of research. Therefore a method that allows for
multi antenna GPS simulations using only one Spirent 6700 GPS simulator and one
frontend (called FireHose and provided by Novatel Inc.) was devised, developed and
validated. The developed method paves the way for one-dimensional and two-
dimensional array simulations which in turn lead to the accomplishment of the desired
goals.
The final phase of the research was the study of array geometries, the selection of the
most suitable configurations and the implementation of the simulations using the Sprient
6700 GPS simulator, the FireHose and Matlab for post-processing and performance
analysis and comparison with predominant DoA estimation methods. The proposed high
resolution method contributes to higher accuracy DoA estimation of the multiple spatially
close-by jammers by employing ULA, UCA and optimum UCA. It also contributes to
more accurate determination of jammer amplitude which is useful to jammer localization
systems that utilize signal strength to determine location. The methods directly contribute
to the robustness enhancement of GPS receivers and especially receivers that are not only
10
developed to detect jammers but to mitigate them. Providing a more accurate DoA
estimate minimizes the GPS signal degradation incurred by the inaccuracy of anti-
jamming receivers that rely on less accurate DoA estimation methods.
1.5 Thesis Outline
The thesis presented is structured as follows: Initially in Chapter 2, an overview of the
GPS system components with particular attention to the receiver is given. The chapter
then introduces the different types of interference and jamming followed by a review of
anti-jamming techniques and finally moving onto the predominant DoA estimation
methods found in literature; thus providing sufficient background knowledge for readers
who will move onto Chapter 3. Special interest is given to CW interference and jamming
in Chapter 3 where the proposed spectral estimation based high resolution jamming
detection method is introduced. The jammers used in this chapter are CW jammers. The
signal processing required is explained along the chapter. The effects of JSR levels
variation, window size and the number of jammers are studied and the proposed method
is compared to the predominant FFT. Chapter 4 primarily reviews the necessary concepts
required in order to understand array processing, a review of the different arrays
geometries such as ULA, URA and UCA is given, and it is followed by an optimization
process that leads to the use of the optimized UCA. The chapter continues by
demonstrating the developed method for multi-dimensional GPS antenna array
simulation which is used throughout the chapter. Experimental results achieved by the
proposed DoA method and its comparison to the predominant DoA estimation methods
are presented and discussed when the method is applied to the aforementioned array
11
geometries. The conclusions, recommendations and future work relating to the research’s
outcomes are subsequently presented in Chapter 5.
12
Chapter 2
GPS System Overview
2.1 Introduction
The role of the Global Positioning System (GPS) has become indispensable in our lives.
Because of its significance, research is very active in this sector and obtaining higher
reliability and higher accuracy is essential to sustain the integrity of the position, velocity
and time (PVT) solutions. Initially a military satellite system that was later on provided
partially to the public, GPS has seen an explosive growth in the number of users. There
are also many industry sectors that rely on it such as precise timing, precise positioning
used in geodesy and structural engineering, aviation, land and maritime navigation...etc.
[20]. This chapter starts with a general introduction of the GPS system’s sectors followed
by the GPS observables, the generic structure of the receiver and finally a summary of the
GPS errors which will pave the way towards the main objectives of this research.
Interference and jamming are then overviewed; where their types, causes, sources, effects
and mitigation techniques are described. Additionally, one of the detection and mitigation
techniques; based on array processing, will be explained for its importance and relevance
to Chapter 4.
2.2 GPS Segments
GPS is made up of three segments as shown in Figure 2-1, the first is the space segment
explained briefly in section 1.3.1, the second is the control segment and the last is the
user segment [21].
13
Figure 2-1: Different GPS segments
2.2.1 Space Segment
GPS constellation as shown in Figure 2-2 consists of 24 operational satellites in six
orbital planes with four satellites in each plane. It has been recently upgraded to 27 by the
US air force [22]. The number of satellites can increase up to 30. The ascending nodes of
the orbital planes are separated by 60 degrees and the orbital planes are inclined 55
degrees. The orbit of each GPS satellite is nearly circular with a semi-major axis of
26578 km and a period of about twelve hours. The satellites continuously orient
themselves to ensure that their solar panels stay pointed towards the Sun, and their
antennas point toward the Earth. Each satellite carries four atomic clocks, is the size of a
car and weighs about 1000 kg [23].
14
Figure 2-2: GPS satellites in orbit [22]
2.2.2 Control Segment
The control segment of GPS is made up of a network of tracking stations positioned all
across the Earth as shown in Figure 2-3 with a master control station (MCS) located at
Colorado Springs, Colorado, USA. This MCS is staffed around the clock [24]. There are
16 monitoring stations located throughout the globe. Six of them are operated from Air
Force bases and 10 from the National Geospatial-Intelligence Agency (NGA) [25]. They
are located around the Earth and their positions are known very precisely. These stations
are equipped with very sophisticated GPS receivers and a cesium oscillator for tracking
all of the satellites in view. The MCS remotely operates these ground control stations.
Among the tasks of the control segment are: (a) monitoring and maintaining satellite
orbits and health by maneuvering and relocation, (b) maintaining GPS time, (c)
predicting satellite ephemerides and clock parameters and (d) updating satellite
navigation messages periodically [20].
15
Figure 2-3: Locations of the different sections of the control segment around the world [25]
2.2.3 User Segment
The user segment consists of users (individuals, commercial or military) with different
radio receivers that receive signals from GPS satellites and estimate their position,
velocity, and time. Users of GPS can be classified into civil and military. The user
segment is experiencing exponential growth since inauguration due to decreasing prices
of receiver and the increasing application of GPS in daily life. When a GPS receiver is
activated, it will acquire the GPS satellite signals [26]. The signal is processed, and the
receiver position is determined by the intersection of the spheres created by the satellite
pseudoranges. A minimum of four satellites with good geometry is required. Three
satellites are required for the position fix, while the fourth is required to account for the
receiver clock offset. An in depth look at the user segment can be found in [27] and [28].
2.3 GPS Signal
GPS signals are transmitted on two frequencies in the UHF band (which covers the
frequency band from 500 MHz to 3 GHz) [29]. The two frequencies are referred to as L1
and L2 and are derived from a common frequency, f0=10.23 MHz, there is a third
frequency that is not present in all satellites called the L5, it is only available on block IIF
16
satellites [30] and, till now, only four satellites from this block have been launched with
the operational L5 signal [31]. The GPS frequencies are generated as follows:
Each frequency is modulated by ranging codes called pseudo-random noise (PRN)
sequence which enable precise range measurements. A Coarse Acquisition (C/A) code is
associated with the Standard Positioning Service (SPS) set aside for civil use while a
Precise Code (P-code) is associated with Precise Positioning Service (PPS) (further
encrypted to P(Y)-code for security reasons) and meant for authorized military users. The
L1 frequency is modulated by both C/A and P-codes whereas L2 is modulated by P-code
only. Figure 2-4 depicts the structure of a L1 signal transmitted by the GPS satellite.
Figure 2-4 : GPS L1 signal structure
fL1 = 154 × f0 = 1575.42 MHz Eq. 2-1
fL2 = 120× f0 = 1227.60 MHz Eq. 2-2
fL5 = 115× f0= 1176.45 MHz Eq. 2-3
90° Phase Shift
Antenna
1.023 MHz C/A code
50 Hz Navigation Data
10.23 MHz P(Y) code
50 Hz Navigation
Data
17
It is necessary for the receiver to identify the position of the satellite in order to convert
range measurements into velocity and position by using the navigation message. This
message is superimposed on both L1 and L2 carriers along with PRN codes. The
navigation message is binary coded data consisting of information on the satellite’s
health, ephemeris (position and velocity), clock bias parameters and almanac (which is a
less precise version of ephemeris but concerning the rest of the constellation) [32]. The
GPS signal received from the 𝑖𝑡ℎ satellite is shown in Eq. 2-4
where
𝐴𝑖 is the amplitude of the 𝑖𝑡ℎ useful signal,
𝑓𝑑𝑖 is the Doppler frequency affecting the 𝑖𝑡ℎ GPS signal
𝜑𝑖is the random phase of the signal
𝑓𝑐 is the GPS carrier frequency
𝑐𝑖(𝑡) is the spreading code
𝑑𝑖(𝑡) is the navigation message
𝜏𝑖 is the delay introduced by the communication channel [7].
𝜂 is the noise encountered at the receiver.
2.4 GPS Observables
There are three essential observables which are pseudo-ranges measurements, Doppler
measurements and carrier phase measurements [32]. Carrier phase is more precise than
pseudo-range and is used for applications such as surveying.
2.4.1 Pseudorange Measurements
Pseudorange is the measurement of the distance between the GPS receiver’s antenna and
the GPS satellite’s antenna which is necessary for the calculation of position [21]. It can
𝑦𝑖(𝑡) = 𝐴𝑖𝑐𝑖(𝑡 − 𝜏𝑖)𝑑𝑖(𝑡 − 𝜏𝑖) cos[2𝜋(𝑓𝑐 + 𝑓𝑑𝑖)𝑡 + 𝜑𝑖] + 𝜂 Eq. 2-4
18
be computed from either the C/A code or the P code. The transit time is measured by the
correlation between the C/A-code, received from a GPS satellite, and its replica generated
at the receiver. Since the clocks, installed on the satellite and the receiver, are not
synchronized, the measured range is biased and called a pseudorange [33]. At least four
pseudorange observations are required to estimate the receiver’s three-dimensional
position and clock offset. The pseudo-range is calculated by computing the difference
between the satellite and receiver times and multiplying it with the speed of light [32] as
seen in Figure 2-5.
Figure 2-5: Pseudorange measurements [21]
2.4.2 Carrier Phase and Doppler
A more accurate technique to acquire the range to the satellites can be achieved using
carrier phase measurements. By adding the number of full carrier cycles to the total
number of fractional cycles at both the receiver and the satellite and then multiplying the
sum by the wavelength of the carrier, we get the range. According to [21] , the range of
the accuracy of the L1 signal is approximately 19 cm. The problem with carrier phase
measurements is that the GPS receiver is unable to obtain the number of complete cycles,
but is able to track the phase changes and thus keep the initial cycle ambiguity (initial
number of complete cycles is unknown) unchanged.
Δt
Satellite code
Receiver replica code
19
A GPS receiver can observe the carrier frequency of the received signal by generating a
replica carrier wave. Based on the Doppler effect, the observed frequency will be
different from the L1 signal or any other carrier frequency emitted from the satellites.
This frequency shift is due to the relative motion of the receiver and the satellite and it is
called Doppler measurement [7]. By providing Doppler measurements and knowing the
satellite velocity, the receiver’s velocity can be measured.
2.5 GPS Receiver
GPS receivers process signals transmitted by GPS satellites, estimate the user-to-satellite
ranges and range rates and compute a position, velocity and time solution .As shown in
Figure 2-6, the GPS receiver consists of four blocks: Antenna, Radio Frequency (RF)
front-end, local oscillator and signal processing block. The latter is illustrated in
Figure 2-7 where there are multiple receiver channels that function simultaneously; one
for each received satellite signal. Each receiver channel is responsible for acquiring a
visible satellite, and then the tracking loops track the code and carrier phase of the signal.
Once the tracking is established, the navigation data is extracted and processed to provide
the pseudorange of the specific receiver channel. The receiver channels pseudoranges are
further processed to compute the position of the receiver.
20
Figure 2-6: Block diagram of a GPS receiver
Figure 2-7: Receiver channels in the signal processing phase [29]
RF Front-end
Signal Processing
Local Oscillator
Antenna
Acquisition
Carrier Tracking
Code Tracking
Navigation Data
Extraction
Pseudorange Calculation
21
2.5.1 The GPS Antenna
The antenna is the first element of the receiver architecture. The transmitted signal is
Right Hand Circularly Polarized (RHCP), so the antenna must be designed to receive
RHCP signals. The antenna gain pattern is an important consideration that indicates how
well the antenna performs at different center frequencies, different polarizations and
different elevation angles. Figure 2-8 illustrates the typical gain patterns of the NovAtel
antenna GPS 702GG that can receive GPS L1 and L2 signals.
Figure 2-8: Typical RHCP and LHCP normalized radiation pattern of NovAtel antenna
GPS-702 GG for GPS L1 and L2 Signals [34]
The pre-amplifier is the first active component succeeding the antenna. It is often housed
in the same enclosure as the antenna element. The antenna may be capable of receiving
multiple frequency bands. Thus, there may be one pre-amplifier per frequency band of
interest, or a single pre-amplifier may cover multiple frequency bands. The primary
purpose of the preamplifier is to amplify the signal at the output of the antenna for further
processing [35]. However, it is vital that this amplifier has a very low noise figure; hence
the amplifier is usually referred to as a Low Noise Amplifier (LNA).
22
2.5.2 RF Front-End
RF front-end represents the second stage in the receiver architecture after the antenna and
LNA. RF front-end filters, down converts and digitizes the received signal. The front-end
conditions the narrowband signal in each GPS band and with the help of the local
oscillator and the low pass filters; down converts the signal to the Intermediate Frequency
(IF) .This process is followed by the digitization of the data using the ADC [35]. Initially,
the signal is amplified after which it is filtered such that unwanted out-of-band signals are
removed. The front-end utilizes a crystal along with a Phase Locked Loop (PLL) to
create the Local Oscillator (LO) signal [36]. Filtering in the front-end achieves a number
of objectives such as rejecting out-of-band signals, reducing the noise content in the
received signal and reducing the effect of aliasing. Wide bandwidth signals, if
appropriately processed, can provide higher resolution measurements in the time domain,
but come at the cost of requiring higher sampling rates, and hence result in larger power
consumption in the receiver [35]. Modern front-ends rely on multi-bit ADCs; therefore an
AGC unit is required to ensure that the power of the received signal is adjusted such that
it remains within the dynamic range of the ADC [36]. The AGC consists of three main
parts, the Variable Gain Amplifier (VGA), the gain controller and the ADC as shown in
Figure 2-9.
23
Figure 2-9: Building blocks of the AGC signal conditioning stage
The digital IF signal consists of quantized In-phase and Quadrature (I&Q) samples, they
follow the Gaussian distribution because the GPS signals are spread spectrum signals and
as previously mentioned in section 2.3 the modulating ranging codes are noise-like.
Figure 2-10 demonstrates the Gaussian distribution of one second of real unjammed GPS
I and Q data sampled at 2.5 MHz.
Figure 2-10: Distribution of 1 sec of real I and Q data sampled at 2.5 MHz
-100 -50 0 50 1000
0.5
1
1.5
2
2.5
3
3.5
4x 10
5
In-phase data
-100 -50 0 50 1000
0.5
1
1.5
2
2.5
3
3.5
4x 10
5
Quadrature-phase data
Gain Controller
ADC VGA Analog IF Signal
Digital IF Signal
24
2.6 GPS Errors
GPS receiver uses the trilateration principle to calculate its position by making range
measurement to at least four satellites. Unfortunately, these measurements contain lots of
errors arising from several sources. The effect of these errors must be mitigated so that an
accurate position is obtained.
2.6.1 Hardware Clock Error
Hardware clock errors originate from both the satellite and the receiver. Despite their
high accuracy, satellite clocks, with time, drift away from the GPS system time [21].
Based on the observed satellite clock data from monitoring stations, the control segment
estimates the correction parameters for satellite clocks and transmits them to the
satellites, which in turn broadcast these parameters in the navigation message every 30
minutes. These parameters enable the user to correct the satellite clock error in the
measured range [1]. The clocks in GPS receivers should be inexpensive for affordability.
Consequently they are far less accurate than the satellite clocks and thus contain a bias.
This clock bias error affects all measurements in same manner. Therefore if four pseudo-
range measurements are available, the clock bias can be estimated along with the three
components required to determine the position of the user [24].
2.6.2 Unintentional Delays
The surrounding environment causes another source of delay. In order to receive the
signal at the receiver’s end, the transmitted GPS signal must traverse the ionosphere and
the troposphere layers. These layers add inevitable delay to the signal. Another source of
delay is the reflection of the satellite signals off surfaces before reaching the receiver and
this is termed multipath delay.
25
2.6.2.1 Ionospheric and Tropospheric Delays
Delays caused by the Ionosphere and the Troposphere are introduced and added to other
delays that affect the transit time. They must be modeled and compensated for at the
receiver.
Ionospheric Delay
The ionosphere is the layer of the atmosphere containing ionized gases (free electrons
and ions). It is found approximately from 60 to 1,000 km above the Earth’s surface [28].
Solar activity changes the ionization level of this layer thus affecting the refractive
indices of the various layers of the ionosphere and, as a result, changing the transit time
of a GPS signal and introducing an extra delay that the receivers must remove.
Tropospheric Delay
The troposphere is the lower part of the atmosphere extending from 8 to 40km above the
earth’s surface. It is mainly composed of dry gases and water vapor. The troposphere is
non-dispersive for GPS frequencies, in comparison with the ionosphere, but it is
refractive and causes a decrease in speed relative to free space [28]. Tropospheric errors
are consistent between L1 and L2 carriers and the delay caused by the troposphere has a
dry and a wet component. The wet component is responsible for 10% of the tropospheric
delay and difficult to model as water vapor content varies locally. The dry component is
better modeled which accounts for 90% of the tropospheric delay.
2.6.3 Receiver Noise
Noise is always present in GPS measurements due to component nonlinearity and thermal
noise. Receiver noise cannot be entirely avoided, and its effect on location precision is a
function of the signal-to-noise ratio of the received signal. This error in receiver
26
technology is usually modeled as white noise [37]. A standard GPS receiver performs a
self-test when switched on. Tests for high precision receivers can be performed to
evaluate their noise performance; more on that topic is explained in [21] and [38].
Table 2-1 demonstrates the approximate errors attributed to different factors that affect
the GPS solution accuracy.
Table 2-1: Typical pseudo-range measurement errors for a single –frequency L1 receiver
[20]
Error Source RMS Range Error
Satellite Clock and Ephemeris Parameters 3 m
Atmospheric Propagation Modelling 5 m
Received Noise and Multipath 1 m
2.6.4 Errors Caused by RF Signals in the Receiver’s Vicinity
GPS signal can be classified as a weak signal, as such; it is vulnerable to unintentional
and intentional RF interference signals that limit the receiver’s ability to provide a PVT
solution. Some of the commonly mentioned terms (oriented towards GPS usage) are
defined below [39]:
Background or Thermal Noise. It is the characteristic noise of the system.
Thermal noise is produced mainly in the first amplifier of the receiver.
Interference Noise: is the noise that is present at the same frequency or
frequencies of the desired signal. It can contribute to partially or fully concealing
the GPS signal. Such properties are exploited in AWGN jammers.
Interference Source: it is any electric device which radiates electromagnetic
signals that interfere with the proper operation of the receiver.
27
Unintentional Interferences: is any interference received from a source that does
not have the specific intent of harming GPS operation but unfortunately does so.
Intentional Interferences: they are RF signals that are purposefully generated to
harm and impair the correct functionality of GPS receivers. They comprise
jamming, spoofing and meaconing.
Jamming: is sending certain RF signals in the GPS operating bandwidth, there
are narrowband and wideband jammer.
Spoofing: is transmitting RF signals that mimic GPS signals’ structure while
enclosing faulty navigation data, thus misleading the receiver and causing it to
produce an erroneous PVT solution. Connecting GPS receivers to GPS simulators
can be considered as a harmless type of spoofing.
Meaconing: is the rebroadcasting of intercepted GPS signals at higher power than
the original ones leading to confusion at the receiver. The received GPS signal is
is recorded for a certain time and later re-transmitted. It only needs to be slightly
stronger than the real GPS signal to have the receiver lock on the wrong signal.
The terms of interest throughout this thesis are interference and jamming; they will refer
to unintentional and intentional interference respectively. Interference and jamming are
two of the main degrading factors of the reliability of the GPS solution [40]. Interference
itself can either be caused by signals emitting from different communication systems or
by identical delayed copies of the GPS satellite signal such as multipath which is briefly
explained in the following section.
28
2.6.4.1 Multipath Delay
Multipath is a major contributor of error due to the delay it introduces. The GPS signal
reaches the receiver by one or more different delayed paths; in some cases the Line of
Sight (LoS) signal is not available. The received signal usually comprises direct LoS and
reflected signals from other objects around the receiving antenna named Non Line of
Sight (NLoS) signals. The multipath effect is more evident in urban environments and is
extremely service degrading in urban canyons as shown in Figure 2-11 where one LoS
component is visible and in Figure 2-12 where a LoS component is not available. The
signal that arrives indirectly is delayed and usually leads to a lower SNR. This can cause
a position error in excess of 10 m. Further details are found in [20,28,1]. The multipath
error is two orders of magnitude lower for carrier phase measurements than for pseudo-
range measurements.
Figure 2-11: LoS signal (red) and its
reflection (green) in an urban canyon
Figure 2-12: Absence of LoS component in an
urban canyon
2.7 Interference and Jamming
As mentioned earlier, interference and jamming signals prevent the RF front-end from
functioning correctly. It has been reported in [1] that no matter what tracking and
acquisition algorithms have been implemented in the digital part of the receiver, they will
29
be rendered useless. The latter condition is true if the GPS signals were totally corrupted
by the nonlinear behavior of the front-end. Spectral characteristics such as the carrier
frequency and bandwidth, 𝑓𝑖𝑛𝑡 and 𝐵𝑖𝑛𝑡 respectively, can lead to a broad classification of
interferences. A higher level classification would include in-band and out-of band
signals. The former would encompass any signal whose carrier frequency is close but
outside the GPS band while the latter would encompass signals found inside the GPS
band. A more specific classification for the in-band interferences would differentiate
between the bandwidths of different interference sources and categorize them into
narrowband and wideband interferences. The different classifications are demonstrated in
Table 2-2 where 𝐵𝐼𝐹 and 𝑓𝐼𝐹 correspond to the GPS signal’s bandwidth and center
frequency after the IF stage [41].
Table 2-2: Classification of interferences based on spectral characteristics
Class Spectral Characteristics
Out-of-band interference 𝑓𝑖𝑛𝑡 > 𝑓𝐼𝐹 + 𝐵𝐼𝐹/2 or 𝑓𝑖𝑛𝑡 < 𝑓𝐼𝐹 −
𝐵𝐼𝐹/2
In-band interference (𝑓𝐼𝐹 − 𝐵𝐼𝐹/2 < 𝑓𝑖𝑛𝑡 < 𝑓𝐼𝐹 + 𝐵𝐼𝐹/2)
Narrowband interference 𝐵𝑖𝑛𝑡 ≪ 𝐵𝐼𝐹
Wideband interference 𝐵𝑖𝑛𝑡 ≈ 𝐵𝐼𝐹
Most interference signals are generated by licensed emitters that do not transmit in the L1
band; nonetheless, harmonics of these signals find their way into the GPS L1 band [9]. A
summary of the different types of interference signals based on their modulation types as
30
well as their potential sources is shown in Table 2-3. Some of the most common
interference and jamming sources are demonstrated in Figure 2-13.
Table 2-3: Types of RF interference and their probable sources
Class Type Potential Sources
Narrowband Continuous Wave
(CW)
Intentional CW jammers or near-band
unmodulated transmitter’s carriers
Narrowband Swept CW Intentional swept CW jammers or frequency
modulation (FM) stations transmitters’
harmonics
Narrowband Phase/frequency
modulation
Intentional chirp jammers or harmonics from
an amplitude modulation (AM) radio station,
citizens band (CB) radio, or amateur radio
transmitter
Wideband Pulse Any type of burst transmitters such as radar or
ultra-wideband (UWB)
Wideband Wideband—matched
spectrum
Intentional matched-spectrum jammers,
spoofers, or nearby pseudolites
Wideband Wideband—
phase/frequency
modulation
Television transmitters’ harmonics or near-
band microwave link transmitters overcoming
the front-end filter of a GPS receiver
Wideband Wideband—band-
limited Gaussian
Intentional matched bandwidth noise jammers
31
Figure 2-13: Common interference and jamming sources
2.7.1 Jamming Attacks on GPS
GPS anti-jamming was always considered a priority of the military sector, unfortunately,
due to primarily, the growing demand of personal privacy devices (PPDs),
[5,42,4,43,44,45] jamming is being viewed as a threat to the public; especially with the
increasing dependence on the GPS system. Jamming attacks have a sole objective of
preventing the receiver from providing the desired solution. Although jamming is illegal
and therefore criminalized, it has been reported that offenders have been using jammers
to prevent stolen vehicles from being tracked or simply to remain undetected. Equally,
PPDs are becoming cheaper and easier to acquire leading to the increased threat of GPS
outages for the everyday consumer [2,3]. Therefore, the struggle against jamming has
been moved from the battlefield to the streets. Different types of PPDs are shown in
Figure 2-14 [46].
Spoofers
Broadcasting Stations’ Harmonics
Jammers
Multipath Propagation GPS
receiver
32
Figure 2-14: Different types of PPDs [47]
A GPS jamming signal is typically transmitted on or around the GPS L1 frequency,
which is 1575.42 MHz. It is not necessarily required to be very complex or powerful due
to the fact that the GPS signals are extremely weak. The signal power level of a received
GPS signal at earth’s surface is approximately -163 dBW, which is below the noise floor
of the receiver (estimated at -138 dB for a 20 MHz bandwidth signal located in the L
band). The jamming signal can effortlessly overwhelm the GPS receiver’s front-end and
prevent the actual signal from being processed. Unfortunately, this occurs due to the fact
that the spreading gain (approximately 30 dB) is not yet applied to the signal. At that
time, even after the spreading gain is applied, the jamming signal is also amplified
therefore preventing further processing of GPS signals. The gain of the amplifier adjusts
itself to the jamming signal, which is more powerful. Consequently, the ADC samples the
jamming signal instead of the much weaker GPS signal [6].
2.7.2 Jamming Effect on GPS
Several studies [44,46,5] have explored the effects of jamming and especially PPDs on
Global Navigation Satellite Systems (GNSS) performance. Jamming deteriorates the
positioning solution accuracy, leads to the total loss of lock on the satellite signals and
therefore impairs the positioning availability. It has been reported that the typical front-
end is saturated in case of broadband AWGN signals, high power narrowband and pulsed
33
signals. Contrarily, the front-end was not saturated when low power narrowband,
spoofing and meaconing signals were applied [39].
Several authors have studied the effects of jamming on acquisition and tracking [9,7,48].
Acquisition success is documented in [48]; where it has been identified that CW has the
most damaging effect compared to swept CW and broadband noise which come in
second place such that 10 and 15 dB Jamming to Signal Ratios (JSR) are required to
prevent acquisition respectively. Additionally, applying sufficiently high power jamming
signals will inflict loss of tracking to one or more channels; hence the receiver will go
into reacquisition mode; since tracking requires a carrier to noise ratio (C/N0) that is 10
dB lower than that for acquisition. When tracking loops are affected, the pseudorange
measurements will be erroneous. To understand the effects of jamming experimentally, a
simulation was conducted using the Spirent GSS 8000 in the TECTERRA geomatics lab
in Calgary. The GPS L1 C/A signal was jammed three times, two of which were swept
CW jamming at a jamming to signal ratio (JSR) of 50db. The final jamming incident was
caused by a Gaussian noise jammer of the same JSR (considered wideband) as shown in
Figure 2-15. The jamming effects are clearly shown in the tracking loops of the software
receiver. The results show loss of both carrier phase lock and code phase lock, thus
leading to service denial in the durations of the jamming which start at 4:00 mins, 9:00
mins and 15 mins. By comparing the navigation data bits in Figure 2-15 and Figure 2-16,
it is clear that during the jamming periods (marked by the red arrow); the navigation bits
cannot be distinguished or decoded thus preventing a navigation solution. Furthermore, in
the correlation results, the prompt correlation peaks (in green) should always be higher
than the early and the late correlation peaks. However, during the jamming period, they
34
are superimposed on the early and late peaks, which prohibit the computation of the code
delay. Moreover, both the phase locked loop and the delay locked loop are affected such
that keeping track of carrier phase and code phase becomes difficult. It is also observed
that when jamming signals were turned off, the receiver resumed normal operation.
Figure 2-16 shows the tracking results of a test conducted under normal operating
conditions that lasted for 29 seconds. Jamming signals were not added during this test.
The figure clearly shows navigation bits and the difference between early, late and
prompt correlation peaks.
Figure 2-15: Tracking results of an experiment after applying swept CW and Gaussian
noise jamming.
35
Figure 2-16: Tracking results of an experiment without applying jamming
2.8 Jamming Detection
The PVT solution is compromised when interference and jamming signals are transmitted
on GPS frequencies. Damage is incurred and performance is degraded whether or not it
was originally intended. Nevertheless, knowledge of such an incident is essential to
determine the reliability of the PVT solution. Detecting the presence of interference and
jamming alerts the user about the state of the receiver’s operating conditions. Detection
can either be achieved in the pre-correlation or the post correlation (also referred to as
baseband processing) stages. In the former, the front-end is relied upon to detect the
interference. The AGC is a commonly utilized tool in interference and jamming
detection. Since the signal power is lower than the noise floor, the AGC is accustomed to
the normal operating levels. Jamming signals’ power is regularly higher than the signal
and lead to a change in operating levels and to detection [39]. The gains observed from
the AGC control loop serve as test statistic for detection [49]. Additionally, the
36
distribution of the ADC bins is also used as a statistical measure for interference
detection. It is important to note that multipath and signals sharing GPS structural
characteristics such as spoofing and meaconing can’t be detected by the front-end [39].
An alternate method of detection is achieved by exploiting the transformation of the
received signal to the frequency domain. FFT is commonly utilized and thresholding is
employed to detect the presence of jammers, this topic is researched in detail in Chapter 3
where one of the contributions of this thesis is demonstrated. In the post correlation
methods; based on the studies of the effects of interference and jamming on base band
processing, measurements from the correlators, the PLL and the DLL along with
variations in C/N0 are exploited for RFI detection. A group of test statistics can also be
used as a measure of the presence of RFI such as the correlator’s output power, its
variance and the carrier phase vacillation, which are all unique to each receiver channel.
These test statistics are computed from I and Q sample values obtained after digitization
[49].
2.9 Anti- Jamming Techniques
Successive detection is followed by mitigation if the receivers are equipped for it. A
multitude of techniques have been proposed for anti-jamming and interference
suppression. They can be classified mainly into single antenna based techniques, which
perform jamming mitigation based on receiver processing, whether in pre-correlation or
post-correlation, and antenna array based techniques, which benefit from the presence of
the array to spatially mitigate the jamming signals by electronically controlling the
radiation pattern of the array. Since all the mitigation techniques used by single antenna
systems are present in the receiver, then it is appropriate to imply that they are all
37
applicable in multi-antenna systems. Figure 2-17 depicts a classification of different anti-
jamming approaches from a perspective that focuses on the utilized antenna system. The
classification is explored in the following section.
Figure 2-17: Classification of anti-jamming techniques
2.9.1 Single Antenna Receiver-Based Solutions
2.9.1.1 Front-End-Based Solution
Mitigation solutions in the front-end of the receiver are usually accomplished either by
means of spectral or temporal excision. Mitigation of certain types of interference such as
pulsed interference can be accomplished using a technique called pulse blanking. It is an
appropriate technique aimed at mitigating pulsed interference. Interference excision is
implemented in the time domain on the samples obtained from the ADC that are above a
certain threshold. Alternatively, frequency excision is implemented in the front-end; it is
based on determining the frequencies occupied by the interference signal. A common
method of spectral excision is the Adaptive Notch Filter (ANF) which is explained in the
following section [41].
Anti-jamming Mitigation
Single Antenna Receiver-Based
Front-end-based Adaptive Gain
Control
Adaptive Notch Filter
Spectral Excision
Switching GNSS Constellations
Integrating with INS
Antenna Array-Based
Beam Forming
Null Steering
38
2.9.1.2 Adaptive Notch Filter (ANF)
ANF is a band-stop filter that attenuates a band of frequencies that is very specific and
narrow, namely a notch, in the frequency spectrum. It is usually used for the extraction of
the jammer signal from the received signal. In the case of a continuous wave (CW)
jammer also known as CW interference (CWI), a high power single tone signal is used to
overwhelm the front-end. The filter will remove all frequency components located in the
notch bandwidth which means removing valuable GPS signal components as well; thus
leading to degradation of the signal [43,50]. Using ANF, the undesired signals can be
mitigated by clipping the frequency bins of the jamming signal and transforming the
signal back into the time domain [5]. A cascade of ANFs is studied in [50] for the
detection of multiple CWI signals using the power of the signal to exploit the statistic
value and internal state associated with the filter. The author also acknowledges the fact
that spatial domain mitigation is more effective but hard to implement in mobile devices.
Adaptive notch filtering is also studied and suggested as a countermeasure to jamming
chirp signals. The ability to track the frequency changes in the chirp signal is being
investigated in [42] and carrier-to-Noise density power ratio (C/ N0) degradation is used
as a measure of performance in this analysis.
2.9.1.3 Switching Between GNSS Constellations
Typically, a jammer is used to jam the signals of specific GNSSs and not all systems at
the same time, nevertheless such jammers can be acquired at an expensive cost. A
solution for this problem is to switch to another GNSS when service denial is sensed,
such as GPS, GLONASS, GALILEO and BEIDOU (which has limited coverage).
39
Switching from one GNSS to another, requires more advanced hardware such as multiple
frequency antennas and multi-constellation capable receivers.
2.9.1.4 Integrating GNSS with Inertial Navigation Systems (INS)
Among the proposed solutions that improve the anti-jamming capability [51] is the
integration of INS with GPS [32]. Since INS cannot be spoofed or jammed unlike GPS; it
can be used as a backup in the absence of GPS. Unfortunately, if the jamming continues
for a certain amount of time, the INS will start to drift. Weapon manufacturers are
producing solutions that combine combat grade INS with an anti-jamming unit, such an
example is the integrated GPS anti-jam system by Rockwell Collins [52].
2.9.2 Antenna Array Based Solutions
An antenna array is a group of antennas for which the outputs are combined to create an
overall radiation pattern that is different from that of each antenna element individually
[53]. These antennas are arranged in space depending on the array geometry. They are
processed such that the radiation pattern of the array increases the gain towards a selected
direction while rejecting signals emitted from other directions [54]. Thus, the interference
signal(s) are rejected in space, rather than in the frequency or time domain. Interference
mitigation is significantly enhanced by using an adaptive antenna array (also known as a
smart antenna) through the capabilities of spatial filtering techniques. These techniques
frequently attain high anti-jamming (AJ) margins. They are effective against both narrow
and especially broadband interference signals that single antenna systems cannot combat
[55]. Hence, adaptive arrays improve SNR in the presence of jamming which enables
signal tracking in environments that would otherwise lack code and carrier phase
40
observables. The most common techniques applied in adaptive spatial filtering are null
steering and beamforming.
2.9.2.1 Null Steering
Null steering, also known as side lobe canceller (SLC), is based on the fact that the GPS
signal power is lower than the noise floor. Null Steering considers any signal with a
power level higher than that of the satellite to be an interference signal. Accordingly the
antenna’s beam is steered away from that source and a null is directed towards it. The
objective is to decrease the output power subject to the constraint that one antenna
element is always on (bottom arm of the receiver in Figure 2-18) while trying to maintain
unit gain in all other directions. The rest of the channels have their output phase and
amplitude adjusted in order to block jammers. This is done during the summation process
shown in right side of Figure 2-18. The number of interference signals from different
directions that can be cancelled has a limit that is usually referred to as the degree of
freedom of the array. For example, if there are N antennas in the array, then only up to N-
1 spatially independent interference signals can be rejected [54]. According to [56] an N
element array can steer up to N-1 spatial nulls. Novatel’s GPS Anti-jamming Technology
(GAJT) [57] which is the commercial name of their null steering antenna is an example
of a null steering. It is a seven element controlled reception pattern antenna (CRPA) that
can create up to six nulls in case of jammers. Null steering has been implemented
commercially such as the GAJT [57] mainly due to its lower complexity compared with
beamforming. For example if an M channel L1 receiver ideally has an M beam adaptive
beamformer associated with it; M beams are required for L1. Contrastingly, a null
steering configuration would only require one output for L1.
41
2.9.2.2 Beamforming
Beamforming on the other hand, enhances signal reception by directing beams towards
desired signals. The desired antenna outputs are weighted using different sets of complex
beamforming coefficients. In the case of GPS, multiple beams are required; one for every
satellite. The presence of individual beams that are formed for each satellite as shown in
Figure 2-19, maximizes the SNR. A factor of N improvement is achieved compared to
the SNR that is gained in a non-interference environment when applying the null steering
approach, where N is the number of array elements [55]. Maximizing the power of the
beam achieved by beamforming will also utilize the available degrees of freedom to
shape the spatial null and minimize GPS signal cancellation. Therefore, the probability
that GPS signals will be preserved increases. It is a known disadvantage of null steering
that there will be a probable reduction in the received GPS signal’s energy level and
especially if the signal’s Direction of arrival (DoA) is close to the interferer’s DoA [58].
42
Figure 2-18: A generic nulling antenna [56]
It is desirable that the direction of the satellite be known either through DoA
estimation or from the receiver’s solution. It has been reported that null steering does not
require any prior knowledge of the DoA of the satellite signals [54], although knowledge
of the DoA of the interference source remains a requirement. The advantages gained from
beamforming come at the cost of a more complex receiver that is not yet mainstream and
available to the market. As reported by [59] it could be in the market soon given the
current advances in antenna minimization. A common disadvantage of both methods is
when the GPS signal direction is in line with the interference signal. Consequently,
rejecting such interference will also block the signal from that satellite.
RF Downconversion
A/D
Receiver channel conditionnning
Receiver channel conditionnning
RF Downconversion
Weight Computation
A/D Receiver channel conditionnning
RF Downconversion A/D
Output to receiver
“1”
43
Figure 2-19: Multiple output beamformer [56]
2.10 DoA Estimation of Jamming Signals
This section deals with the application of array processing for the DoA estimation
of the jamming signals. As previously mentioned, whether nullsteering or beamforming
are chosen, the DoA estimates are essential to complete the function of either one. As
such, light must be shed on array processing and DoA estimation techniques. A thorough
review of array processing data model can be found in [11,60,61,62,2] and a quick
summary is given in the following section.
RF Downconversion
A/D
Receiver channel conditionnning
Receiver channel conditionnning
RF Downconversion
Weight Computation
A/D Receiver channel conditionnning
RF Downconversion A/D
“1”
Beam 1
Beam 2
Beam M
Unique Weights For Each Beam
44
2.10.1 Array Data Model
The following data model is based on an example that assumes the presence of S
electromagnetic far-field sources (GPS satellites) of directions 𝚯 where 𝚯 =
[𝜃1, 𝜃2 … , 𝜃𝑆] and 𝜃𝑘 is the angle of arrival of the signal corresponding to 𝑘𝑡ℎ source
such that 𝑘 = 1,2, … , 𝑆 .The satellite signals impinge on an array of 𝑀 GPS antennas
equally separated by distance 𝑑 as shown in Figure 2-20, for simplicity of the diagram;
only the signal from one satellite is displayed.
Figure 2-20: Different arrival times of satellite signals impinging on ULA
The signal received by the 𝑙th antenna at time t, is expressed in the discrete time matrix
notation. Time is referred to by sample number (𝑛) where 𝑛 = 0 is the first sample and
= 0 ,1 … . 𝑇 , where 𝑇 is the entire sampling period. The sampled received signal is
modelled as shown in Eq. 2-5 and is simplified in Eq. 2-6 and Eq. 2-7
. . . .
45
[
𝑦0[𝑛]
𝑦1[𝑛]⋮
𝑦𝑀−1[𝑛]
] = [
𝑎0(𝜃0) 𝑎0(𝜃1) … 𝑎0(𝜃𝑆−1)
𝑎1(𝜃0) ⋱ 𝑎1(𝜃𝑆−1)⋮
𝑎𝑀−1(𝜃0)…
⋮𝑎𝑀−1(𝜃𝑆−1)
] [
𝑠0[𝑛]
𝑠1[𝑛]⋮
𝑠𝑀−1[𝑛]
] + [
𝑒0[𝑛]
𝑒1[𝑛]⋮
𝑒𝑁−1[𝑛]
] Eq. 2-5
𝑦𝑛 = [𝒂(𝜃0) 𝒂(𝜃1) . . . 𝒂(𝜃𝑆−1)] 𝐬𝒏 + 𝐞𝒏 = 𝐀𝐬𝒏 + 𝐞𝒏 Eq. 2-6
𝒚𝒏 = 𝐀(𝚯)𝒔𝒏 + 𝒆𝒏 Eq. 2-7
where 𝑙 = 0 … 𝑀 − 1, 𝑦𝑙[𝑛] is the discrete time received signal at sample 𝑛 for
antenna𝑙, the 𝑆 × 𝑀 matrix 𝐀(𝚯) is called the steering matrix (array response matrix)
where every column corresponds to the steering vector of the signal 𝑠𝑘(𝑡) (a GPS
satellite signal). The vector 𝒆𝒏 represents the uncorrelated noise present at each antenna
element. 𝒚𝒏 , 𝒔𝒏 and 𝒆𝒏 are 𝑀 × 1 vectors where 𝒆𝒏 is considered as zero mean
spatially white noise (i.e. the additive noise is uncorrelated in both time and space
domain) whose variance is σ2 present at the 𝑙𝑡ℎ antenna at time t. The propagation delay
across the array associated with the 𝑘𝑡ℎ source and the 𝑙𝑡ℎ antenna is τl(θk). No delay is
measured at the reference element (τ0(θk) = 0) and the measured delay is maximum at
the last element of the array as shown in Eq. 2-8.
Calculation of the spatial covariance matrix of the input 𝒚𝒏 is a necessary step for DoA
estimation. As per Eq. 2-9 , 𝑹𝒚𝒚 is the spatial covariance matrix where 𝐸[ ] represents the
expectation and 𝒚𝒏𝑯 is the Hermitian (complex conjugate) transpose of 𝒚𝒏.
τM−1(𝜃𝑘) = 𝑀 − 1 × τ1(θk) Eq. 2-8
𝑹𝒚𝒚 = 𝑬 [𝒚𝒏 𝒚𝒏𝑯] Eq. 2-9
46
DoA estimation is a very mature field that affects many communication systems,
as such, a lot of progress has been established. For this reason the following classification
of DoA estimation techniques will be made. The two different categories consist of
conventional DoA estimation and high resolution DoA estimation methods. The former
includes the classical beamforming method and Capon’s beamforming method while the
latter encompasses the Multiple Signal Classification (MUSIC) method along with its
many variants and the Estimation of Signal Parameters via Rotational Invariance
Technique (ESPRIT) which has also generated many variations.
2.10.2 Conventional Beamforming
The conventional beamforming method (CBF) is also known as the delay-and-
sum beamformer and also as the Bartlett method and the Classical method. It is based on
searching the region of interest (angle wise) usually in discrete steps. Once a certain
direction produces the largest output power, then it is estimated that the desired signal is
in that direction. As the direction 𝜃 is varied incrementally across the space of access, the
array response vector 𝐀(𝚯) is calculated and the output power of the beamformer is
measured as per Eq. 2-10
This quantity is also referred to as the spatial spectrum and the estimate of the true DoA
is the angle 𝜃 that corresponds to the peak value of the output power spectrum. The
method is also referred to as Fourier method since it is a natural extension of the classical
Fourier based spectral analysis with different window functions [11]. Unfortunately, the
𝑃𝐶𝑙𝑎𝑠𝑠𝑖𝑐𝑎𝑙(𝜃) =𝐀𝑯(𝚯)𝑅𝒚𝒚𝐀(𝚯)
𝐀𝑯(𝚯)𝐀(𝚯) Eq. 2-10
47
spatial spectrum has a resolution threshold, an array with only a few elements is not able
to form neither narrow nor sharp peaks and thus has a limited ability to resolve closely
spaced signals sources [12] as shown in the comparison in Figure 2-21.
2.10.3 Capon’s Beamformer
Capon’s Beamformer is also known as the Minimum Variance Distortionless
Response (MVDR). The Capon method is an attempt to solve the problem of the poor
resolution that comes with the delay-and-sum method and it results in a significant
improvement [13]. In this method, the output power is minimized with the constraint that
the gain in the desired direction remains unity. Solving this constraint optimization
problem for the weight vector [11] we obtain Eq. 2-11 which is used to obtain the
Capon’s Spatial Spectrum [63] in Eq. 2-12 :
where 𝑅𝒚𝒚−1 is the inverse of the spatial covariance matrix 𝑅𝒚𝒚. The MVDR method
requires an extra matrix inversion but if compared to the CBF, it shows greater resolution
in a majority of the cases. The disadvantages of the MVDR method are the expensive
computations of 𝑅𝒚𝒚 and its failure to locate highly correlated (coherent) source signals
[14]. Figure 2-21 shows a comparison between the Classical and the MVDR
beamformers, it is evident that MVDR confirms the claims that it has higher resolution.
This simulation was conducted with three sources situated at -13,-5, and 40 degrees
𝑤 =𝑅𝒚𝒚
−1𝐀(𝚯)
𝐀𝑯(𝚯)𝑅𝒚𝒚−1𝐀(𝚯)
Eq. 2-11
𝑃𝐶𝑎𝑝𝑜𝑛 = 𝑤𝐻𝑅𝒚𝒚𝑤 =1
𝐀𝑯(𝚯)𝑅𝒚𝒚−1𝐀(𝚯)
Eq. 2-12
48
respectively, noise variance is 1 and the antenna array is a 10 element ULA with
separation of λ/2 where λ is the wavelength of the received signal.
Figure 2-21: Comparison of Classical and MVDR Beamformers with an 8 degree separation
2.10.4 DoA Estimation Using Subspace Techniques
In this section DoA estimators that exploit the signal subspace are explained.
These DoA estimators possess high-resolution estimation capabilities and are also called
high resolution and super resolution DoA estimation techniques. Signal subspace
methods originated in spectral estimation research [64]; where they estimate the
autocovariance of a model that incorporates signal with additive noise. A matrix whose
Eigen structure gives rise to the signal and noise subspaces is generated from the model.
Similarly, this method can be used in DoA estimation of jammers by processing the
spatial covariance matrix as shown in Figure 2-22
49
Figure 2-22 : Eigen decomposition of antenna array signals for jamming detection and DoA
estimation [60]
2.10.5 Multiple Signal Classification (MUSIC)
MUSIC was developed by Schmidt in 1986 [15]. It was shown in details in [60]
that the signal subspace is where the steering vectors corresponding to the incoming
signals are. It was also shown that they are orthogonal to the noise subspace. So, in order
to estimate the DoAs, a search could be conducted through all the possible steering
vectors so that the ones that are orthogonal to the noise subspace are found. If 𝑎(𝜃) is the
steering vector corresponding to one of the incoming signals from the direction 𝜃 and
𝑎𝐻(𝜃) is its conjugate transpose, thus 𝑎𝐻(𝜃)𝑸𝑛 = 0, where 𝑸𝑛 is the noise subspace. In
real life scenarios, 𝑎(𝜃) will not be completely orthogonal to the noise subspace due to
errors in estimating 𝑸𝑛. Eq. 2-13 is referred to as the MUSIC spectrum.
Constructing the Covariance
Matrix
Eigen Decomposition and
Subspace Formation
w0
w1
wM-1
( − 1)
Weight update
Jamming Detection
DoA Estimation
𝑃𝑀𝑈𝑆𝐼𝐶(𝜃) =1
𝐀𝑯(𝚯)𝑸𝑛𝑸𝑛𝐻𝐀(𝚯)
Eq. 2-13
50
The MUSIC spectrum will produce a large value when θ is equal to the DoA of one of
the signals. The MUSIC method is known for outperforming conventional methods,
additionally, it can be used with a variety of array geometries. Figure 2-23 compares
MUSIC to the previously mentioned methods taking into consideration the same
simulation settings used in Figure 2-21.
Figure 2-23: Comparison between MUSIC, MVDR and the classical beamformer with
closely spaced sources
It is clear from Figure 2-23 that MUSIC is the only method that has estimated the
presence of three sources with exceptional DoA estimation for the closely spaced
sources, as the simulation shows that the two close sources are located at −13 and −5.1
degrees
2.10.5.1 Spatial Smoothing
Among the disadvantages of MUSIC is that it is not able to identify DoAs of
correlated signals and is computationally expensive as it requires a search over the
51
function PMUSIC for the peaks. A solution for this problem is found in [16]. It is a
technique named “Spatial Smoothing” it can be introduced to overcome this problem. In
fact, spatial smoothing is essential in a multipath propagation environment and to apply
this technique; the array must be divided up into smaller subarrays. The possibility of
overlapping the subarrays can be considered. By averaging the spatial covariance matrix
of each subarray, one single spatially smoothed covariance matrix can be formed, and is
subsequently introduced to the MUSIC algorithm.
2.10.6 Search-Free DoA Estimation Techniques
Other subspace-based DoA estimation methods that avoid any spectral search,
whose computational complexity is often lower than that of the spectral MUSIC
algorithm have been proposed, which provide a significantly improved performance as
compared to the traditional spectral MUSIC algorithm. Among the most popular
algorithms are root-MUSIC [17], [18] and ESPRIT. However, root-MUSIC and ESPRIT
can be applied only in the case of specific array geometries. In particular, root-MUSIC is
applicable only if the sensors are located on a uniform grid, whereas ESPRIT requires
that the array consists of two identical and identically oriented subarrays.
2.10.7 Root MUSIC
The root MUSIC algorithm was proposed by Barabell, unfortunately, it is only
applicable for Uniform Linear Arrays (ULA). He has shown in [17] that the root MUSIC
algorithm provides improved resolution relative to the ordinary MUSIC method
discussed in section 2.10.5 especially at low SNRs where MUSIC does not perform well.
In root MUSIC, by recalling 𝑃𝑀𝑈𝑆𝐼𝐶(𝜃) =1
𝑎𝐻(𝜃)𝑸𝑛𝑸𝑛𝐻𝑎(𝜃)
and computing 𝑃𝑀𝑈𝑆𝐼𝐶(𝜃)−1
52
and with a short mathematical derivation [60], the polynomial with is the equivalent to
𝑃𝑀𝑈𝑆𝐼𝐶(𝜃)−1 is evaluated on the unit circle. Since the MUSIC spectrum will have 𝑟
peaks, then 𝑃𝑀𝑈𝑆𝐼𝐶(𝜃)−1 will have 𝑟 valleys and thus the denominator will have 𝑟 zeros
on the unit circle. The remaining zeros of thus the denominator will be located away from
the unit circle. The authors of [12] showed that in the absence of noise, thus the
denominator will have roots that are located precisely on the unit circle, but unlike in the
presence of noise; the roots will be located close to the unit circle. The root MUSIC
algorithm reduces the complexity of the estimation process of the DoAs to finding the
roots of a (2𝑀 + 1) order polynomial where M is the number of array elements.
2.10.8 Estimation of Signal Parameters via Rotational Invariance Technique
(ESPRIT)
ESPRIT algorithm was proposed by [18], it is computationally efficient and more
robust if compared to MUSIC as it does not search the entire spectrum. ESPRIT is based
on separating the ULA of M sensors into two identical subarrays consisting of the same
number of sensors. Figure 2-24 shows what is defined as a doublet, it is a matched pair
of sensors with identical displacement ,it should be noted that the number of doublets 𝑚
has a varying range from M/2 in case of non-overlapping as shown in Figure 2-24 to
M-1 in case of fully overlapping as in Figure 2-25. The latter does not necessitate the
presence of two separate arrays. For instance, a ULA consisting of four identical
elements with constant inter-element spacing 𝑑 can be considered as two arrays of three
matched pairs, the first subarray would be made up of the first three elements and the
second would be made up of the last three elements in a manner that the first and the
second elements would form the first pair, the second and the third elements would form
53
the second pair; and the third and the fourth elements would form the third and final pair.
This would be a case of M-1 doublets where the two subarrays are displaced by the
distance d.
Figure 2-24: The doublets between the two subarrays X and Y, non-overlapping subarrays
Figure 2-25: The doublets between the two subarrays X and Y, overlapping subarrays
ESPRIT exploits that the two subarrays are rotationally invariant and employs the
configuration shown in in Figure 2-24 to model the output of the array [11]. Assuming
that the signals induced on the 𝑙𝑡ℎ pair due to a narrowband source in the direction 𝜃 be
denoted by 𝑥𝑚(𝑡) and 𝑦𝑚(𝑡) ; also assuming the phase difference between these two
signals depends on the time travelled by the plane wave from one element of the doublet
to the other given that the distance between the two elements is Δ0 (measured in
wavelengths), Eq. 2-14 is obtained as follows.
𝑦𝑙(𝑡) = 𝑥𝑙(𝑡)𝑒𝑗2𝜋∆0𝑐𝑜𝑠𝜃 Eq. 2-14
54
Note that Δ0 is the magnitude of the displacement vector that sets the reference direction;
and that all angles are measured with reference to this vector. 𝑥(𝑡) and 𝑦(𝑡) are the
signals received by the two 𝐾 element subarrays, as shown below in Eq. 2-15 and
Eq. 2-16
where A is a 𝐾 × 𝑀 matrix whose columns denote the M steering vectors corresponding
to M directional sources associated with the first subarray, 𝚽 is an 𝑀 × 𝑀 diagonal
matrix with its 𝑚𝑡ℎdiagonal element given by Eq. 2-17
and 𝑠(𝑡) denotes M satellite signals induced on a reference element, and 𝒏𝑥(𝑡) and 𝒏𝑦(𝑡)
denote the noise induced on the elements of the two subarrays. By Comparing both the
𝒙(𝑡) and 𝒚(𝑡) equations, the steering vectors corresponding to the M directional sources
associated with the second subarray are given by A𝚽. Let 𝑈𝑥 and 𝑈𝑦 denote two 𝐾 × 𝑀
matrices whose columns represent M eigenvectors corresponding to the largest
eigenvalues of the two array correlation matrices 𝑅𝑥𝑥 and 𝑅𝑦𝑦 respectively. As these two
sets of eigenvectors span the same M-dimensional signal space, as a consequence the two
matrices 𝑈𝑥 and 𝑈𝑦 are related by a unique nonsingular transformation matrix 𝜓, that
is 𝑈𝑥𝜓 = 𝑈𝑦. Similarly, 𝑈𝑥 and 𝑈𝑦 are related to the steering vector matrices 𝐴 and A𝚽
by another unique nonsingular transformation matrix T since the same signal subspace is
𝒙(𝑡) = 𝐴𝑠(𝑡) + 𝑛𝑥(𝑡) Eq. 2-15
𝒚(𝑡) = 𝐴Φ𝑠(𝑡) + 𝑛𝑦(𝑡) Eq. 2-16
𝚽𝑚,𝑚 = 𝑒𝑗2𝜋∆0𝑐𝑜𝑠𝜃𝑚 Eq. 2-17
55
spanned by 𝑈𝑥 = 𝐴𝑇 and 𝑈𝑦 = 𝐴𝚽𝑇. By substituting for 𝑈𝑥 and 𝑈𝑦 and given that 𝐴 is
a full rank matrix, we obtain Eq. 2-18 as shown below.
According to this equation, the eigenvalues of 𝜓 are equal to the diagonal elements of Φ,
and the columns of the matrix T are eigenvectors of 𝜓. This is the core relationship in the
development of the ESPRIT algorithm [18]. It requires an estimate of 𝜓 from the
measurement x(t) and y(t). Moreover, an Eigen decomposition of 𝜓 will provide its
eigenvalues, and by equating them to Φ we get the DoA estimates with 𝑙 = 1, … , 𝑀 as
shown in Eq. 2-19
Many versions of the ESPRIT algorithm were developed by relying on different means of
estimating 𝜓 from the array signal measurements [19] [18].
𝑇𝜓𝑇−1 = 𝚽 Eq. 2-18
𝜃𝑚 = cos−1 {𝐴𝑟𝑔(𝜆𝑚)
2𝜋Δ0} Eq. 2-19
56
Chapter 3
Enhanced GPS Narrowband Jamming Detection Using
High Resolution Spectral Estimation
3.1 Introduction
As discussed in the previous chapter, GPS interference and jamming have a damaging
effect on the proper operation of GPS receivers. Interference signals within the GPS C/A-
code bandwidth are named narrowband interference [40]. There exists a multitude of
solutions that were proposed in order to confront the challenge of accurate positioning in
the presence of RFI and jamming. The initial requirement that enables solving the
problem of jamming is to successfully and effectively detect the presence of the
malicious signal. Interference detection is essential to allow the receiver to activate the
suitable interference suppression methods. Unfortunately, the application of incorrect
interference suppression techniques will not provide anticipated enhancements. Once the
interference has been detected and characterized, an appropriate interference mitigation
algorithm could be applied by the receiver.
Not only does the detection of interferers determine which interference suppression
technique will be utilized; it can also be used to enhance the acquisition process by
making it more robust. The knowledge of the type of jammer, for example whether it is
fixed or swept, enables the improvement of the receiver operating characteristic (ROC)
results when determining a searching algorithm in system-level acquisition or in the cell-
level. [8]
57
3.2 Continuous Wave Interference:
Continuous Wave (CW) signals are one of the most common interference types; they can
be approximated as pure sinusoids. Undeniably, the majority of electronic devices and
communication systems generate CW signals that can be responsible for producing
harmonics in the GNSS bands. For example, CW signals are inadvertently transmitted by
television ground stations in the GPS L1 frequency band [7] causing unintentional
interference. Therefore, CWs constitute a very common type of interference. The spectral
and temporal characteristics of a CW can be summarized as follows. In the time domain,
they can be approximated as pure sinusoids, meanwhile, in the frequency domain; their
spectrum is concentrated around certain frequencies [7]. Consequently, the bandwidth
occupied by CWs is considered narrow, and the signals are narrowband, such as AM or
FM [65]. In Figure 3-1, a single frequency CW jammer is demonstrated.
Figure 3-1: Spectrum describing CW interference [36]
3.2.1 Out -of -Band Signals
In the conversion of an analog signal to the digital domain, the frequency content
becomes limited to between zero and half the sampling frequency (namely, the Nyquist
sampling criterion). Consequently, all signals located outside the sampling range will be
folded into the sampled signal, thus increasing the noise; this is demonstrated in
Figure 3-2. An ideal band-pass filter would have a pass-band before and after which all
Freq.
Band of
interest
Interference Amplitude
fc
58
frequencies are completely rejected, unfortunately, all filters have a certain amount of roll
off band in which considerable increases in attenuation are achieved. The presence of
powerful signals located outside the band of interest but very close to its edges can be
unsafe and cause these signals to creep into the desired band. If the utilized filters are not
sharp enough to block the unwanted signals; the effect of the out of band interference will
be the similar to that of an in band signal with an equivalent received power.
Out of band RFI can be generated by powerful transmitters such as television
broadcasting stations which transmit very high power signals. The creation of relatively
low power harmonics that act as out of band interferences is attributed to flaws in filter
design. Difficulties arise due to the fact that 60dB attenuation might be insufficient if the
transmitter broadcasts at power of 1MW. Therefore, the 60dB attenuation would cause a
transmission of a relatively powerful 1W interferer [36].
Figure 3-2: Out of band signal being falsely considered as an in band signal [36].
Amplitude
Amplitude
Freq. Freq.
Freq. Freq.
Amplitude
Sampling
59
3.2.2 Impacts of CW Interference
Due to its low power when received on earth, the GPS signal is susceptible to jamming
over long distances. Brown et al. examined the effects inflicted upon the GPS acquisition
in a GPS receiver due to a single CW jammer generating a 4-Watt signal [66]. As shown
in Figure 3-3, the results demonstrated that the jammer power was detected at more than
2 kilometers away, C/A-code acquisition was lost at 1.4 km from the jammer; and GPS
tracking was compromised when the jammer was within 1 km of the receiver. They
concluded that airborne users were more affected by a ground-based jammer than the
civilians who were on the ground because they had a direct line-of-sight (LOS) to the
jammer for distances of up to 145 km.
Figure 3-3: Effective range of a 4 Watt GPS jamming device [66]
Figure 3-4 and Figure 3-5 illustrate the distribution of a CW signal and depict how the
gradual increase of the CW jammer’s power affects the distribution of the combined GPS
1 km
1.4 km
2.1 km
Tracking is lost
Acquisition is lost
Jamming power is
detected
60
plus jamming signal and systematically transforms it until it is similar to that of the CW
jammer.
Figure 3-4: Distribution of a CW signal
Figure 3-5 Histogram of a GPS L1 signal before and after being subjected to moderate and
high power CW jamming.
-250 -200 -150 -100 -50 0 50 100 150 200 2500
2
4
6
8
10
12x 10
4
Amplitude of the signal x(t)
Distribution of a high power CW signalP
robabili
ty D
ensity F
untion
-100 -80 -60 -40 -20 0 20 40 60 80 1000
0.5
1
1.5
2
2.5x 10
4
-300 -200 -100 0 100 200 3000
0.5
1
1.5
2
2.5x 10
4
-80 -60 -40 -20 0 20 40 60 800
5
10
15x 10
4
No jamming JSR =5 dB JSR = 20dB
-250 -200 -150 -100 -50 0 50 100 150 200 2500
2
4
6
8
10
12x 10
4
Amplitude of the signal x(t)
Distribution of a high power CW signal
Pro
babili
ty D
ensity
Funtio
n
-250 -200 -150 -100 -50 0 50 100 150 200 2500
2
4
6
8
10
12x 10
4
Amplitude of the signal x(t)
Distribution of a high power CW signal
Pro
babi
lity
Den
sity
Fun
tion
-250 -200 -150 -100 -50 0 50 100 150 200 2500
2
4
6
8
10
12x 10
4
Amplitude of the signal x(t)
Distribution of a high power CW signal
Pro
babi
lity
Den
sity
Fun
tion
61
3.2.3 Related Work
CW Interference can harm the operation of several blocks and parameters of a GPS
receiver. This effect is witnessed on synchronization and acquisition steps, pseudorange
measurements and positioning solutions [1,7,8]
It has been proven that CW interference affects the GPS C/A code signal in a manner that
is mostly visible in the characteristics of the frequency spectrum of the signal [28].
Additionally, in [48] Deshapande has demonstrated that CW interference is responsible
for more impairment of the acquisition process unlike AM/FM interference sources. He
concluded that it distorts the signal spectrum and affects the carrier tracking loop. A
tracking loop will lock onto the frequency of a single tone interference signal given that
the CW power level is significantly high. Therefore, inaccurate carrier phase and Doppler
measurements will be generated. The location of the CW interference relative to the L1
center frequency is also important; if the jammer is sufficiently close to the L1 center
frequency, it is assumed that degradation in the carrier to noise ratio(C/ N0) is fast,
mainly due to the correlation process. A CW interferer whose center frequency was
closest to the L1 frequency repeatedly caused false lock [9]. A commonly used method
for the excision of CW signals is the notch filter (NF). In [65], Montloin demonstrated
that CW removal using NFs introduced an average position error of several tens of
centimeters; it was found that linear phase FIR notch filters were more accurate than
Infinite Impulse Response IIR filters.
A detection algorithm based on notch frequency estimation and interference signals
discrimination method based on AGC input histogram were proposed in [8]. Early efforts
to detect CW jamming for GPS applications were conducted in [49] . Detection metrics
62
have been tested when CW jamming was applied. Such metrics can be obtained in the
post-processing stage of the receiver, for instance correlator output power, carrier phase
indecision and the measured variation in correlator output power. These metrics were
recommended as test statistics to monitor the operational integrity of the GPS receiver
[8]. Additionally, AGC, ADC, and antenna based techniques are utilized in the pre-
processing stage of the receiver to detect RFI [67].
Narrowband CW interference affects the acquisition process of the GNSS receiver. The
study of those effects lead to the development of a descriptive statistical model. The
author presented theoretical results of probability of detection and false-alarm of a GNSS
receiver when CW interference was existent [68]. Additionally, [8] proposed a low
complexity solution that enhances the receiver performance in terms of quantization and
acquisition metrics and contributes to the improvement of consumer-grade GPS receivers
in terms of robustness to interference. Also, an adaptive FIR notch filter based on least
mean squares is proposed and implemented to adaptively detect, locate and mitigate
narrowband interferences. Z. Jiang proposed an algorithm that relied on spectral analysis
to mitigate high power interference [10]. It was concluded that a fixed detection
threshold, as implied in previous studies is insufficient. Therefore, an adaptive detection
threshold that is a function of the standard deviation of the normalized spectrum and the
correlators’ power output was suggested. Related to the same field of detection metrics,
the combination of the AGC levels and the correlators’ output power was exploited to
detect and characterize the RFI but the difference is that the receiver simply relied on a
standard correlator [67].
63
3.3 Proposed Method
Recognizing the importance of CW jamming and understanding the damage it can
cause, an accurate method for the detection and frequency estimation of single and
multiple simultaneous CW jamming signals using high resolution spectral estimation is
proposed. The spectral variant of FOS is presented to improve the detection and
frequency estimation of jammers obtained from FFT. Figure 3-6 depicts the proposed
jamming detection method, after obtaining a digitized signal from the front-end, the
proposed method is applied to the received signal and if a single or multiple jammers is
included in the signal, it is recognized. The method operates in the pre-correlation section
of the receiver.
Figure 3-6: Proposed jamming detection method applied in the pre-correlation stage of the
receiver
Radio Frequenc Frontend
Local
Oscillator
Signal Processing
-
64
3.4 Jammer Detection
This section introduces a new high resolution technique capable of detecting single and
multiple CW jamming signals and accurately estimating their frequencies. The proposed
technique is highly efficient compared to FFT which is widely adopted as a spectral
estimation method for jammer detection. Figure 3-7 demonstrates how a relatively small
window size (1ms) can lead to failure in the accurate estimation of the jamming signal
which is located at 10.2 kHz. FFT would only able to either detect the jammer at 10 or at
11 kHz as shown. The jammer was detected at 10 kHz due to the higher magnitude at that
frequency.
Figure 3-7 Estimated jammer frequency due to FFT (window size = 1 ms)
The proposed method is based on a general purpose modelling technique called FOS.
Several techniques have been proposed for pre-correlation jamming detection [69,70,71].
An Eigenvalue-based jamming detection method is proposed in [69]. This method
provided jamming detection of narrowband as well as wideband signals provided that the
baseband complex signals are sampled at a suitable sampling rate. In [70] jamming
detection is achieved using statistical testing that is based on Short Time Fourier
Transform (STFT) of the received signals. Similarly, time filtering was employed to
detect jammers in [71] followed by a transformation of the signal to the frequency
9.4 9.6 9.8 10 10.2 10.4 10.6 10.8 11 11.20
5
10
window time=0.001sec Window offset of 1 MAX TTA = 3 JSRdB
=0 msethresh=1e-06
Ma
gn
itu
de
Frequency (KHz)
FFT
Jammer
9.4 9.6 9.8 10 10.2 10.4 10.6 10.8 11 11.20
5
10
window time=0.002sec Window offset of 1 MAX TTA = 3 JSRdB
=0 msethresh=1e-06
Ma
gn
itu
de
Frequency (KHz)
FOS
FFT
9.4 9.6 9.8 10 10.2 10.4 10.6 10.8 11 11.20
5
10
window time=0.004sec Window offset of 1 MAX TTA = 3 JSRdB
=0 msethresh=1e-06
Ma
gn
itu
de
Frequency (KHz)
FOS
FFT
9.4 9.6 9.8 10 10.2 10.4 10.6 10.8 11 11.20
5
10
window time=0.005sec Window offset of 1 MAX TTA = 3 JSRdB
=0 msethresh=1e-06
Ma
gn
itu
de
Frequency (KHz)
FOS
FFT
9.4 9.6 9.8 10 10.2 10.4 10.6 10.8 11 11.20
5
10
window time=0.006sec Window offset of 1 MAX TTA = 3 JSRdB
=0 msethresh=1e-06
Ma
gn
itu
de
Frequency (KHz)
FOS
FFT
9.4 9.6 9.8 10 10.2 10.4 10.6 10.8 11 11.20
5
10
window time=0.007sec Window offset of 1 MAX TTA = 3 JSRdB
=0 msethresh=1e-06
Ma
gn
itu
de
Frequency (KHz)
FOS
FFT
9.4 9.6 9.8 10 10.2 10.4 10.6 10.8 11 11.20
5
10
window time=0.008sec Window offset of 1 MAX TTA = 3 JSRdB
=0 msethresh=1e-06
Ma
gn
itu
de
Frequency (KHz)
FOS
FFT
9.4 9.6 9.8 10 10.2 10.4 10.6 10.8 11 11.20
5
10
window time=0.009sec Window offset of 1 MAX TTA = 3 JSRdB
=0 msethresh=1e-06
Ma
gn
itu
de
Frequency (KHz)
FOS
FFT
9.4 9.6 9.8 10 10.2 10.4 10.6 10.8 11 11.20
5
10
window time=0.01sec Window offset of 1 MAX TTA = 3 JSRdB
=0 msethresh=1e-06
Ma
gn
itu
de
Frequency (KHz)
FOS
FFT
Estimated
frequency
Frequency (kHz)
65
domain by means of FFT. Generally, numerous jamming detection methods depend on
spectral analysis methods such as FFT to estimate the frequency content of the signal.
The basic principle of the FFT-based RFI detection is to identify the frequency and
amplitude of the interfering source [72,73]. This technique is limited by FFT spectral
resolution given by:
where f𝑠 is the sampling frequency and 𝑁𝑠 is the number of points in the record. Subject
to the SNR, it has been shown that FOS can achieve frequency resolutions up to 10 times
the frequency resolution of the FFT [74]. From Eq. 3-1, one can conclude that a long
record length yields improved spectral resolution, which is needed to accurately model a
time-series. This imposes a limitation on the FFT resolution; therefore relatively long
data records are required to achieve adequate detection of jamming sources. This results
in potential inaccuracies in the spectral estimation of the CW jammers’ frequencies. This
spectral estimate can degrade significantly in the presence of multiple CW jammers with
contiguous frequencies as clearly shown in Figure 3-8, where FFT will only detect one
jammer at 10 kHz although three jamming signals were applied to the received signal.
Figure 3-8 FFT detection of multiple simultaneous CW jammers
FFT Resolution = f𝑠/𝑁𝑠 Eq. 3-1
9.2 9.4 9.6 9.8 10 10.2 10.4 10.6 10.8 11 11.20
50
100
window time=0.001sec Window offset of 1 MAX TTA = 21 JSRdB
=3 msethresh=1e-05
Mag
nit
ud
e
Frequency (KHz)
FFT
FOS
9.2 9.4 9.6 9.8 10 10.2 10.4 10.6 10.8 11 11.20
50
100
150
window time=0.001sec Window offset of 1 MAX TTA = 21 JSRdB
=6 msethresh=1e-05
Mag
nit
ud
e
Frequency (KHz)
FFT
FOS
9.2 9.4 9.6 9.8 10 10.2 10.4 10.6 10.8 11 11.20
100
200
window time=0.001sec Window offset of 1 MAX TTA = 21 JSRdB
=9 msethresh=1e-05
Mag
nit
ud
e
Frequency (KHz)
FFT
FOS
9.2 9.4 9.6 9.8 10 10.2 10.4 10.6 10.8 11 11.20
100
200
300
window time=0.001sec Window offset of 1 MAX TTA = 21 JSRdB
=12 msethresh=1e-05
Mag
nit
ud
e
Frequency (KHz)
FFT
FOS
9.2 9.4 9.6 9.8 10 10.2 10.4 10.6 10.8 11 11.20
100
200
300
window time=0.001sec Window offset of 1 MAX TTA = 21 JSRdB
=15 msethresh=1e-05
Mag
nit
ud
e
Frequency (KHz)
FFT
FOS
9.2 9.4 9.6 9.8 10 10.2 10.4 10.6 10.8 11 11.20
200
400
window time=0.001sec Window offset of 1 MAX TTA = 21 JSRdB
=18 msethresh=1e-05
Mag
nit
ud
e
Frequency (KHz)
FFT
FOS
9.2 9.4 9.6 9.8 10 10.2 10.4 10.6 10.8 11 11.20
200
400
600
window time=0.001sec Window offset of 1 MAX TTA = 21 JSRdB
=21 msethresh=1e-05
Mag
nit
ud
e
Frequency (KHz)
FFT
FOS
9.2 9.4 9.6 9.8 10 10.2 10.4 10.6 10.8 11 11.20
200
400
window time=0.001sec Window offset of 1 MAX TTA = 21 JSRdB
=24 msethresh=1e-05
Mag
nit
ud
e
Frequency (KHz)
FFT
FOS
9.2 9.4 9.6 9.8 10 10.2 10.4 10.6 10.8 11 11.20
200
400
600
window time=0.001sec Window offset of 1 MAX TTA = 21 JSRdB
=27 msethresh=1e-05
Mag
nit
ud
e
Frequency (KHz)
FFT
Jammers
9.2 9.4 9.6 9.8 10 10.2 10.4 10.6 10.8 11 11.20
500
1000
window time=0.001sec Window offset of 1 MAX TTA = 21 JSRdB
=30 msethresh=1e-05
Mag
nit
ud
eFrequency (KHz)
FFT
FOS
9.4 9.6 9.8 10 10.2 10.4 10.6 10.8 11 11.20
5
10
window time=0.001sec Window offset of 1 MAX TTA = 3 JSRdB
=0 msethresh=1e-06
Ma
gn
itu
de
Frequency (KHz)
FFT
Jammer
9.4 9.6 9.8 10 10.2 10.4 10.6 10.8 11 11.20
5
10
window time=0.002sec Window offset of 1 MAX TTA = 3 JSRdB
=0 msethresh=1e-06
Ma
gn
itu
de
Frequency (KHz)
FOS
FFT
9.4 9.6 9.8 10 10.2 10.4 10.6 10.8 11 11.20
5
10
window time=0.004sec Window offset of 1 MAX TTA = 3 JSRdB
=0 msethresh=1e-06
Ma
gn
itu
de
Frequency (KHz)
FOS
FFT
9.4 9.6 9.8 10 10.2 10.4 10.6 10.8 11 11.20
5
10
window time=0.005sec Window offset of 1 MAX TTA = 3 JSRdB
=0 msethresh=1e-06
Ma
gn
itu
de
Frequency (KHz)
FOS
FFT
9.4 9.6 9.8 10 10.2 10.4 10.6 10.8 11 11.20
5
10
window time=0.006sec Window offset of 1 MAX TTA = 3 JSRdB
=0 msethresh=1e-06
Ma
gn
itu
de
Frequency (KHz)
FOS
FFT
9.4 9.6 9.8 10 10.2 10.4 10.6 10.8 11 11.20
5
10
window time=0.007sec Window offset of 1 MAX TTA = 3 JSRdB
=0 msethresh=1e-06
Ma
gn
itu
de
Frequency (KHz)
FOS
FFT
9.4 9.6 9.8 10 10.2 10.4 10.6 10.8 11 11.20
5
10
window time=0.008sec Window offset of 1 MAX TTA = 3 JSRdB
=0 msethresh=1e-06
Ma
gn
itu
de
Frequency (KHz)
FOS
FFT
9.4 9.6 9.8 10 10.2 10.4 10.6 10.8 11 11.20
5
10
window time=0.009sec Window offset of 1 MAX TTA = 3 JSRdB
=0 msethresh=1e-06
Ma
gn
itu
de
Frequency (KHz)
FOS
FFT
9.4 9.6 9.8 10 10.2 10.4 10.6 10.8 11 11.20
5
10
window time=0.01sec Window offset of 1 MAX TTA = 3 JSRdB
=0 msethresh=1e-06
Ma
gn
itu
de
Frequency (KHz)
FOS
FFT
Estimated
frequency
Frequency (kHz)
66
A solution for the detection of multiple CW jammers was proposed in [71] by cascading
a fixed number of filters to detect multiple jammers. A major limitation presents itself
when the number of jammers is higher than the number of cascaded filters. Furthermore,
the presence of multiple jammers with frequency separation less than FFT resolution has
not yet been addressed. Finally, having the ability to detect low power jammers (i.e., far
from the receiver and potentially approaching it) creates an opportunity for early jammer
warning and detection. Accordingly, the performance of jamming detection based on
spectral estimation can be significantly enhanced by using a robust spectral estimation
method that overcomes the limitations of FFT.
This chapter explores the application of an innovative high resolution spectral estimation
method to pre-correlation frequency based jamming detection. The new method is
introduced to extend the capabilities of pre-correlation detection by substantially
increasing its accuracy when dealing with single jammers; and by enabling the
simultaneous detection of multiple in-band CW jamming signals transmitted at close-by
frequencies.
3.4.1 Application of FOS to Spectral Estimation
FOS is a general purpose modelling technique that can be used for time series analysis
and frequency analysis [75] .It possesses features that can significantly enhance CW
jamming detection based on spectral estimation [76,77,75]. These features include the
remarkably high frequency resolution spectrum of FOS and its noise rejection capabilities
for white and colored noise.
67
FOS constructs a functional expansion of an input, namely 𝒚(𝑛) using an arbitrary set of
non-orthogonal candidate functions. The functional expansion of the input in terms of
the arbitrary candidate functions is given by:
𝒚(𝑛) = ∑ 𝑎𝑚 𝑝𝑚(𝑛) + 𝑒(𝑛)
𝑀
𝑚=0
Eq. 3-2
where 𝑎𝑚 are the weights of the functional expansion where 𝑚 = 0,1 … 𝑀 and 𝑒(𝑛) is
the modelling error. FFT produces a functional expansion of the input using a set of
complex sinusoidal functions that ensure an integer number of cycles in the record length
[78]. Thus, when FFT models a frequency that does not have an integer number of cycles
in the record length, energy is spread into all the other frequencies, which is a
phenomenon known as spectral leakage [79]. By using candidate functions that are non-
orthogonal, FOS is more likely to enable modeling of this frequency between two FFT
bins with a single term [77]. FOS calculations are based on a functional expansion using
orthogonal basis functions such that
𝒚(𝑛) = ∑ 𝑔𝑚𝑤𝑚(𝑛) + 𝜀(𝑛)
𝑀
𝑚=0
Eq. 3-3
where 𝑤𝑚(𝑛) is a set of orthogonal functions derived from the candidate functions
𝑝𝑚(𝑛), 𝑔𝑚is the weight, and 𝜀(𝑛) is an error term. The orthogonal functions are
obtained from the candidate functions 𝑝𝑚(𝑛)using the Gram Schmidt (GS)
orthogonalization algorithm. The GS algorithm relates the orthogonal functions to the
candidate functions through GS coefficients 𝛼𝑚𝑟 given by:
𝛼𝑚𝑟 =𝑝𝑚(𝑛)𝑤𝑟(𝑛)̅̅ ̅̅ ̅̅ ̅̅ ̅̅ ̅̅ ̅̅ ̅̅
𝑤𝑟2̅̅ ̅̅ ̅
Eq. 3-4
68
The GS coefficients mr and the orthogonal weights gm can be found recursively using
the equations [80]:
𝑤0(𝑛) = 𝑝0(𝑛) Eq. 3-5
𝐷𝑚0 = 𝑝𝑚(𝑛)𝑝0(𝑛)̅̅ ̅̅ ̅̅ ̅̅ ̅̅ ̅̅ ̅̅ ̅ Eq. 3-6
𝐷𝑚𝑜 = 𝑝𝑚(𝑛)𝑝𝑟(𝑛)̅̅ ̅̅ ̅̅ ̅̅ ̅̅ ̅̅ ̅̅ ̅ − ∑ 𝛼𝑟𝑖
𝑟−1
𝑖=0
𝐷𝑚𝑖 Eq. 3-7
𝛼𝑚𝑟 =𝑝𝑚(𝑛)𝑤𝑟(𝑛)̅̅ ̅̅ ̅̅ ̅̅ ̅̅ ̅̅ ̅̅ ̅̅
𝑤𝑟2̅̅ ̅̅ ̅
=𝐷𝑚𝑟
𝐷𝑟𝑟 Eq. 3-8
𝐶0 = y(𝑛)𝑝0(𝑛)̅̅ ̅̅ ̅̅ ̅̅ ̅̅ ̅̅ ̅ Eq. 3-9
𝐶𝑚 = y(𝑛)𝑝𝑚(𝑛)̅̅ ̅̅ ̅̅ ̅̅ ̅̅ ̅̅ ̅̅ − ∑ 𝛼𝑚𝑟
𝑚−1
𝑟=0
𝐶𝑟 Eq. 3-10
and
𝑔𝑚 =𝐶𝑚
𝐷𝑚𝑚 Eq. 3-11
Accordingly, FOS skips the point-by-point computation of 𝑤𝑚(𝑛) as it is implicitly
defined by the Gram-Schmidt coefficient 𝛼𝑚𝑟.
In its final stage, FOS recursively computes the original functional expansion’s weights,
𝑎𝑚 (Eq. 3-2), based on the weights of the orthogonal series expansion, 𝑔𝑚. The value of
𝑎𝑚 can be found recursively using
𝑎𝑚 = ∑ 𝑔𝑖
𝑀
𝑖=𝑚
𝑣𝑖 , 𝑣𝑚 = 1 Eq. 3-12
where,
69
𝑣𝑖 = − ∑ 𝛼𝑖𝑟
𝑖−1
𝑟=𝑚𝑣𝑟 𝑖 = 𝑚 + 1, 𝑚 + 2, … , 𝑀 Eq. 3-13
From the previous equations, it is observed that FOS requires the calculation of the
correlation between the candidate functions as shown in Eq. 3-6 and Eq. 3-7 and the
calculation of the correlation between the input and the candidate functions as shown in
Eq. 3-9 and 3-10. The correlation between the input and the candidate function
𝑦(𝑛)𝑝𝑚(𝑛)̅̅ ̅̅ ̅̅ ̅̅ ̅̅ ̅̅ ̅̅ ̅ is normally calculated point-by-point once at the start of the algorithm and
then stored for later quick retrieval.
The MSE of the orthogonal function expansion is demonstrated as [75]:
𝜀2̅̅ ̅ = 𝑦2(𝑛)̅̅ ̅̅ ̅̅ ̅̅ − ∑ 𝑔𝑚2
𝑀
𝑚=0𝑤𝑚
2(𝑛)̅̅ ̅̅ ̅̅ ̅̅ ̅̅ Eq. 3-14
Subsequently, the MSE reduction attributed to the 𝑚𝑡ℎ candidate function is given by:
𝑄𝑚 = 𝑔𝑚2𝑤𝑚
2(𝑛)̅̅ ̅̅ ̅̅ ̅̅ ̅̅ = 𝑔𝑚2𝐷𝑚𝑚 Eq. 3-15
FOS can fit a model with a small number of model terms; this is achieved by fitting terms
that minimize the mean squared error (MSE) in order of their significance. There are
three conditions that are responsible for stopping the search algorithm. If any of them
becomes true the search is halted. If a predefined maximum number of terms to be fitted
is reached, the search stops. In addition, if the ratio of MSE to the mean squared value of
the input signal is less than a pre-determined limit, the search is stopped. Finally, if the
addition of an extra term to the model would yield an MSE reduction that is not greater
than that expected if the current residual 𝜀(𝑛) were white Gaussian noise, the search will
stop. Spectral analysis with FOS is accomplished by selecting candidates 𝑝𝑚(𝑛)that are
70
pairs of sine and cosine terms at each of the frequencies of interest. The candidate
functions 𝑝𝑚(𝑛)are given by:
𝑝2𝑚(𝑛) = cos(𝜔𝑚𝑛)
𝑝2𝑚+1(𝑛) = sin (𝜔𝑚𝑛)
Eq. 3-16
Where = 1, … , 𝑃 , 𝜔𝑚 is the digital frequency of the candidate pair while 𝑃 is the number
of candidate pairs. When a sinusoidal pair (sine and cosine pair) is fitted at each
candidate frequency, the magnitude and phase at that candidate frequency is computed
[81]. Two noteworthy differences exist between FOS and conventional Fourier transform
techniques (i.e. discrete Fourier transform (DFT) or FFT) [74,77,80]. The first difference
is that FOS produces a sinusoidal series representation that is more efficient and frugal
(selects fewer components) by choosing the most significant sinusoidal components first.
Meanwhile, the second difference is that the frequencies of the selected sinusoids
shouldn’t necessarily be commensurate nor integer multiples of the fundamental
frequency corresponding to the record length [80]. Therefore, finer frequency resolution
in the spectral model is achieved.
3.4.2 FOS-based Jamming Detection
In this research FOS is employed for narrowband CW jamming detection. As FOS is
generally known to be a data dependent algorithm [76,82,74,77]; the accuracy of the
model produced by FOS depends on the data record being modelled, the candidate
functions being used to compute correlations, and the stopping conditions (thresholds) in
the algorithm. The chosen FOS candidate frequencies have a finer resolution compared to
FFT in order to achieve better detection accuracy. Candidate frequencies can be selected
so that the candidate functions focus on a particular frequency range of interest. For
71
example, the candidates can be spaced with high resolution on a range of interest and
outside the range of interest; the candidates can be spaced by FFT resolution intervals.
It is desirable to have the minimum number of candidate frequencies in the spectral
estimate which represent the most significant components of received signal. However,
creating a model that incorporates fewer than necessary terms will result in an inaccurate
representation of the received signal. On the other hand, the excessive inclusion of terms
will add noise terms into the received signal’s spectral estimate as well as increase the
computation time.
FOS stops modeling when the addition of a new frequency pair does not increase the
MSE reduction more than the reduction obtained from fitting white Gaussian noise
(WGN). Thus, a candidate acceptance threshold, requiring a frequency pair to fit a
minimum percentage of the overall energy in the signal, is set. Such a threshold allows
FOS to reject frequency terms that model the noise.
3.4.3 Experimental Work
The proposed narrowband CW jamming detection method is assessed using simulated
data records. The main simulated data included raw digitized GPS and jamming signals.
GPS simulated data was obtained using Spirent GSS 6700 GPS simulator and Novatel
FireHose Digital Frontend. The jamming signals were simulated using Matlab and added
to the recorded GPS data sets.
The Spirent GSS 6700 GPS L1 simulator was employed to simulate the GPS signals
based on scenarios created using Spirent’s SimGEN platform and controlled by the
SimReplay Plus software. These signals were used in all the conducted experiments.
Subsequently, the output of the simulator is a raw RF GPS signal that is fed to the
72
Novatel FireHose frontend [83]. The FireHose is responsible for digitizing and down-
converting the GPS signal (obtained from the GSS 6700) into a baseband sampled In-
phase/Quadrature (I/Q) signal. The FireHose, shown below in Figure 3-9, is a dual
frequency L1/L2 software receiver and frontend designed for software receiver and
interference detection systems. The FireHose is patented by NovAtel and is not available
for commercial use. It encloses a NovAtel dual frequency OEMV-1DF GNSS receiver,
along with a GNSS frontend processor which is based on a Smart Fusion
FPGA/Microprocessor chip. Samples are provided for L1 and L2 frequency bands at
sampling frequencies that span from 2.5 to 20 MHz. The experimental setup in the
NavINST lab at the Royal Military College is shown in Figure 3-10 , it consists of the
Spirent GSS 6700 connected to the FireHose frontend. The simulated GPS data was
digitized and recorded using the FireHose Frontend at a sampling frequency of 2.5 MHz.
Data extracted from the FireHose together with the simulated jamming signals was
processed using MATLAB.
Figure 3-9: The Novatel FireHose Frontend Figure 3-10: Simulation setup using the
Spirent GSS 6700 in the NAVINST lab.
73
3.4.4 Simulation Scenarios
All scenarios were conducted according to the procedure demonstrated in
Figure 3-11.The proposed technique was examined through 5 simulation scenarios. The
first scenario studied the case of single CW jammer using simulated single tone CW
jammer signal at frequency 10.2 kHz. The performance of the proposed method based
versus the FFT based one was examined at different sizes of the data records (i.e. window
size) and variable JSR’s. In the second scenario, two CW Jammers at equal power were
added to GPS simulated data. In this scenario, the window size was maintained constant
at 1 ms which is the length of the L1 C/A code while JSR was varied. The third scenario
was similar to the second with a slight variation of adding a third CW jammer. In
Scenario 5 the effects of increasing the window size to 2 ms were studied. Finally, in
scenario 6, the case of three jammers of unequal power was studied at different window
sizes.
Figure 3-11: Demonstration of the procedure for conducting scenarios
G
U G
6700
G
b
74
3.4.5 Results and Discussions
In the first scenario, window sizes were varied from 1 ms to 20 ms and JSR was
increased from 0 dB to 40 dB. The obtained results did not show significant variation at
different JSRs. The results displayed in Figure 3-12 correspond to a JSR of 0 dB.
Frequency detection error was approximately zero for FOS and it reached 200 Hz in case
of FFT.
Figure 3-12 also illustrates the decrease in frequency detection error for FFT as window
sizes increased. On the other hand, FOS preserved a steady performance among the
different widow sizes. The cases of 5, 10, 15 and 20 ms windows differ from other
window sizes. The FFT resolutions corresponding to the aforementioned windows are
200, 100, 66.67 and 50 Hz. The error demonstrated by FFT occurred due to spectral
leakage when the jammer frequencies were not aligned with the frequency bins. On the
other hand, the leakage disappeared when the bins located at multiples of the window
sizes coincided with the jammer frequency of 10.2 kHz.
75
Figure 3-12: Detection error of FOS vs FFT at JSR = 0 dB
The effect of different values of JSR on detection accuracy was studied in the second
scenario. Two jammers having frequencies located at 10.2 kHz and 10.3 kHz, whose
difference is lower than the FFT window size of 1 kHz were introduced in scenario 2.
The window size was maintained constant and JSR of each source was varied from 0 dB
to 15 dB. As shown in Figure 3-13, FOS successfully detected both jammers; frequency
estimation was accurate for one jammer; as for the second, estimation was slightly
erroneous until the JSR of each jammer was increased to 13 dB. It is clear from this
experiment that increasing the number of jammers while maintaining the same window
size poses a challenge in detection accuracy at lower JSR. When FFT was employed for
detection in this scenario, only one signal was detected at 10 kHz thus highlighting the
substantial advantage of using FOS over FFT.
0 2 4 6 8 10 12 14 16 18 200
20
40
60
80
100
120
140
160
180
200
FOS vs FFT jammer detection error in Hz at JSRdB
=0dB with MSE=1e-05
window size in milliseconds
Err
or
in d
ete
cti
on
(H
z)
FOS 0dB
FFT 0dB
Window size (ms)
76
Figure 3-13: Detection error of the second jammer component.
The third scenario studied the effect of adding another closely spaced jammer totaling the
number of jammers to 3 at frequencies 10.2, 10.3 and 10.4 kHz. In this scenario, the
window size was kept constant at 1 ms and JSR was varied from 0 dB to 30 dB per
jammer. Figure 3-14 shows the effect of a weak jammer signal on accurate detection. At
first, only two signals are correctly detected out of three within the desired FFT
resolution of 1 kHz, although other higher magnitude frequency components exist outside
the search spectrum. Increasing the JSR minimizes the error in frequency estimation for
the two visible components. This trend continues until the JSR of each signal is 11 dB.
Figure 3-15 represents a superimposition of the individual subplots that are found in
Figure 3-14. It illustrates the effect of increasing combined JSR from 0 to 90 dB. The
0 5 10 15 20 25 300
0.02
0.04
0.06
0.08
0.1
0.12
0.14
0.16
0.18
0.2
Combined JSR of the two equal power jammers(dB)
Magnitude e
rror
in F
OS
dete
ction (
KH
z)
first component
second component
Mag
nit
ud
e er
ror
in F
OS
det
ecti
on
(H
z)
77
highest magnitude was observed using FFT at the 10 kHz frequency; it corresponds to the
final iteration where the combined JSR was 90 dB. At that iteration, FFT yielded higher
magnitudes due to spectral leakage. When all three jammers were not powerful enough,
the proposed method produced erroneous frequency estimates. The low magnitude
components highlighted in green correspond to iterations where JSR values were small,
and subsequently insufficient for correct frequency estimation. The error in frequency
estimation is attributed to the low JSR.
78
Figure 3-14: Effect of jammer power on FOS detection of three jammers.
9.2 9.4 9.6 9.8 10 10.2 10.4 10.6 10.8 11 11.20
20
40
60
window time=1 msec JSRdB
=0 M
ag
nit
ud
e
Frequency (KHz)
FFT
FOS
9.2 9.4 9.6 9.8 10 10.2 10.4 10.6 10.8 11 11.20
50
100
window time=1 msec JSRdB
=3
Mag
nit
ud
e
Frequency (KHz)
FFT
FOS
9.2 9.4 9.6 9.8 10 10.2 10.4 10.6 10.8 11 11.20
50
100
150
window time=1 msec JSRdB
=6
Mag
nit
ud
e
Frequency (KHz)
FFT
FOS
9.2 9.4 9.6 9.8 10 10.2 10.4 10.6 10.8 11 11.20
100
200
window time= 1 msec JSRdB
=9
Mag
nit
ud
e
Frequency (KHz)
FFT
FOS
9.2 9.4 9.6 9.8 10 10.2 10.4 10.6 10.8 11 11.20
100
200
300
window time=1 msec JSRdB
=12
Mag
nit
ud
e
Frequency (KHz)
FFT
FOS
9.2 9.4 9.6 9.8 10 10.2 10.4 10.6 10.8 11 11.20
100
200
300
window time=1 msec JSRdB
=15
Mag
nit
ud
e
Frequency (KHz)
FFT
FOS
9.2 9.4 9.6 9.8 10 10.2 10.4 10.6 10.8 11 11.20
200
400
window time=1 msec JSRdB
=18
Mag
nit
ud
e
Frequency (KHz)
FFT
FOS
9.2 9.4 9.6 9.8 10 10.2 10.4 10.6 10.8 11 11.20
200
400
600
window time= 1 msec JSRdB
=21
Mag
nit
ud
e
Frequency (KHz)
FFT
FOS
9.2 9.4 9.6 9.8 10 10.2 10.4 10.6 10.8 11 11.20
200
400
window time= 1 msec JSRdB
=24
Mag
nit
ud
e
Frequency (KHz)
FFT
FOS
9.2 9.4 9.6 9.8 10 10.2 10.4 10.6 10.8 11 11.20
200
400
600
window time=1 msec JSRdB
=27
Mag
nit
ud
e
Frequency (KHz)
FFT
FOS
79
Figure 3-15: Effect of JSR on correct detection.
Scenario 4 bares a similarity to scenario 3; although window size was increased to 2 ms thus
increasing the FFT resolution from 1 kHz to 0.5 kHz. JSR was progressively increased from 0 to
15 dB as shown in Figure 3-16. The FOS-based method enabled precise detection of the 3
jammers at 0 dB JSR. Figure 3-16 also shows that although FFT resolution was increased to 0.5
kHz, FFT was incapable of detecting all three jammers. This suggests that FOS detection
performance can be enhanced by increasing data record length but at the same time this will
increase the computational load of the jamming detection. In scenario 5, three jammers were
added at frequencies 10.2, 10.3 and 10.4 kHz and two cases were studied. In this scenario each
jammer had a different power level, as shown in Figure 3-17 and Figure 3-18. Additionally, the
window sizes ranged from 1 to 10 ms. The JSR attributed to jammers 1, 2 and 3 in case (A) was
0, 10 and 15 dB respectively as shown in Figure 3-17. For case (B) JSR was 10, 0 and 15 dB as
shown in Figure 3-18. Case (A) results showed that using a 1 ms window, only two jammers
9.2 9.4 9.6 9.8 10 10.2 10.4 10.6 10.8 11 11.20
100
200
300
400
500
600
700
800
Mag
nit
ud
e
Frequency (kHz)
FFT
FOS
80
were detected out of three. Contrarily, case (B) results detected three jammers accompanied by a
frequency estimation error of 200 Hz for the first component. It can be established that accurate
results are achieved using FOS for both cases (A and B) after increasing the window size to a
value of 2 ms and above. Furthermore, FOS detection performance remained constant after 2 ms,
therefore eliminating the need for larger window sizes. FFT performance became comparable to
FOS from a frequency estimation perspective after increasing the FFT window size to 10 ms.
However, FOS outperformed FFT in terms of the accuracy of the detected components
amplitudes.
81
Figure 3-16: FOS vs FFT jammer detection at different JSR and a 2 ms window size
9.2 9.4 9.6 9.8 10 10.2 10.4 10.6 10.8 11 11.20
5
10
15
20
25
30
window size= 2 msec JSR 1 dB
=0 JSR 2 dB
=0 JSR 3 dB
=0
Magnitude
Frequency (KHz)
FFT
FOS
9.2 9.4 9.6 9.8 10 10.2 10.4 10.6 10.8 11 11.20
5
10
15
20
25
30
35
40
45
window size= 2 msec JSR 1 dB
=5 JSR 2 dB
=5 JSR 3 dB
=5
Magnitude
Frequency (KHz)
FFT
FOS
9.2 9.4 9.6 9.8 10 10.2 10.4 10.6 10.8 11 11.20
10
20
30
40
50
60
70
window size= 2 msec JSR 1 dB
=10 JSR 2 dB
=10 JSR 3 dB
=10
Magnitude
Frequency (KHz)
FFT
FOS
9.2 9.4 9.6 9.8 10 10.2 10.4 10.6 10.8 11 11.20
20
40
60
80
100
120
window size= 2 msec JSR 1 dB
=15 JSR 2 dB
=15 JSR 3 dB
=15
Magnitude
Frequency (KHz)
FFT
FOS
82
Figure 3-17: FOS vs FFT jammer detection at different window sizes for 3 jammers at different JSR (A)
9.2 9.4 9.6 9.8 10 10.2 10.4 10.6 10.8 11 11.20
50
100
150
window size=1 ms JSR 1 dB
=0 JSR 2 dB
=10 JSR 3 dB
=15
Magnitude
Frequency (kHz)
FFT
FOS
9.2 9.4 9.6 9.8 10 10.2 10.4 10.6 10.8 11 11.20
50
100
150
window size=2 ms JSR 1 dB
=0 JSR 2 dB
=10 JSR 3 dB
=15
Magnitude
Frequency (kHz)
FFT
FOS
9.2 9.4 9.6 9.8 10 10.2 10.4 10.6 10.8 11 11.20
50
100
150
window size=4 ms JSR 1 dB
=0 JSR 2 dB
=10 JSR 3 dB
=15
Magnitude
Frequency (kHz)
FFT
FOS
9.2 9.4 9.6 9.8 10 10.2 10.4 10.6 10.8 11 11.20
50
100
150
window size=5 ms JSR 1 dB
=0 JSR 2 dB
=10 JSR 3 dB
=15
Magnitude
Frequency (kHz)
FFT
FOS
9.2 9.4 9.6 9.8 10 10.2 10.4 10.6 10.8 11 11.20
50
100
150
window size=10 ms JSR 1 dB
=0 JSR 2 dB
=10 JSR 3 dB
=15
Magnitude
Frequency (kHz)
FFT
FOS
83
Figure 3-18: FOS vs FFT jammer detection at different window sizes for 3 jammers at different JSR (B)
9.2 9.4 9.6 9.8 10 10.2 10.4 10.6 10.8 11 11.20
50
100
150
window size=1 ms JSR 1 dB
=10 JSR 2 dB
=0 JSR 3 dB
=15
Magnitude
Frequency (kHz)
FFT
FOS
9.2 9.4 9.6 9.8 10 10.2 10.4 10.6 10.8 11 11.20
50
100
150
window size=2 ms JSR 1 dB
=10 JSR 2 dB
=0 JSR 3 dB
=15
Magnitude
Frequency (kHz)
FFT
FOS
9.2 9.4 9.6 9.8 10 10.2 10.4 10.6 10.8 11 11.20
50
100
150
window size= 4 ms JSR 1 dB
=10 JSR 2 dB
=0 JSR 3 dB
=15
Magnitude
Frequency (kHz)
FFT
FOS
9.2 9.4 9.6 9.8 10 10.2 10.4 10.6 10.8 11 11.20
50
100
150
window size=5 ms JSR 1 dB
=10 JSR 2 dB
=0 JSR 3 dB
=15
Magnitude
Frequency (kHz)
FFT
FOS
9.2 9.4 9.6 9.8 10 10.2 10.4 10.6 10.8 11 11.20
50
100
150
window size=10 ms JSR 1 dB
=10 JSR 2 dB
=0 JSR 3 dB
=15
Magnitude
Frequency (kHz)
FFT
FOS
84
3.5 Conclusion
This chapter introduced a new jamming detection method based on high resolution
spectral estimation for GPS signals using FOS. The proposed method featured several
spectral estimation capabilities including high frequency resolution, elimination of
spectral leakage and rejection of background noise. The aforementioned abilities resulted
in a significant enhancement to the GPS jamming detection process.
The performance of the proposed method was examined using GPS data sets generated
by the Spirent GSS 6700 and acquired by the Novatel FireHose frontend GPS receiver.
The proposed method demonstrated more consistent and accurate jamming detection than
conventional FFT based methods in the presence of a single jammer with less sensitivity
to window size. FFT could only match FOS-based detection in specific cases. It provided
resolutions similar to that of the proposed method at a window size 10 times larger than
that used by FOS. Moreover, our method accurately and efficiently estimated the number
and frequencies of multiple in-band CW jammers whose frequency separation is lower
than that of FFT as demonstrated in scenarios 2 and 3. Nevertheless, increasing the
number of jammers degrades the detection accuracy at lower JSR. In General, the
proposed FOS-based method can expand the capabilities of present GPS receivers to
provide robust anti-jamming in challenging environments.
85
Chapter 4
DoA Estimation of GPS Narrowband Interference
4.1 Introduction
GPS interference detection can be achieved in the time, space, or frequency
domain as previously demonstrated in Chapter 3 or in joint variables domain such as
time-frequency or space–time. Generally, spatial signal processing using antenna arrays
is considered one of the most effective techniques for narrowband interference detection
and especially suppression. In Chapter 2, it was stated that antenna arrays enable
interference mitigation through spatial nulling and beamsteering using adaptive
beamforming and high-resolution DoA estimation methods [84] .
DoA estimation of signals from multiple sources plays a key role in the
interference suppression process. Adaptive antenna arrays use the DoA information to
produce nulls in DoA of interference signals and steer the main beam towards the desired
GPS signal. Various DoA estimation algorithms have been developed for array signal
processing applications. The selection of the DoA estimation algorithm is a crucial
element of adaptive antenna array design. Selecting the proper DoA estimation algorithm
enables proper signal DoA estimation, which directly affects null steering and
beamforming processes. Moreover, one of the main problems of null steering and
beamforming occurs when the jammer is in the same direction or very close to the GPS
signal’s DoA. This implies that jamming detection that is based on high resolution DoA
estimation can significantly enhance the anti-jamming process. The DoA estimation
86
techniques that are predominantly used are those that are based on Classical DoA, Capon
and MUSIC algorithms [84]. It has been reported that, Eigen-decomposition based
methods such as MUSIC have high resolution DoA estimation performance [85]. These
methods are based on exploiting the Eigen-structure of the input covariance matrix.
Applying MUSIC to GPS anti-jamming has shown significant enhancement to overall
system performance [86]. However, the performance of MUSIC is limited by coherence
of the jamming sources [87].
This thesis proposes a new high-resolution DoA estimation method to enhance the
GPS jamming detection capabilities using antenna arrays. The proposed method is based
on the orthogonal search technique in order to create bases for the steering function. The
coefficients of the steering function are determined via a unique orthogonal search
method that provides high accuracy DoA estimation. The main advantage of this method
is based on its super high resolution and robust DoA estimation. This method can
enhance anti-jamming techniques based on adaptive array systems, specifically in
situations where the DoA of jamming signals is very close to the DoA of the GPS signal.
Hence, it can enhance the null-steering process and avoid reduction in GPS signal levels
while rejecting jamming signals. Furthermore, the high-resolution capabilities of the
proposed method can resolve jamming signals arriving at very close DoAs and hence
improves the overall performance of the anti-jamming method.
The proposed method is compared to Classical DoA estimation and MUSIC using GPS
L1 signals obtained from a Spirent 6700 simulator; the arrays utilized were linear and
circular arrays. The jamming signals were simulated as CW sources originating from
different directions. In addition, this thesis studies the effect of array geometry on DoA
87
estimation efficiency and proposes an optimum array configuration for enhancing GPS
jamming detection and mitigation. The recommended configuration is examined using
GPS data obtained from the Spirent 6700. A comparison between the performance of the
proposed method and the aforementioned predominant DoA estimation methods is
conducted using the recommended array configuration. For each array configuration, the
performance of the investigated methods is evaluated in case of single and multiple
interference signals. Our results showed that the proposed method could introduce
significant enhancement to GPS jamming detection and hence provide significant
improvement for overall system performance.
4.2 DoA Estimation Using Nonlinear Signal Modeling
This research studies the application of FOS in DoA estimation of jamming
signals. FOS is a highly efficient general purpose modeling method that can be used for
time series analysis and frequency analysis [76]. FOS is known to have features that can
significantly enhance the DoA estimation of CW jamming signals [76,88,82]. They
include a high frequency resolution spectrum and an ability of rejecting white and
colored noise. FOS has been used in numerous applications, including identification of
generalized nonlinear difference equation models, spectrum estimation and DoA
estimation for uniform linear arrays [88,82]. The original orthogonal search (OS)
algorithm is mainly based on suboptimal minimization of the mean squared error (MSE)
between observed and modeled data. The algorithm constructs a functional expansion of
an input using an arbitrary set of non-orthogonal candidate functions. The functional
expansion of the input in terms of the arbitrary candidate functions is given by:
88
𝒚(𝑛) = ∑ 𝑎𝑚 𝑝𝑚(𝑛) + 𝑒(𝑛)
𝑀
𝑚=0
Eq. 4-1
Where the weights 𝑎𝑚 , (where 𝑚 = 0, … , 𝑀) are the weights of the functional
expansion, and 𝑒(𝑛) is the modelling error. Generally, the signal arriving at the antenna
array is a superposition of complex sinusoidal signals and noise. The total signal received
at the 𝑛𝑡ℎ sensor at time 𝑡𝑖 from 𝑆 sources is given by [88]:
𝑦𝑚(𝑡𝑖) = ∑ 𝑃𝑘(𝑡𝑖) exp (𝑗2𝜋𝑓𝑘(𝑡𝑖 − 𝜏𝑚(𝜃𝑘))) +
𝑆−1
𝑘=0
𝑒𝑚(𝑡𝑖) Eq. 4-2
where 𝑖 = 0, … , 𝑇 − 1 is the signal time index
𝑚 = 0, … , 𝑀 − 1 is the sensor index number
𝑘 = 0, ⋯ , 𝑆 − 1 is the signal index number
𝜃𝑘 Is the DoA of the signal from 𝑘𝑡ℎ source
𝑃𝑘 is the complex amplitude of the 𝑘𝑡ℎ source
𝑃𝑘(𝑡𝑖) = |𝑃𝑘(𝑡𝑖)| exp(𝑗𝜑(𝑡𝑖)) and
𝜑(𝑡𝑖) is the phase of the 𝑘𝑡ℎ signal at its source
For 𝐿 real signals 𝑆 = 2𝐿 and the complex amplitudes are conjugate pairs (i.e. 𝑃𝑘+1 =
𝑃𝑘∗ and 𝑓𝑘+1 = −𝑓𝑘 ). The reference is taken at the first element=0, thus 𝜏0(𝜃𝑟𝑘) = 0.
The following model may describe the received signal at the antenna array:
𝑦(𝑛) = ∑ 𝑎𝑘(𝜃𝑘)𝑠𝑘(𝑛)
2𝐿−1
𝑘=0
+ 𝑒(𝑛) Eq. 4-3
where 𝑎𝑘(𝜃𝑟) is 𝑀 × 1 antenna array steering vector for kth
source
𝑠𝑘 is the kth
source signal received by the antenna array
89
When dealing with real signals, each parameter 𝜃𝑘 has two linearly independent steering
vectors: either complex-conjugate exponentials with complex coefficients incorporating
signal phase information, or sine and cosine components with real coefficients may be
used. In this research real (sine and cosine) steering vectors were used. Nevertheless, the
complex exponential notation is retained for consistency with the literature.
OS uses Gram-Schmidt procedure to create an orthogonal basis for the steering vectors,
and a model equivalent to Eq. 4-3 is formed using a set of orthogonal basis vectors{𝑤𝑘}:
𝑦(𝑛) = ∑ 𝑤𝑘(𝜃0, 𝜃1, ⋯ , 𝜃𝑟)𝑔𝑘(𝑛)
2𝐿−1
𝑘=0
+ 𝑒(𝑛) Eq. 4-4
where 𝑟 = 𝑘/2 for 𝑘 even and 𝑟 = (𝑘 − 1)/2 for 𝑘 odd
𝑤𝑗∗ = 𝑤𝑘 for 𝑖 ≠ 𝑘 and * denotes complex-conjugate transpose.
{𝑔𝑘(𝑛)} are the coeffecients of orthogonal basis vectors. The orthogonal basis vectors
{𝑤𝑘} are derived from the candidate functions {𝑎𝑘(𝜃𝑟)} using the following formula:
𝑤𝑘 = 𝑎0 − ∑ α𝑘𝑖𝑤𝑖
𝑘−1
𝑖=0
Eq. 4-5
where 𝛼𝑘𝑖 = 𝑤𝑘∗𝑎𝑘
𝑤𝑘∗𝑤𝑘
and 𝑤0 = 𝑎0
The estimated coefficient 𝑔𝑘(𝑛) that should minimize the mean square error (MSE) are
given by:
𝑔𝑘(𝑛) =𝑤𝑘
∗ 𝑦(𝑛)
𝑤𝑘∗𝑤𝑘
Eq. 4-6
The total MSE is:
90
𝜀2̅̅ ̅ =1
𝑇∑
1
𝑀
𝑇−1
𝑛=0
‖𝒚(𝑛) − ∑ 𝑔𝑘(𝑛)𝑤𝑘
𝑆−1
𝑘=0
‖
2
=1
𝑀𝑇∑ 𝑦∗(𝑛)𝑦(𝑛)
𝑇−1
𝑛=0
− ∑ 𝑄𝑘
𝑆−1
𝑘=0
Eq. 4-7
where
𝑄𝑘 =1
𝑀𝑇∑
[ 𝑤𝑘∗𝒚(𝑛)]2
𝑤𝑘∗𝑤𝑘
𝑇−1
𝑛=0
=𝑤𝑘
𝐻𝑤𝑘
𝑀𝑇∑ 𝑔𝑘(𝑛)
𝑇−1
𝑛=0
Eq. 4-8
𝑄𝑘 is the amount of reduction in MSE gained by adding the candidate basis vector 𝑤𝑘 to
the model. The calculation of 𝑄𝑘 involves the correlations of 𝑤𝑘 with themselves, the
steering vectors 𝑎𝑘 and the data 𝒚(𝑛) are. Therefore, the orthogonal search can discard
the need to explicitly create the orthogonal basis vectors 𝑤𝑘.
FOS works by exploiting these observations and uses Cholesky factorization to calculate
the aforementioned correlations. It starts by creating a variable 𝐷𝑘𝑚, which represents the
correlation between the candidate and the steering vector. This is given by:
𝐷𝑘𝑚 = 𝑤𝑚𝑎𝑘 = 𝑎𝑘∗ 𝑎𝑘 − ∑ 𝛼𝑚𝑗
∗
𝑚−1
𝑗=0
𝐷𝑘𝑗 Eq. 4-9
where 𝑚 = 1 … 𝑘 and 𝛼𝑘𝑗 =𝐷𝑘𝑗
𝐷𝑗𝑗
Another variable 𝐶𝑛𝑘 is used to denote the correlation between the candidate and the
observations vector and defined by:
91
𝐶𝑛𝑘 = 𝑤𝑘∗𝒚(𝑛) = 𝑎𝑘
∗𝒚(𝑛) − ∑ 𝛼𝑘𝑗∗
𝑘−1
𝑗=0
𝐶𝑛𝑗 Eq. 4-10
Accordingly the coefficients of orthogonal basis vectors 𝑔𝑘(𝑛) and the amount of
reduction in MSE 𝑄𝑘 are given by:
𝑔𝑘 (𝑛) =𝐶𝑛𝑘
𝐷𝑘𝑘 Eq. 4-11
𝑄𝑘 = ∑𝐶𝑛𝑘
2
𝑀𝑇𝐷𝑘𝑘
𝑇−1
𝑛=0
Eq. 4-12
The fact that 𝑤𝑘∗𝑤𝑘 = 𝑤𝑘
∗𝑎𝑘 , was utilized in formulating Eq. 4-11 and Eq. 4-12. Given
that 𝑤0 = 𝑎0 the initial values for 𝐷𝑘𝑚 and 𝐶𝑛𝑘 are given by:
𝐷𝑚0 = 𝑎0∗𝑎𝑚 and 𝐶𝑛0 = 𝑎0
∗𝑦(𝑛) Eq. 4-13
The model in Eq. 4-4 is created by adding the candidate functions corresponding to one
direction of arrival at a time. To choose the following estimate θ𝑟, the set of parameters
are accessed to provide the value that yields maximum MSE reduction 𝑄𝑘 Since a real
signal with sine and cosine components is used, the MSE reduction contributed by a
single candidate parameter is in fact the sum of the MSE reductions induced by the two
corresponding steering vectors of that candidate (i.e. 𝑄𝑘 and𝑄𝑘+1). Hence the succeeding
estimate θ𝑟 is given by:
θ𝑟 = 𝑎𝑟𝑔[max (𝑄𝑘 + 𝑄𝑘+1)] Eq. 4-14
Once θ𝑟 is chosen, a set of two terms 𝑔𝑘(𝑛)𝑤𝑘 and 𝑔𝑘+1(𝑛)𝑤𝑘+1 is added to the model
terms. This procedure is repeated L times, once the L parameter estimates are chosen; the
92
values of the signal coefficients 𝑠𝑘(𝑛) can be found recursively using the following
formula:
�̂�𝑘(𝑛) = ∑ 𝑔𝑚(𝑛)𝑣𝑚
𝑆−1
𝑚=𝑘
Eq. 4-15
where 𝑣𝑘 = 1
𝑣𝑚 = − ∑ 𝛼𝑚𝑗
𝑚−1
𝑗=𝑘𝑣𝑗 𝑚 = 𝑘 + 1, k + 2, … , 𝑆 − 1
4.2.1 Stopping Conditions for FOS Algorithm
FOS algorithm can be stopped using one of the following criteria:
a) Reaching a predefined maximum number of terms to be fitted.
This requires the knowledge of the number of narrowband interference signals
impinging upon the antenna array. However, the orthogonal search
approaches may be combined with any of a number of statistical criteria to
determine when to stop adding terms to the model, thus providing an estimate
of the number of signals [88].
b) Reaching a pre-defined threshold for MSE Reduction.
This criterion is achieved when the ratio of MSE to the mean squared value of
the input signal is below a pre-defined threshold. Accordingly, the limitation
of knowing the number of expected interference signals is waved. However, it
may lead to an increase in the processing time. Thus this ratio should be
carefully selected to avoid excessive processing time.
93
c) When adding another term to the model, the MSE reduction is less than the
MSE reduction that would be gained if white Gaussian noise was added.
4.2.2 FOS complexity:
FOS necessitates floating point operations of the order of 𝐶𝑀𝑇 + 𝐶𝐿2 where 𝐶 is the
number of candidate steering vectors searched [82]. Moreover, if the elements of the
array and therefore the data samples are not equally spaced, FOS will require a higher
order which becomes 𝐶𝑀𝑇 + 𝐶𝐿2 + 𝐶𝑀𝐿 [88].
4.2.3 Candidate Function Selection for FOS Algorithm
The process of candidate function selection plays a key role in the FOS algorithm.
In this research, the selected candidate functions are pairs of sine and cosine functions
corresponding to the DoAs of the search domain. Accordingly, the selected candidate
functions for a ULA using the model shown in Eq. 4-3 are given by [82]:
𝑎𝑘(𝜃𝑟) = {𝑎𝑘(𝑚, 𝜃𝑟)} Eq. 4-16
where 𝜃𝑟 is the DoA of 𝑟𝑡ℎ signal .
When 𝜃𝑟 is measured from the line of the antenna array, Eq. 4-17 is used.
𝑎𝑘(𝑚, 𝜃𝑟) = cos (2𝑚𝜋(𝑑
𝜆)cos (𝜃𝑟))
𝑎𝑘+1(𝑚, 𝜃𝑟) = sin (2𝑚𝜋 (𝑑
𝜆) cos(𝜃𝑟)) Eq. 4-17
On the other hand, when 𝜃𝑟 is measured from the line perpendicular to the antenna array
the utilized equation is Eq. 4-18, which is shown below:
𝑎𝑘(𝑚, 𝜃𝑟) = cos (2𝑚𝜋(𝑑
𝜆)𝑠𝑖𝑛(𝜃𝑟))
94
𝑎𝑘+1(𝑚, 𝜃𝑟) = sin (2𝑚𝜋(𝑑
𝜆)sin (𝜃𝑟)) Eq. 4-18
where 𝑚 = 0,1, … , 𝑀 − 1 , 𝑖 = 0,1 … , 𝑇 − 1, 𝑑 is the spacing between elements, and 𝜆
is the wavelength of the received signal.
The coefficients 𝑠𝑘(𝑛) are given by [82]:
𝑠𝑘 = √2𝑃𝑟 cos (𝜋𝑡𝑖 + 𝜑𝑟(𝑡𝑖))
𝑠𝑘+1 = √2𝑃𝑟 sin (𝜋𝑡𝑖 + 𝜑𝑟(𝑡𝑖)) Eq. 4-19
where 𝑃𝑟 and 𝛹𝑟(𝑡𝑖) are the power and phase of 𝑟𝑡ℎsource respectively.
4.3 Antenna Array Design
In the following sections, array design for several array geometries will be studied
from a different perspective that will ultimately serve the goal of DoA estimation.
Therefore, in order to adhere to the common literature notations, the total number of
elements in an array will be referred to as 𝑁 elements while 𝑛 will represent the index of
an element in the array and 𝑛 = 0, 1, … , 𝑁 − 1.
4.3.1 Uniform Linear Array
The ULA is capable of filtering the electromagnetic environment in which it
operates, based on the location of the signals in its vicinity. Array geometry has a crucial
effect on the DoA estimation abilities of antenna arrays. Recently, array geometry
optimization has been extensively studied using various methods. In general, the impact
95
of array arrangement on performance was investigated by considering standard arrays
such as the uniform linear array shown below in Figure 4-1 [89].
Figure 4-1:Uniform linear array
For the general ULA shown in Figure 4-1 with inter-element spacing 𝑑𝑛 and M identical
elements, the signal received by the antenna array is given by:
𝑦(𝑡) = 𝑠(𝑡) ( ∑ 𝑒−𝑗𝑘.𝑑𝑛
𝑀−1
𝑛=0
) Eq. 4-20
where 𝑠(𝑡) is the transmitted signal.
The quantity in parenthesis is known as the array factor (AF). This factoring is regularly
termed pattern multiplication. It can be utilized when the identical array elements are
oriented in the same direction. Moreover the radiation pattern of the array is the result of
the multiplication of element radiation pattern and array factor.
When the elements of the ULA are all located on the z-axis, 𝑑𝑛 = (0,0, 𝑧𝑛). The AF then
becomes:
M M-1
dn dn dn dn dn dn
96
𝐴𝐹 = ( ∑ 𝑒−𝑗2𝜋𝜆
𝑧𝑛𝑐𝑜𝑠𝜃
𝑀−1
𝑛=0
) Eq. 4-21
where 𝑛 = 0, … , 𝑀 − 1
𝜃 is the direction of arrival of received signal.
For an M-element ULA let 𝑑𝑛 = (0,0,𝑛𝜆
2) , hence the AF becomes
𝐴𝐹 = ∑ 𝑒−𝑗𝑛𝜋𝑐𝑜𝑠𝜃
𝑀−1
𝑛=0
Eq. 4-22
The array factor formula can be simplified using the following identity [89]
∑ 𝑐𝑛 =1 − 𝑐𝑀
1 − 𝑐
𝑀−1
𝑛=0
Eq. 4-23
Therefore the array factor for ULA is given by:
𝐴𝐹 =1 − 𝑒−𝑗𝑀𝜋𝑐𝑜𝑠𝜃
1 − 𝑒−𝑗𝜋𝑐𝑜𝑠𝜃 Eq. 4-24
After factoring, Eq. 4-24 reduces to:
𝐴𝐹 = (𝑒−𝑗
𝑀𝜋2
𝑐𝑜𝑠𝜃
𝑒−𝑗𝜋2
𝑐𝑜𝑠𝜃)
𝑠𝑖𝑛 (𝑀𝜋
2 𝑐𝑜𝑠𝜃)
𝑠𝑖𝑛 (𝜋2 𝑐𝑜𝑠𝜃)
Eq. 4-25
The AF magnitude pattern for ULA is shown in Figure 4-2. The magnitude of the array
factor is plotted for an array with 𝑀 = 9 elements. Based on the magnitude of the array
factor, maximum energy is received or transmitted by the array when𝜃 = 90° or when 𝜃
= 270°. Introducing weights and varying the orientation of the ULA can manipulate this
array factor. Since the array factor is a linear function of the weights, therefore, choosing
weights is a minor procedure [90,91,92]. On the other hand, the array factor is a nonlinear
97
exponential function of the element positions; consequently, array geometry optimization
is more challenging [89].
Figure 4-2 Array factor for ULA
Other important parameters of array factors include beamwidth and sidelobe level. The
beamwidth is usually defined as null-to-null or half-power beamwidth. The null-to-null
beamwidth is the angular distance between the first nulls around the main beam [89]. The
half-power beamwidth is the angular distance between the half-power points (i.e. 3 dB
points on the array factor) around the main beam. The sidelobe level is usually defined as
the maximum value of the array factor that is found outside the main beam.
4.3.2 Two-Dimensional Arrays
The one-dimensional array factor is either a function of the DoA elevation or the
DoA azimuth depending on its orientation. Hence, the array can filter the received signals
based on single parameter which is the elevation angle 𝜃 or azimuth angle 𝜙 but cannot
2
4
6
8
10
30
210
60
240
90
270
120
300
150
330
180 0
98
accommodate both parameters in filtering received signals. That is why two-dimensional
arrays are used in order to distinguish between received signals based on azimuth and
elevation [93]. The most common configurations for two-dimensional arrays are
rectangular and circular arrays [84].
4.3.2.1 Rectangular Arrays
For a two-dimensional rectangular array with elements on the x-y plane such as the one
shown in Figure 4-3, the array factor becomes [89]:
𝐴𝐹 = ( ∑ 𝑒−𝑗2𝜋𝜆
(𝑥𝑛𝑠𝑖𝑛𝜃𝑐𝑜𝑠𝜙+𝑦𝑛𝑠𝑖𝑛𝜃𝑠𝑖𝑛𝜙)
𝑀−1
𝑛=0
) Eq. 4-26
Figure 4-3 Rectangular array geometry
a
𝑟
z
y
x
𝜙
b
M
99
The array factor is a function of both spherical angles and can therefore filter the
received signals based on their azimuth and elevation angles. The effect of the array on
the received signal as a function of the angle of arrival can be demonstrated by examining
the array factor. The array factor for rectangular array (𝑚 × 𝑚) with uniform weights
and 𝑑 = (𝑎𝜆
2,
𝑏𝜆
2, 0) where 𝑎, 𝑏 = 0,1,2, … . . 𝑚 − 1 is given by:
𝐴𝐹 = ( ∑ ∑ 𝑒−𝑗𝜋 𝑠𝑖𝑛𝜃(𝑎𝑐𝑜𝑠𝜙+𝑏𝑠𝑖𝑛𝜙)
𝑚−1
𝑎=0
𝑚−1
𝑏=0
) Eq. 4-27
by applying the identity formula
𝐴𝐹 = (1 − 𝑒−𝑗𝑚𝜋𝑠𝑖𝑛𝜃𝑐𝑜𝑠𝜙
1 − 𝑒−𝑗𝜋𝑠𝑖𝑛𝜃𝑐𝑜𝑠𝜙) (
1 − 𝑒−𝑗𝑚𝜋𝑠𝑖𝑛𝜃𝑠𝑖𝑛𝜙
1 − 𝑒−𝑗𝜋𝑠𝑖𝑛𝜃𝑠𝑖𝑛𝜙)
= (𝑒−𝑗𝑚𝜋𝑠𝑖𝑛𝜃𝑐𝑜𝑠𝜙 2⁄
𝑒−𝑗𝜋𝑠𝑖𝑛𝜃𝑐𝑜𝑠𝜙 2⁄) (
𝑒−𝑗𝑚𝜋𝑠𝑖𝑛𝜃𝑠𝑖𝑛𝜙 2⁄
𝑒−𝑗𝜋𝑠𝑖𝑛𝜃𝑠𝑖𝑛𝜙 2⁄)
∗ (𝑠𝑖𝑛(𝑚𝜋𝑠𝑖𝑛𝜃𝑐𝑜𝑠𝜙 2⁄ )
𝑠𝑖𝑛(𝜋𝑠𝑖𝑛𝜃𝑐𝑜𝑠𝜙 2⁄ )) (
𝑠𝑖𝑛(𝑚𝜋𝑠𝑖𝑛𝜃𝑠𝑖𝑛𝜙 2⁄ )
𝑠𝑖𝑛(𝜋𝑠𝑖𝑛𝜃𝑠𝑖𝑛𝜙 2⁄ ))
Eq. 4-28
To simplify the process of plotting, the following variables are presented:
𝑢 =𝑘𝑥𝜆
2𝜋= 𝑠𝑖𝑛𝜃𝑐𝑜𝑠𝜙 and 𝑣 =
𝑘𝑦𝜆
2𝜋= 𝑠𝑖𝑛𝜃𝑠𝑖𝑛𝜙
hence,
𝐴𝐹 = (𝑒−𝑗
𝑚𝜋𝑢2
𝑒−𝑗𝜋𝑢2
) (𝑒−𝑗
𝑚𝜋𝑣2
𝑒−𝑗𝜋𝑣2
) (sin (𝑚𝜋𝑢 2)⁄
sin (𝜋𝑢 2)⁄) (
sin (𝑚𝜋𝑣 2)⁄
sin (𝜋𝑣 2)⁄)
Eq. 4-29
The array factor formula for two-dimensional arrays and the beamwidth are more
difficult to study in 3D. Usually, two dimensional arrays are studied in certain planes (i.e.
elevation and azimuthal planes) and given in half-power or null-to-null form, similarly to
100
the one-dimensional case. The sidelobe level remains, similar to the previous section, the
peak value of the array factor outside of the main beam.
An illustration of the AF magnitude pattern is shown in Figure 4-4, Figure 4-5 and
Figure 4-6. The shown pattern corresponds to a square array of 9 elements uniformly
arranged in the x,y plane with equal space of 𝜆/2. Rectangular arrays have the
disadvantage of the edge effect, where the outer elements in the finite array have
degraded performance when compared with the central array elements. It can also be
depicted from the AF magnitude shown in Figure 4-4 that the rectangular array response
varies along the elevation direction which limits the capabilities of rectangular antenna
arrays in beam steering.
Figure 4-4 Array factor magnitude for rectangular array
101
Figure 4-5 Array factor pattern for rectangular array
Figure 4-6 XY-Plane view of array factor pattern for rectangular array
102
4.3.2.2 Circular Array
The planar array geometries can be divided into three other sub-categories;
circular, rectangular, and square. The circular arrays stand out of the three as they do not
have edge elements. In the absence of edge constraints, the beam pattern of a circular
array can be electronically rotated. Moreover, the circular arrays also have the capability
to compensate the effect of mutual coupling by breaking down the array excitation into a
series of symmetrical spatial components [94,95].
Figure 4-7 Circular array geometry
The array factor of a circular antenna array which is shown in Figure 4-7 is given by [94]:
𝐴𝐹(𝜃, 𝜙) = ∑ 𝑒𝑗(𝑘𝑏𝑠𝑖𝑛𝜃 cos(𝜙−𝜙𝑛))
𝑀
𝑛=1
Eq. 4-30
where 𝜙𝑛 is the angular position of element 𝑛
b
n n +1
z
y
x
𝜙𝑛
M
𝜙
103
𝑀 is the number of antenna elements
𝑏 is the radius of the circle
and 𝑘 is the wave number , 𝑘 =2𝜋
𝜆.
The array factor for circular antenna array is a function of both spherical angles and can
therefore filter the received signals based on their azimuth and elevation angles. The
effect of the array on the received signal as a function of the angle of arrival can be
demonstrated by examining the array factor. The magnitude pattern of the circular array
factor is shown in Figure 4-8, Figure 4-9 and Figure 4-10. This pattern is drawn for a
circular array of nine elements that are equally spaced on a circular circumference with
radius of 𝜆
2. Circular arrays configuration does not have the edge effect. It is clear from
the AF magnitude shown in Figure 4-8 that the circular array response is symmetrical
around the elevation direction, which enhances the capabilities of circular antenna arrays
in beam steering.
104
Figure 4-8 Array factor magnitude for circular array
Figure 4-9 Array factor pattern for circular array
105
Figure 4-10 XY-plane view of array factor pattern for circular array
A major factor that is considered in array design is aliasing. Aliasing occurs when signals
propagating in separate directions yield similar steering vectors. In that case, the array’s
response towards the two directions will be identical, so that the array cannot differentiate
between the two directions. Equivalently, in spectral analysis, when the sampling rate is
too small in time, frequency distinction cannot be achieved.
For a ULA, plane waves from distinct directions with duplicate steering vectors will exist
if the spacing between elements is larger than λ / 2. Similarly, for uniformly spaced
rectangular arrays with elements arranged on the x-y plane, there distinct directions with
duplicate steering vectors will exist if the element spacing in either the x or y directions is
larger than λ / 2. Upon the occurrence of aliasing, the main beam may be replicated
somewhere else in the pattern. These replicated beams are denoted as grating lobes.
106
For arrays not having a non-uniform structure, the distance between elements can be
much larger than λ / 2 without introducing aliasing. In this case, two distinct DoAs will
not produce duplicate steering vectors. Nevertheless, although aliasing technically does
not occur, there may be steering vectors that are very similar, hence, grating lobes could
exist. Determining whether or not this takes place for an arbitrary array is very
challenging. In general, if a non-uniform array is utilized, the array factor can be tested to
guarantee that grating lobes are not created [89]. Mathematical studies on the uniqueness
of steering vectors can be found in [96].
4.4 Array Geometry Optimization
An antenna array is commonly used for communication systems operating in
environments with a large quantity of co-channel interference. In this case, an antenna
array is necessary for blocking interference spatially separated from the desired signal
direction. Given that interference signals’ DoAs are unknown, it would not be practical to
optimize the array geometry for specific interference conditions.
Statistical characterization of expected interference DoAs and relative power may
enhance interference rejection. However, these characteristics cannot be ensured in
highly dynamic environments. GPS receivers can reject some interference arriving at
relatively low elevations [28]. Consequently, the DoA of interference signals can be
limited to a defined range of elevations. By optimizing array geometry with respect to
interference characteristics, it is possible to reduce the expected interference power that is
not mitigated by the array.
107
Optimal array geometry was determined in [89] by solving an optimization problem. The
solution was based on the autocorrelation matrix of received signal, given by:
𝑅𝑋𝑋 = 𝐸[𝑋(𝑡)𝑋𝐻(𝑡)] Eq. 4-31
In this equation the expectation is over time. The autocorrelation matrix 𝑅𝑋𝑋 varies for
each interference condition. Hence the expected autocorrelation matrix 𝑆𝑋 provides a
general representation of interference conditions, 𝑆𝑋 is defined as
𝑆𝑋 = 𝐸𝐼 [𝑅𝑋𝑋 ] Eq. 4-32
where the expectation operator is now over the interference situations. If all the antenna
elements have similar physical orientation, 𝑆𝑋 can then be otherwise calculated by
considering the antenna array elements as isotropic antenna elements and adjusting the
power levels in the interference environment.
In GPS systems, the majority of jamming signals have a larger power than the
signal of interest. Otherwise, they would be rejected to an extent at the stage of signal
despreading. Therefore, the array’s main goal is to minimize the output power while
restricting one of the weights in the array to be unity. This is analogous to sidelobe
cancellation [86] and was previously utilized in a 7-element adaptive array developed by ,
American weapons contractor, Raytheon for battling interference in GPS systems [97].
The power minimization technique significantly enhances the antenna array
capabilities to reduce the amount of received interference power. This is considered a
suboptimal technique with respect to MSE since the power minimization does not attempt
to place the maximum of the array factor towards the signal of interest. However, when
the interference power is much stronger than the power of the desired signal, this
108
technique can be employed to produce weights that are close to those produced by the
minimum mean square error (MMSE) techniques [89].
The output power from the array at any time is
𝑦∗𝑦 = 𝑤𝐻𝑋𝑋𝐻𝑤 Eq. 4-33
In fixed interference cases, the average output power �̅� becomes
�̅� = 𝑤𝐻𝑅𝑋𝑋𝑤 Eq. 4-34
𝑃 is therefore a measure of the average output power for a given interference
environment
𝑃 = 𝑤𝐻𝑆𝑋𝑤 Eq. 4-35
One of the weights is restricted to be unity so that the power minimization algorithm does
not nullify all the weights. Furthermore, in order to minimizing the effect of mutual
coupling, the separation between elements is required to be at least 𝜆
4. Let 𝑑𝑖𝑗 be the
separation between elements i and j. The problem of finding an optimal array for
interference suppression can be written as an optimization problem. The minimization
variables are the complex weights and the values of 𝑑𝑖𝑗 . It is desired to obtain
min(𝑤𝐻𝑆𝑋𝑤) where the minimization is subject to 𝑤𝐻𝑒1 = 1, where 𝑒1𝑇 = [1 0 0 ⋯ 0],
such that Eq. 4-36 is achieved.
𝑑𝑖𝑗 ≥𝜆
4 , 𝑓𝑜𝑟 𝑖 ≠ 𝑗 Eq. 4-36
Assuming the locations of the antenna elements are known (or held fixed), the optimal
weight vector for this problem can be found by using Lagrange multipliers. The
minimum value of the objective function for a fixed geometry becomes
109
min(𝑤𝐻𝑆𝑋𝑤) =1
𝑒1𝑇𝑆𝑋
−1𝑒1
Eq. 4-37
The optimization procedure minimizes the objective function over all array geometries
that meet the optimization problem in Eq. 4-36.
The results in [97,89] suggested that the interference suppression capabilities are greatest
when array elements are spaced as closely as possible. Therefore, a trade-off is to be had
between interference suppression and largely spaced arrays used for either diversity or
mutual coupling minimization. Based on array configuration analysis, the selected
optimum array configuration in this research is circular array of 7-elements and central
element. The selected array configuration is shown in Figure 4-11.
Figure 4-11 Circular array with central element
b
n n +1
z
y
x
𝜙𝑛
M
𝜙
110
The radius of the circle is equal to 𝜆
2 and the 7 antenna elements are equally spaced on a
circular circumference with angular spacing of 2𝜋
7. The array factor of the selected
configuration is given by [94]:
𝐴𝐹(𝜃, 𝜙) = 1 + ∑ 𝑒𝑗(𝑘𝑏𝑠𝑖𝑛𝜃 cos(𝜙−𝜙𝑛))
𝑀
𝑛=1
Eq. 4-38
where 𝜙𝑛 is the angular position of element 𝑛
𝑀 is the number of antenna elements
𝑏 is the radius of the circle
𝑘 is the wave number , 𝑘 =2𝜋
𝜆
The array factor magnitude pattern for the selected optimum configuration is shown in
Figure 4-12, Figure 4-13 and Figure 4-14.
Figure 4-12 Array factor magnitude for circular array with central element
111
The AF magnitude for circular array with central element is similar to that of the circular
array previously shown in Figure 4-8, except that it is superior in the aspect of lower
side lobes levels. This provides the circular array with a central element a more enhanced
capability of interference rejection.
Figure 4-13: Array factor pattern for circular array with central element
112
Figure 4-14 XY-Plane view of array factor pattern for circular array with central
element
4.5 High resolution FOS-based DoA Estimation Using a 2D Circular Array with
Central Element
The high resolution FOS-based-DoA estimation algorithm was modified for processing
signals received by a circular array with central element. The received signal model
defined by Eq. 4-3 was modified using the array factor of 2D circular array with central
element given in Eq. 4-36. Generally, the direction of arrival estimate for a 2D array is
defined by elevation 𝜃 and azimuth 𝜙. Accordingly, the received signal model in Eq. 4-3
was modified to include the azimuth variable in the steering vector. The modified
received signal model is given by:
𝑦(𝑛) = ∑ 𝑎𝑘(𝜃𝑟𝑘, 𝜙𝑟𝑘)𝑠𝑘(𝑛)
2𝐿−1
𝑘=0
+ 𝑒(𝑛) Eq. 4-39
113
where 𝑎𝑘(𝜃𝑟𝑘, 𝜙𝑟𝑘) is an 𝑀 × 1 antenna array steering vector for the kth
source
𝑠𝑘 is the kth
source signal received by the antenna array.
Similarly, the orthogonal basis model for Eq. 4-39 is modified to include the azimuth
direction. Therefore the orthogonal basis signal model is defined as:
𝑦(𝑛) = ∑ 𝑤𝑘(𝜃0, 𝜃1, ⋯ , 𝜃𝑟 , 𝜙0, 𝜙1, ⋯ , 𝜙𝑟)𝑔𝑘(𝑛)
2𝐿−1
𝑘=0
+ 𝑒(𝑛) Eq. 4-40
where 𝑟 = 𝑘/2 for 𝑘 even and 𝑟 = (𝑘 − 1)/2 for 𝑘 odd
𝑤𝑗∗ = 𝑤𝑘 for 𝑖 ≠ 𝑘 and * denotes complex-conjugate transpose.
The model in Eq. 4-40 is created by adding the candidate functions corresponding to one
direction of arrival at a time. To choose the succeeding estimate (θ𝑟 , ϕ𝑟), the set of
parameter is accessed to provide the value that yields maximum MSE reduction 𝑄𝑘 Since
a real signal with sine and cosine components is used, the MSE reduction contributed by
a single candidate parameter is in fact the sum of the MSE reductions induced by the two
corresponding steering vectors of that candidate (i.e 𝑄𝑘 and 𝑄𝑘+1). Hence the succeeding
estimate (θ𝑟 , ϕ𝑟), is given by:
(θ𝑟 , ϕ𝑟), = 𝑎𝑟𝑔[max (𝑄𝑘 + 𝑄𝑘+1)] Eq. 4-41
Once (θ𝑟 , ϕ𝑟) are selected, a set of two terms 𝑔𝑘(𝑛)𝑤𝑘 and𝑔𝑘+1 (𝑛)𝑤𝑘+1 is added to the
model terms. This procedure is repeated L times, once the L parameter estimates are
chosen; the values of the signal coefficients 𝑠𝑘(𝑛) can be found recursively using the
same formulas shown in Eq. 4-15.
114
Candidate function selection for FOS algorithm
The selected candidate functions for a circular array with central element were
defined using the new steering vector form. The general form of the candidate function is
given by:
𝑎𝑘(𝜃𝑟 , 𝜙𝑟) = {𝑎𝑘(𝑚, 𝜃𝑟 , 𝜙𝑟)} Eq. 4-42
where 𝜃𝑟 𝑎𝑛𝑑 𝜙𝑟 are the elevation and azimuth of the DoA of the 𝑟𝑡ℎ signal
respectively.
The candidate functions for the circular array with central element, 𝜃𝑟 measured
from the line perpendicular to the antenna array are given by:
𝑎𝑘(𝑚, 𝜃𝑟 , 𝜙𝑟) = cos (2𝑘𝑏𝜋𝑠𝑖𝑛𝜃𝑟 cos(𝜙𝑟 − 𝜙𝑚))
𝑎𝑘+1(𝑚, 𝜃𝑟 , 𝜙𝑟) = sin (2𝑘𝑏𝜋𝑠𝑖𝑛𝜃𝑟 cos(𝜙𝑟 − 𝜙𝑚))
Eq. 4-43
where 𝑚 = 1, … , 𝑀 − 1 , 𝑖 = 0,1 … , 𝑇 − 1
𝑎𝑘(0, 𝜃𝑟 , 𝜙𝑟) = 1 𝑎𝑛𝑑 𝑎𝑘+1(0, 𝜃𝑟 , 𝜙𝑟) = 0
𝜙𝑛 is the angular position of element n
𝑀 is the number of circular antenna elements
and 𝑏 is the radius of the circle.
In this research, an initial candidate number is defined for 360 azimuth angles and
90 elevation angles. Hence the number of candidates is 360 X 90.
4.6 Experimental Setup
In order to verify the proposed method, examination in a controlled environment
such as a hardware simulator is necessary. In this research Novatel Digital GNSS
Antenna (DGA) with software receiver and SPIRENT GSS6700 simulator running
115
SimREPLAYPlus software are utilized to verify the proposed method. The simulation of
antenna array requires DGA for each array element which imposes additional complexity
in data processing to synchronize received data among array elements and ensure that all
elements of the array are identical. In this thesis a new method that enables the
simulation of multi-dimensional GPS antenna array systems using limited hardware
resources is introduced.
One RF-frontend and a single RF output SPIRENT GSS6700 simulator running
SimREPLAYPlus software are utilized to verify the proposed method. Initially, the
desired experimental parameters such as number of antenna elements, array geometry,
GPS signal frequency (i.e. L1, L2 and L5), and consequently antenna element spacing,
are defined. Upon designating the zero-phase element’s (reference element) location in
terms of latitude, longitude and height, the remaining elements of the array are mapped
according to the pre-defined experimental configuration. Once the locations of all
antenna elements are set, a simulation scenario is created using the simulator software.
The simulator software enables the reiteration of the simulation scenario without
changing any parameters except for the location of the antenna element in 3D space.
Figure 4-15 illustrates the simulation sequence for a given 𝑀-element array, where each
iteration corresponds to one of the array elements where 𝑛 = 1, … , 𝑀.
Figure 4-15: Simulation sequence of an M-element array of GPS antennas
Simulation #1 Simulation #2 Simulation #n
116
The RF GPS signal output from the hardware simulator is acquired, digitized, down-
converted and stored using DGA. Finally, post-processing software implemented in
Matlab is used for further testing. The developed technique substitutes for the presence of
a physical array and multiple frontends. Hence, he array is assumed to be calibrated and
hence mutual coupling is neglected. As mentioned earlier, single RF output Spirent
GSS6700 GPS simulator running SimREPLAYPlus software was employed. The former
generates RF GPS L1 signals while the latter controls the simulator and provides scenario
editing and replaying capabilities. The simulator is connected to the Novatel FireHose
frontend which provides the raw I and Q baseband samples as shown in Figure 4-16a.
The FireHose settings adopted for the experiment are shown in Table 4-1 below. The
antenna configuration used in the first experiment was a 7 element ULA with constant
element spacing of 9.5 cm.
Table 4-2 shows the locations of the 7 elements, which were used in the simulation runs.
Table 4-1: Settings adopted for data collection using FireHose frontend
Simulation Parameter Parameter Value
Sampling Frequency (MHz) 2.5
Sampling Complex ( I and Q )
Quantization Bits 8
117
Table 4-2: Location of the array elements used in the experiment
Latitude Longitude Altitude (m)
Degrees Minutes Degrees Minutes
38 41.1306350753688 -96 30.2343023307719 100
38 41.1306387187878 -96 30.2342370968847 100
38 41.1306423622068 -96 30.2341718630009 100
38 41.1306460056250 -96 30.2341066291172 100
38 41.1306496490418 -96 30.2340413952265 100
38 41.1306532924587 -96 30.2339761613428 100
38 41.1306569358752 -96 30.2339109274590 100
The raw samples obtained from the frontend are post-processed and tested using software
developed by the NavINST research group. The experiment was conducted on March
18th
, 2014 at 10:00 am. Ten GPS satellites were available at the time of the experiment.
Figure 4-16b shows the skyplot of the available satellites. Only 5 satellites were used for
the DoA estimation methods verification. The 5 satellites’ space vehicle numbers (SVN)
are: 2,4,5,9 and 30. They were chosen specifically due to their positions and elevations in
order to prevent any two satellites from having similar elevations since this is a one
dimensional array and azimuth is not measured. The concentric circles and the diameters
represent the elevation and azimuth angles in degrees respectively.
118
Figure 4-16: Experimental setup and Skyplot of the experiment.
(a) Experimental setup of the Spirent 6700 and the Firehose
(b) Skyplot of available satellites and the ones used, provided by the developed software receiver.
The acquisition results, shown in Figure 4-17, produced by the NavINST software
receiver using the data obtained from the FireHose verified the correctness of the data by
acquiring 5 satellites as planned in the Spirent simulation scenario. Furthermore, DoA
estimates using classical DoA estimation method were computed as shown in Table 4-3.
15
30
45
60
75
30
210
60
240
90270
120
300
150
330
180
0
04
10
1214 05
24
02
09
30
29
GPS (Used)
GPS (Available)
15
30
45
60
75
30
210
60
240
90270
120
300
150
330
180
0
04
10
1214 05
24
02
09
30
29
GPS (Used)
GPS (Available)
(a)
(b)
119
Comparing the estimated DoAs of selected satellite signals with their simulated elevation
and azimuth, four satellites were successfully detected which confirms the accuracy of
the proposed simulation technique.
Figure 4-17: Acquisition results confirming the utilized satellites after the rest of the
available satellites have been switched off using SimReplay Plus
0 5 10 15 20 25 300
0.5
1
1.5
2
2.5
3
3.5
4
4.5
5Acquisition results
PRN number (no bar - SV is not in the acquisition list)
Acq
uisi
tion
Met
ric
Not acquired signals
Acquired signals
120
Table 4-3: Experimental results for DoA estimation of 5 satellites.
PRN 4 9 2 5 30
Elevation type Low Low/
Moderate
Moderate/
High
High Moderate
Elevation angle 24° 153° 59.67° 106° 132.12°
Classical 38.8 155.8 64.9° N/A 125.6
Error 14.8° 2.8° 5.23° N/A 6.52
4.7 Simulation and Results
The method explained in the previous section was used to generate the GPS L1
signals that were used in this chapter; the simulated signals’ synchronization was later
conducted. This ensures perfect time stamp alignment of recorded sections of GPS
signals. Although mutual coupling was not considered, the experimental setup arranged
the array elements as wide as possible to ensure that the effect of mutual coupling is
negligible since the effect was not considered in this research and the simulated data had
to conform to this assumption.
Two antenna array geometries were examined in this research, namely one-
dimensional and two-dimensional arrays. The former involved a ULA configuration
composed of 7 equally spaced elements separated by 𝜆𝐿1
2 . The latter encompassed two
array configurations which are the 7 element UCA and its expansion, the optimum array
geometry, which was constructed using 8 elements. Seven elements were arranged on the
circumference of a circle with radius of 𝜆𝐿1
2 and one element at the center of the circle.
The elements were arranged on the circle’s circumference with an equal angular space of
121
2𝜋
7. The performance of the proposed method is evaluated in terms of the number of
jammers detected and the resolution of the method. Therefore, several scenarios have
been devised; they differ depending on the geometry of the array, the number of jammers,
their power and their spatial locations.
Three sets of scenarios are studied; namely for the three geometries 1) ULA, 2)
UCA and 3) the optimum array (UCA with central element). Scenarios tested the
detection capabilities of the proposed method and compare it to MUSIC and Classical
DoA estimation methods. The scenarios test a single jammer and multiple closely spaced
jammers with equal Jamming to signal ratios. For all array configurations the test
jamming signal levels were 15 and 45 dB. The process of introducing jamming signals is
achieved by adding sinusoidal signals to the output of DGA. The simulated jammer
frequencies ranged between 100 to 400 Hz. This is to ensure that the introduced jamming
signals fall within the bandwidth of the DGA output L1 signals which are baseband
signals with 2 kHz bandwidth.
4.7.1 Results for ULA
The following section demonstrates the results obtained from the different
scenarios that were implemented in order to evaluate the performance of the developed
high resolution DoA estimation method for the ULA geometry. The proposed method
was tested with single and multiple jamming signals arriving at different JSRs.
4.7.1.1 Single Jammer
The results obtained for a single jammer simulated at a frequency of 100 Hz and
arriving at an elevation angle of 50o are shown in Figure 4-18 and Figure 4-19. The
122
figures illustrate the normalized output for Classical DoA, MUSIC and FOS DoA
methods.
Figure 4-18 DoA estimation of one jammer at JSR=15 dB
It can be demonstrated from the figures that Classical DoA, MUSIC and FOS detected
the single jammer with high accuracy at JSRs of 15 and 45 dB. When detection of a
single GPS jamming signal is required, the performance of all three methods is
satisfactory as the jammer of interest is usually arriving at relatively high power.
0 20 40 60 80 100 120 140 160 1800
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
Angle in degrees
Norm
aliz
ed S
patial S
pectr
um
Classical
Proposed Method
MUSIC
123
Figure 4-19: DoA estimation of one jammer at JSR=45 dB
The main contribution of the proposed method when estimating the DoA of a single
jammer is the high accuracy in the detection of jamming signals’ amplitude. This is
mainly related to the nature of operation of FOS on which the method is built. It operates
by constructing a signal model determined by candidate functions corresponding to the
detected signals.
4.7.1.2 Multiple Jamming Signals
The performance analysis of FOS versus MUSIC and classical DoA was
examined in terms of resolution, tolerance to JSR and tolerance to jamming signals
coherence. For this purpose, three Jamming signals were simulated with frequencies of
0 20 40 60 80 100 120 140 160 1800
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
Angle in degrees
Classical
Proposed Method
MUSIC
124
100 Hz, 200 and 300 Hz and arriving at elevations of 500, 600 𝑎𝑛𝑑 700 respectively.
The tolerance to jamming signals coherency was examined by repeating this scenario
with three jamming signals at frequencies of 100,105 and 110 Hz.
The results obtained for 3 jammers arriving at elevations of (500, 600 𝑎𝑛𝑑 700 ) are
shown in Figure 4-20 and Figure 4-21 at different JSR’s. The figures show the
normalized output for Classical DoA estimation, MUSIC and FOS-based DoA estimation
methods. The JSRs of the jamming signals used in Figure 4-20 and Figure 4-21 are 15
and 45 dB respectively.
Figure 4-20 DoA estimation of 3 jammers at JSR=15 dB and frequencies of 100Hz, 200Hz
and 300Hz
0 20 40 60 80 100 120 140 160 1800
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
Angle in degrees
Norm
aliz
ed S
patial S
pectr
um
Classical
Proposed Method
MUSIC
125
Figure 4-21 DoA estimation of 3 jammers arriving at JSR=45 dB and frequencies of 100Hz,
200Hz and 300Hz
The results shown in the above figures demonstrate that the proposed method
FOS-based method outperformed both of MUSIC and Classical DoA due to its high
tolerance to variation of JSR and jamming signals coherency. MUSIC showed good
performance in detecting three jammers at JSR of 45 dB with zero error in estimated
DoAs, but its performance degraded at JSR of 15 dB as it detected two jammers only. On
the other hand FOS had stable performance in terms of the number of jammers detected
at JSR of 15 dB and 45 dB as it detected 3 jammers accurately with zero error in
estimated DoAs. FOS had a slight degradation at JSR of 15 dB where its power allocation
for detected jammer was not equally divided among the three jammers that were initially
simulated with equal power.
0 20 40 60 80 100 120 140 160 1800
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
Angle in degrees
Norm
aliz
ed S
patial S
pectr
um
Classical
Proposed Method
MUSIC
126
4.7.2 Results for UCA
The proposed method (FOS) performance was examined using a UCA
configuration described earlier with 7 elements equally spaced on the circumference of a
circle with a radius of 𝜆𝐿1
2.
4.7.2.1 Single Jammer
A single jammer was simulated as a 100 Hz sinusoidal signal arriving at elevation
of 400 and an azimuth of 400. Figure 4-22, Figure 4-23 and Figure 4-24, demonstrate
that the performance of FOS, MUSIC and Classical DoA in 2D single jammer detection
is almost identical and shows clear detection of the jamming signal with zero error in
both elevation and azimuth. The advantage of FOS and MUSIC is that their spatial
spectrum has much higher resolution compared to that of Classical DoA.
Figure 4-22 DoA estimation of one jammer using Classical DoA at JSR=45 dB
0
20
40
60
80
100
0
100
200
300
400
0
0.2
0.4
0.6
0.8
1
Elevation (Degrees)
X: 40
Y: 40
Z: 1
Classical Number of Jammers=1 JSRdB
= 45
Azimuth (Degrees)
Norm
alized S
patial S
pectr
um
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
127
Figure 4-23: DoA estimation of a single jammer using MUSIC at JSR = 45 dB
Figure 4-24 DoA estimation of one jammer using FOS at JSR = 45 dB
02040
6080100
0100
200300 400
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
Elevation (Degrees)
MUSIC Number of Jammers=1 JSRdB
= 45
Azimuth (Degrees)
X: 40
Y: 40
Z: 1
Norm
aliz
ed S
patial S
pectr
um
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
0100200300400
0 20 40 60 80 100
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
Azimuth (Degrees)
X: 40
Y: 40
Z: 1
FOS Number of Jammers=1 JSRdB
=15
Elevation (Degrees)
Norm
aliz
ed S
patial S
pectr
um
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
128
4.7.2.2 Multiple Jamming Signals
In this scenario, three equal power jammers were simulated at frequencies of 100,
200 and 300 Hz with elevations and azimuths separations of 10°. The DoAs of the three
jammers were simulated at (30°, 30°), (40°, 40°) and (50°, 50°). This experiment is
conducted at JSRs of 15 and 45dB. Results obtained from MUSIC and Classical DoA
estimation methods are shown in Figure 4-25 to Figure 4-28. In this scenario the
performance of FOS is compared to MUSIC only as their performance versus Classical
DoA method was already examined for ULA and showed superior performance in all
cases over Classical DoA.
Figure 4-25: DoA estimation of 3 closely spaced jammers using MUSIC at JSR = 15 dB
0 10 20 30 40 50 60 70 80 900
200
400
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
Azimuth (Degrees)
Elevation (Degrees)
X: 48
Y: 48
Z: 1
MUSIC Number of Jammers=3 JSRdB
= 15
X: 33
Y: 33
Z: 0.4891
Norm
aliz
ed S
patial S
pectr
um
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
129
Figure 4-26: DoA estimation of 3 closely spaced jammers using FOS at JSR = 15 dB
Table 4-4: Simulation results of 3 closely spaced jammers at JSR=15dB
Jammer 1 2 3
Elevation 30° 40° 50°
Azimuth 30° 40° 50°
MUSIC (33°,33°) N/A (48°, 48°)
Error (3°, 3°) N/A (2°, 2°)
FOS (29°,29°) (40°,40°) (50°,50°)
Error (1°, 1°) (0°, 0°) (0°, 0°)
0 10 20 30 40 50 60 70 80 900
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
X: 50
Y: 50
Z: 0.2353
Azimuth (Degrees)
FOS Number of Jammers=3 JSRdB
=15
X: 40
Y: 40
Z: 1
X: 29
Y: 29
Z: 0.2347
Norm
aliz
ed S
patial S
pectr
um
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
130
Figure 4-27: DoA Estimation of 3 closely spaced jammers using MUSIC at JSR = 45 dB
Figure 4-28: DoA estimation of 3 closely spaced jammers using FOS at JSR = 45 dB
0102030405060708090 0200
4000
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
Azimuth (Degrees)
X: 30
Y: 30
Z: 0.2455
Elevation (Degrees)
X: 39
Y: 39
Z: 0.4481
MUSIC Number of Jammers=3 JSRdB
= 45
X: 49
Y: 49
Z: 1
Norm
aliz
ed S
patial S
pectr
um
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
0200
4000 10 20 30 40 50 60 70 80 90
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
Elevation (Degrees)
X: 50
Y: 50
Z: 0.9535
Azimuth (Degrees)
X: 40
Y: 40
Z: 1
FOS Number of Jammers=3 JSRdB
=45
X: 29
Y: 29
Z: 0.9535
Norm
aliz
ed S
patial S
pectr
um
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
131
Table 4-5: Simulation results of 3 closely spaced jammers at JSR=45dB
Jammer 1 2 2
Elevation 30° 40° 50°
Azimuth 30° 40° 50°
MUSIC (30°,30°) (39°, 39°) (49°, 49°)
Error (0°, 0°) (1°, 1°) (1°, 1°)
FOS (29°,29°) (40°,40°) (50°,50°)
Error (1°, 1°) (0°, 0°) (0°, 0°)
Results shown in Figure 4-26 and Figure 4-28 showed steady performance for
FOS in terms of resolution and number of detected jammers for both cases of JSR 15 dB
and 45 dB. The degradation in FOS was observed in its allocation of power to jamming
signals as it was not accurately allocated for jammers arriving at JSR of 15 dB. The
increase in the level of noise affected FOS performance and caused most of the power to
be allocated to the jammer arriving at 40°. MUSIC performed similarly to FOS at JSR of
45 dB as shown in Figure 4-27 with a slight decrease in accuracy as shown in Table 4-5
with maximum error of 1° in both elevation and azimuth of the jamming signals.
Additionally, MUSIC had a much lower accuracy in power allocation for the three
jamming signals. The degradation in MUSIC performance was more evident at lower
JSR where MUSIC was able to detect only two jamming sources as shown in
Figure 4-25. The accuracy of the detected DoA elevation and azimuth of jamming signals
132
at 15 dB JSR are close for both FOS and MUSIC as shown in Table 4-4 with slight
increase of 1° in the error of jamming signals detected by MUSIC.
Examining the performance of FOS versus MUSIC in resolving very closely spaced jammers:
In this scenario, the performance of 2D FOS is compared to that of MUSIC by
introducing 2 close-by signals at 100 Hz and 200 HZ separated by 5° in both azimuth and
elevation, thus further challenging the detection methods. The two jammers under
investigation were simulated with DoA elevation and azimuth of (400, 400)
and (450, 450) respectively. Results are shown in Figure 4-29 to Figure 4-32 and
numerical results are listed in Table 4-6 and Table 4-7.
Figure 4-29: DoA estimation of 2 closely spaced jammers using MUSIC at JSR = 15 dB
0 10 20 30 40 50 60 70 80 900
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
Elevation (Degrees)
MUSIC Number of Jammers=2 JSRdB
= 15
X: 43
Y: 43
Z: 1
Norm
aliz
ed S
patial S
pectr
um
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
133
Figure 4-30: DoA estimation of 2 closely spaced jammers using FOS at JSR = 15 dB
Table 4-6: Simulation results of 2 closely spaced jammers at JSR=15dB
Jammer 1 2
Elevation 40° 45°
Azimuth 40° 45°
MUSIC (43°,43°) N/A
Error (3°, 3°) N/A
FOS (40°,40°) (45°,45°)
Error (0°, 0°) (0°, 0°)
0 10 20 30 40 50 60 70 80 900
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
X: 45
Y: 45
Z: 0.8815
Elevation (Degrees)
FOS Number of Jammers=2 JSRdB
= 15
X: 40
Y: 40
Z: 1
Norm
aliz
ed S
patial S
pectr
um
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
134
Figure 4-31: DoA estimation of 2 closely spaced jammers using MUSIC at JSR = 45 dB
Figure 4-32: DoA estimation of 2 closely spaced jammers using FOS at JSR = 45 dB
01020304050607080900
200400
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
Azimuth (Degrees)
Elevation (Degrees)
X: 40
Y: 40
Z: 0.4602
X: 45
Y: 45
Z: 1
MUSIC Number of Jammers=2 JSRdB
= 45
Nor
mal
ized
Spa
tial S
pect
rum
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
0 10 20 30 40 50 60 70 80 900
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
X: 45
Y: 45
Z: 1
Elevation (Degrees)
FOS Number of Jammers= 2 JSRdB
= 45
X: 40
Y: 40
Z: 0.979
Nor
mal
ized
Spa
tial S
pect
rum
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
135
Table 4-7: Simulation results of 2 closely spaced jammers at JSR=45dB
Jammer 1 2
Elevation 40° 45°
Azimuth 40° 45°
MUSIC (40°,45°) (40°, 45°)
Error (0°, 0°) (0°, 0°)
FOS (40°,40°) (45°,45°)
Error (0°, 0°) (0°, 0°)
The above results of the FOS-based method and MUSIC performance in detecting
two very-closely spaced jammers at a separation of 5° showed similar performance in
accuracy of jammer detection with an advantage of FOS having zero error detection in
both cases of JSR 15 dB and 45 dB and almost steady performance in resolving the two
jammers. On the other hand MUSIC was affected by the increase in noise level as it
failed to resolve the two jammers at JSR 15 dB and had an error of 3 degrees in detecting
the DoAs of jamming signals at JSR 15dB.
4.7.3 Results of Optimum Array
The procedures followed in uniform circular array experiments were used to
examine the performance of FOS versus MUSIC in jamming detection using optimum
array configuration. The configuration of optimum array is similar to UCA configuration
136
with 7 elements equally spaced on circumference of a circle with radius of 𝜆𝐿1
2 and an
additional element at the center of the circle.
4.7.3.1 Single Jammer
The results obtained for a single jammer arriving at an elevation of 400 and an
azimuth of 400are shown in Figure 4-33 to Figure 4-36. The figures show the normalized
output for, MUSIC and FOS DoA methods. Both methods had similar detection
performance of single jammer.
Figure 4-33: DoA estimation of one jammer using MUSIC at JSR=15 dB (optimum array)
01002003000 10 20 30 40 50 60 70 80 90
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
Azimuth (Degrees)
X: 40
Y: 40
Z: 1
FOS Number of Jammers=1 JSRdB
=15
Elevation (Degrees)
Norm
aliz
ed S
patial S
pectr
um
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
137
Figure 4-34: DoA estimation of one jammer using FOS at JSR=15 dB (optimum array)
Figure 4-35 DoA estimation of one jammer using MUSIC at JSR=45 dB (optimum array)
01002003000 10 20 30 40 50 60 70 80 90
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
Azimuth (Degrees)
X: 40
Y: 40
Z: 1
FOS Number of Jammers=1 JSRdB
=15
Elevation (Degrees)
Nor
mal
ized
Spa
tial S
pect
rum
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
020
4060
80100
0
100
200
300
400
0
0.2
0.4
0.6
0.8
1
Elevation (Degrees)
MUSIC Number of Jammers=1 JSRdB
= 45
X: 40
Y: 40
Z: 1
Azimuth (Degrees)
Norm
aliz
ed S
patial S
pectr
um
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
138
Figure 4-36: DoA estimation of one jammer using FOS JSR=45 dB (optimum array)
4.7.3.2 Multiple Jamming Signals
The results obtained for 3 jammers arriving at an elevation of (300, 400 𝑎𝑛𝑑 500 )
and azimuth angles of (300, 400 𝑎𝑛𝑑 500) respectively are shown in Figure 4-37 to
Figure 4-40. The figures below show the normalized output for MUSIC and 2D FOS
DoA methods. The performance of MUSIC in jamming detection using optimum array
configuration did not show significant improvement compared to UCA performance.
Conversely, FOS performance showed more significant improvement compared to
MUSIC when the central element was added at the center of UCA as described in the
optimum array configuration. The performance of FOS at JSR of 15 dB and 45 dB has
become almost steady in terms of resolution of jamming sources and power allocation for
each source as demonstrated in Table 4-8 and Table 4-9.
0100
200300
400
0 20 40 60 80 100
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
Azimuth (Degrees)
X: 40
Y: 40
Z: 1
FOS Number of Jammers=1 JSRdB
=45
Elevation (Degrees)
Nor
mal
ized
Spa
tial S
pect
rum
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
139
Figure 4-37: DoA estimation of 3 jammers using MUSIC at JSR=15 dB (optimum array)
Figure 4-38: DoA estimation of 3 jammers using FOS at JSR=15 dB (optimum array)
0 10 20 30 40 50 60 70 80 900
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
X: 48
Y: 48
Z: 1
Elevation (Degrees)
MUSIC Number of Jammers=3 JSRdB
= 15
X: 32
Y: 32
Z: 0.3793
Norm
aliz
ed S
patial S
pectr
um
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
0
200
0 10 20 30 40 50 60 70 80 90
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
Azimuth (Degrees)
Elevation (Degrees)
X: 50
Y: 50
Z: 0.7044
FOS Number of Jammers=3 JSRdB
=15
X: 40
Y: 40
Z: 1X: 29
Y: 29
Z: 0.704
Norm
aliz
ed S
patial S
pectr
um
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
140
Table 4-8: Simulation results of 3 closely spaced jammers using optimum array at
JSR=15dB
Jammer 1 2 3
Elevation 30° 40° 50°
Azimuth 30° 40° 50°
MUSIC (32°,32°) N/A (48°, 48°)
Error (2°, 2°) N/A (2°, 2°)
FOS (29°, 29°) (40°,40°) (50°,50°)
Error (1°, 1°) (0°, 0°) (0°, 0°)
Normalized Amplitude 0.704 1 0.7044
Figure 4-39: DoA estimation of 3 jammers using MUSIC at JSR=45 dB (optimum array)
0 10 20 30 40 50 60 70 80 900
200
400
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
X: 49
Y: 49
Z: 1
MUSIC Number of Jammers=3 JSRdB
= 45
Elevation (Degrees)
X: 39
Y: 39
Z: 0.3543
X: 30
Y: 30
Z: 0.2667
Norm
aliz
ed S
patial S
pectr
um
Azimuth (Degrees)
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
141
Figure 4-40: DoA estimation of 3 jammers using FOS at JSR=45 dB (optimum array)
Table 4-9: Simulation results of 3 closely spaced jammers using optimum array at
JSR=45dB
Jammer 1 2 3
Elevation 30° 40° 50°
Azimuth 30° 40° 50°
MUSIC (30°, 30°) (39°, 39°) (49°, 49°)
Error (0°, 0°) (1°, 1°) (1°, 1°)
Normalized Amplitude 0.2267 0.3543 1
FOS (29°,29°) (40°,40°) (50°,50°)
Error (1°, 1°) (0°, 0°) (0°, 0°)
Normalized Amplitude 0.9642 1 0.9642
0
2000 10 20 30 40 50 60 70 80 90
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
X: 50
Y: 50
Z: 0.9642
FOS Number of Jammers=3 JSRdB
=45
Elevation (Degrees)
X: 40
Y: 40
Z: 1
X: 29
Y: 29
Z: 0.9642
Azimuth (Degrees)
Norm
aliz
ed S
patia
l Spect
rum
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
142
Examining the performance of FOS versus MUSIC in resolving very closely spaced jammers:
A similar procedure to the one used on the UCA in 4.7.2 was applied to examine
the resolution capabilities of the proposed method versus MUSIC using the optimum
array configuration. Results are shown in Figure 4-41 to Figure 4-44 and in Table 4-10
and Table 4-11 for JSRs of 15 and 45 dB respectively. Similar to section 4.7.2 MUSIC
was not able to detect both the jammers at JSR of 15 dB and was only able to do so when
the JSR was higher at 45 dB
Figure 4-41: DoA estimation of 2 closely spaced jammers using MUSIC at JSR=15 dB
(optimum array)
0 10 20 30 40 50 60 70 80 900200
400
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
X: 43
Y: 43
Z: 1
Elevation (Degrees)
MUSIC Number of Jammers=2 JSRdB
= 15
Azimuth (Degrees)
Norm
aliz
ed S
patial S
pectr
um
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
143
Figure 4-42 DoA estimation of 2 closely spaced jammers using FOS at JSR=15 dB (optimum
array)
Table 4-10: Simulation results of 2 closely spaced jammers using optimum array at
JSR=15dB
Jammer 1 2
Elevation 40° 45°
Azimuth 40° 45°
MUSIC (43°,43°) N/A
Error (3°, 3°) N/A
FOS (40°,40°) (45°,45°)
Error (0°, 0°) (0°, 0°)
Normalized Amplitude 1 0.9225
02000 10 20 30 40 50 60 70 80 90
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
Azimuth (Degrees)
X: 45
Y: 45
Z: 0.9225
Elevation (Degrees)
FOS Number of Jammers=2 JSRdB
=15
X: 40
Y: 40
Z: 1
Nor
mal
ized
Spa
tial S
pect
rum
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
144
Figure 4-43: DoA estimation of 2 closely spaced jammers using MUSIC at JSR=45 dB
(optimum array)
Figure 4-44: DoA estimation of 2 closely spaced jammers using FOS at JSR=45 dB
(optimum array)
0 10 20 30 40 50 60 70 80 90
0200
400
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
Azimuth (Degrees)
Elevation (Degrees)
X: 45
Y: 45
Z: 1
MUSIC Number of Jammers=2 JSRdB
= 45
X: 40
Y: 40
Z: 0.5817
Nor
mal
ized
Spa
tial S
pect
rum
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
0
2000 10 20 30 40 50 60 70 80 90
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
Azimuth (Degrees)
X: 45
Y: 45
Z: 0.9974
Elevation (Degrees)
FOS Number of Jammers=2 JSRdB
=45
X: 40
Y: 40
Z: 1
Nor
mal
ized
Spa
tial S
pect
rum
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
145
Table 4-11: Simulation results of 2 closely spaced jammers using optimum array at JSR=45
dB
Jammer 1 2
Elevation 40° 45°
Azimuth 40° 45°
MUSIC (40°,40°) (45°,45°)
Error (0°, 0°) (0°, 0°)
Normalized Amplitude 0.5817 1
FOS (40°,40°) (45°,45°)
Error (0°, 0°) (0°, 0°)
Normalized Amplitude 1 0.9974
The above results showed similar improvement in comparison to UCA
performance for both FOS and MUSIC with slight improvement in the power allocation
of the FOS-based proposed method for jamming signals at JSR = 15 dB. MUSIC
performance at JSR of 15 dB was not improved by using optimum array configuration
and its detection capabilities for two very closely spaced jammers remained limited at
JSR of 15 dB. The results displayed in Table 4-11 demonstrate MUSIC’s drawbacks in
terms of power assignment and sensitivity to noise.
146
4.8 Conclusion
In this chapter, the DoA estimation for interference detection was studied. The main
factors impacting the process of interference detection through DoA were investigated in
order to design an optimum jamming detection system using antenna array. The
suggested optimum configuration is intended for the process of jamming detection and
can be also beneficial in jamming mitigation. A new procedure was introduced to
simulate GPS received signals using a single RF output Spirent GSS 6700 GPS simulator
and a single RF input Novatel FireHose frontend. Post-processing validated the GPS
signals used.
Several Antenna array configurations were reviewed to demonstrate their effect on DoA
estimation capabilities of antenna arrays. Furthermore, this section of the thesis
investigated the process of optimum array design using mathematical optimization. The
main contribution of this chapter is the application of nonlinear signal modeling
techniques for jamming detection utilizing ULAs and UCAs. The nonlinear signal
modeling technique employed for DoA estimation was based on FOS. The nature of
operation of the proposed method provides additional capabilities to DoA estimation.
This was shown in its capability of achieving steady performance in jamming detection at
different JSR’s. Moreover, when the optimum array configuration was studied, the
proposed method showed significant improvement in estimating the magnitude of the
spatial spectrum when compared to MUSIC. The proposed method consistently
outperformed MUSIC, which only started to show similar performance improvements
from the optimum array configuration when more powerful jamming signals were
simulated.
147
Spirent simulated GPS signals received by antenna arrays were used in the conducted
simulations to test the performance of the propose method against MUSIC and Classical
DoA estimation. Jammer signals were simulated using MATLB and added to the
simulated GPS signals. Configurations such as the ULA, UCA and the optimized UCA,
which is constructed by adding a central element to the circular antenna array, were
utilized in this study. The main benefit gained from applying FOS was to achieve higher
accuracy in interference DoA estimation in terms of magnitude and angle. This was
illustrated in single and multiple jamming detection scenarios for the aforementioned
array configurations. The proposed method demonstrated an improved ability to
accurately detect the amplitudes of the single jammers. As for multiple interference
detection, for ULA, the capabilities of jamming detection of Classical DoA are limited by
the resolution; MUSIC on the other hand is much more sensitive to noise and performs
worse at lower signal power. This became evident when the proposed method detected all
three jammers and MUSIC only detected two at relatively low JSR. The method also
demonstrated higher accuracy in detection of the 3 sources of interference with slight
variations in their amplitude at relatively low JSR 15 dB. Since the proposed method is
not optimized for real time operation, it is limited by its higher computational complexity
compared to the Classical method. Nevertheless, the added complexity is justified by the
high gains in accuracy. The proposed method is also limited by the number of elements in
the array and will therefore perform better and detect a higher number of closely spaced
jammers when applied to larger arrays.
148
Chapter 5
Conclusions and Recommendations
The enhancement of the anti-jamming capabilities of GPS receivers was the main
objective of this thesis. The research conducted herein focused on increasing the overall
detection accuracy of the GPS receivers in both the frequency and space domains by
introducing new methods that are based on high resolution signal processing. A brief
outline of the outcomes of this thesis along with the conclusions reached is given in this
chapter. Moreover, the future research that can extend and improve the work
accomplished by this thesis is presented in addition to the recommendations that are
based upon its findings.
5.1 Conclusions
To realize the objectives of this thesis, two new methods that are based on high
resolution signal processing techniques were introduced. The first method which operated
in the frequency domain and relied on spectral analysis was introduced for regular users
which constitute the vast majority of GPS receivers in mind. Using this method, these
receivers that only rely on one antenna can detect jammers with higher accuracy than that
provided by the predominant methods such as FFT. The second method targeted
professional grade anti-jamming equipment that takes advantage of the space domain.
The proposed method is based on high resolution DoA estimation of jammers in order to
provide smart antennas with the most accurate estimate of the jammers’ DoAs such that
beam steering and null steering processors nullify the jammer signals without erroneously
149
nullifying parts of the GPS signal thereby decreasing its effectiveness. These methods
were designed to operate in the pre-correlation section of GPS receivers where mitigation
is more effective.
An overview of the GPS system and receiver structures, followed by an
introduction to the types of interference, jamming and anti-jamming techniques was
provided in Chapter 2. It was concluded by a review of different DoA estimation
methods. Special interest was given to CW interference and jamming in Chapter 3, in
which an enhanced GPS narrowband jamming detection method, that uses high resolution
spectral estimation, was introduced. The proposed method is based on orthogonal search
and has featured several spectral estimation capabilities including high frequency
resolution, elimination of spectral leakage and rejection of background noise; which
subsequently resulted in a significant enhancement to the GPS jamming detection
process. The proposed method consistently demonstrated a twofold superiority over the
predominant FFT for both the accuracy of the jammers’ frequency estimation and the
ability to detect multiple closely spaced jammers.
FFT could only match the proposed detection method in specific cases. It
provided resolutions similar to that of the proposed method at an FFT window size which
was 10 times larger than that used by FOS. Moreover, our method accurately and
efficiently estimated the number and frequencies of multiple in-band CW jammers whose
frequency separation is lower than that of FFT resolution step. This twofold advantage
was as demonstrated in scenarios 2 and 3. Nevertheless, increasing the number of
jammers degrades the detection accuracy at lower JSRs. As previously demonstrated, at
low JSRs and using a 1 ms window, the proposed frequency domain method was only
150
able to detect the three closely spaced jammers (100 Hz separation between each other)
when their individual JSRs reached 27 dB. On the other hand, FFT’s disadvantage
became clear when it detected the three jammers as one jammer regardless of the
individual JSR levels. This inability to detect multiple jammers can only be improved by
increasing the window size.
The spatial domain aspect of the GPS jamming detection enhancement by DoA
estimation of jammers was tackled in Chapter 4. In addition to accurate DoA estimation,
the introduced method achieved high resolution detection of jammers whose spatial
separation is minimal; which is a major goal of this part of the thesis. The method was
studied, developed and tested for both one-dimensional and two dimensional arrays in the
form of ULA, UCA and optimum UCA. Its performance was compared to Classical DoA
estimation and MUSIC using several scenarios that differ in the number of jammers, their
spacing and their power levels. Additionally, a new method that enables the simulation of
multi-dimensional GPS antenna array systems using limited hardware resources was
developed and introduced in order to test the effectiveness of the proposed high
resolution DoA estimation method.
The proposed FOS-based method uses nonlinear signal modeling to provide the DoA
estimates. The method is based on an orthogonal search technique that creates a basis for
the steering function of the array. The steering function coefficients are determined via a
unique orthogonal search method that provides the high accuracy DoA estimates of the
jammers. When applying FOS, one candidate was generated per degree, so when 360
degrees were searched, FOS generated 360 candidates, which subsequently gave control
over the angle being searched. In addition, sub-degree resolution searches became
151
accessible. When these searches were conducted using Classical DoA estimation, a
multitude of DoA estimates that have the same magnitude was provided for the jammer
being search for.
The proposed method clearly outperformed the predominant methods in terms of
accuracy at lower JSRs and in terms of the number of closely spaced jammers detected as
previously demonstrated in Chapter 4. The proposed method prevailed against MUSIC in
both close-by jammers’ scenarios. When two jammers with a 5 degree separation were
being tested at 15 dB JSR, MUSIC detected them as only one jammer with a 40 percent
prediction error; in contrary to the proposed method which successfully detected the
presence of two jammers and accurately estimated their DoAs without error.
Additionally, in a more demanding scenario which involved three closely spaced
jammers, the proposed method demonstrated superiority by correctly detecting the
presence of all three, accurately estimating the DoAs of two jammers and erring in the
DoA estimation of the third jammer by one degree. MUSIC was only able to detect two
of the three jammers thus suffering a 33 percent detection error. Moreover, both
estimations incurred a two degree error which corresponds to a 50 percent higher error
for that jammer compared to the proposed method.
5.2 Recommendations and Future Work
The research conducted in this thesis along with the provided experimental results brings
forth the following recommendations:
152
1. The detection accuracy achieved in both the frequency and spatial domains by
using the proposed methods imply that further optimization and real-time
implementations of the methods are in order.
2. Since the proposed frequency domain method is designed to operate in the pre-
correlation section of the receiver, it can be easily implemented as a hardware
add-on structure to receivers or as a software upgrade to current software
receivers.
3. The proposed DoA estimation method merits the same attention for the
substantial improvements an anti-jamming system stands to gain by applying this
method, although it is not recommended for consumer grade GPS receivers due to
the supplementary cost incurred by the necessary added hardware. Nevertheless it
is highly recommended for military grade anti-jamming receivers.
4. It is highly recommended to use a central element with circular arrays and apply
the interference power minimization optimization because its yields more
accurate jammer amplitude estimation.
Based on the scope and findings of the thesis, the predicted future research work is
presented as follows:
1. In this thesis, only the GPS L1 signal has been experimented on, mainly due to
the fact that GPS is the most dominant and widespread GNSS. Other systems such
as Galileo, GLONASS and BeiDou operate on different frequencies; nevertheless,
they are still prone to jamming. As these systems become more established and
integrated into mainstream receivers, so will the jammers designed to impede the
receivers’ performance. Therefore, it is recommended that this work be expanded
153
into other GNSS frequencies and systems in addition to bearing in mind a unified
jamming detection platform that operates on multi-GNSS receivers.
Modifications to the methods proposed in this thesis should be applied when
considering other GNSSs, nonetheless, the approaches and the methodologies
followed remain similar. It is foreseeable that the application of this research to
different GNSSs will pose different challenges depending on the signals under
study.
2. The prevalence of CW interference and jamming is among the motivating factors
of this study. Unfortunately, other jamming techniques are steadily making their
way into the GPS jamming market, thus increasing the likelihood of GPS
jamming and complicating the anti-jamming receiver’s design and
implementation in order to combat such signals. Hence, the methods proposed in
this thesis should be applied to other jamming types (based on bandwidth and
modulation) that specifically target GPS and should be modified accordingly.
3. External aiding sensors such as INS are typically integrated on different levels
with GPS receivers in order to provide a position solution in case of GPS outages.
A study of the benefits of augmenting integrated GPS/INS systems with a
jamming detection and potentially a jamming mitigation system on the overall
performance of the receiver is deemed required and beneficial.
4. Since the DoA estimation was conducted on one and two dimensional arrays, the
proposed method should be extended to include three dimensional array
geometries, which would highly benefit the anti-jamming process by exploiting
154
the advantages provided by three dimensional arrays such as spatially mitigating
low elevation satellite multipath and similar direction jamming.
155
References
[1] K. Hegarty, Understanding GPS Principles and Applications, 2nd ed. Norwood: Artech
House Inc., 2006.
[2] J. Arribas, et al., Microwave and Millimeter CaveCircuits and Systems, Emerging design,
Technologies and Applications, A. Georgiadis, et al., Eds. Chichester,United Kingdom: John
Wiley & Sons, Ltd, 2013.
[3] D. Borio, C. O'Driscoll, and J. Fortuny, "Jammer impact on Galileo and GPS receivers," in
International Conference on Localization and GNSS, Turin, 2013.
[4] D. A. Goward, "Expert advice: the low cost of protecting America," GPS World, Jan. 2014.
[5] R. Bauernfeind, T. Kraus, A. S. Ayaz, D. Dötterböck, and B. Eissfeller, "Analysis, detection
and mitigation of in-car GNSS jammer interference in intelligent transport systems," in
Deutscher Luft-und Raumfahrtkongress, Berlin, 2012.
[6] U. Hunkeler, J. Colli-Vignarellii, and C. Dehollaini, "Effectiveness of GPS-jamming and
counter-measures.," in International Conference on Localization and GNSS (ICL-GNSS),
Starnberg, 2012.
[7] D. Borio, "A statistical theory for GNSS signal acquisition," PhD thesis, Politecnico Di
Torino, Torino, 2008.
[8] M. Abdizadeh, "GNSS signal acquisition in the presence of narrowband interference," PhD
Thesis, Geomatics Engineering, University of Calgary, Calgary, 2013.
[9] N. Kim, "Interference effects on GPS receivers in weak signal environments," M.Sc. Thesis,
Geomatics Engineering, University of Calgary, Calgary, 2006.
[10] Z. Jiang, "Mitigation of narrow band interference on software receivers based on spectrum
analysis," M.Sc., Geomatics Engineering, University of Calgary, Calgary, 2004.
[11] C. A. Balanis and P. I. Ioannides, Introduction to Smart Antennas. Phoenix: Morgan &
Claypool, 2007.
[12] J. C. Liberti and T. S. Rappaport, Smart Antennas for Wireless Communications: IS-95.
Upper Saddle River: Prentice Hall, 1999.
[13] S. Bellofiore, "Smart antenna systems formobile platforms," PhD Dissertation, Phoenix,
2002.
[14] D. H. Johnson, "The application of spectral estimation methods to bearing estimation
156
problems," in IEEE Proc., Vol.70, 1982, pp. 1018-1028.
[15] R. O. Schmidt, "Multiple emitter location and signal parameter estimation,” IEEE
Transactions," on Antennas and Propagation, , pp., 1986, vol. vol. AP-34, p. 276–280, 1986.
[16] W. Xu and M. ”. Kaveh, "Comparative study of the biases of MUSIC-like estimators,"
Signal Processing, vol. 50, pp. 39-55, 1996.
[17] A. J. Barabell, "Improving the resolution performance of Eigenstructure-based direction
finding algorithms," in IEEE International Conference on Acoustics, Speech and Signal
Processing, 1983, p. 336–339.
[18] R. Roy and T. Kailath, "ESPRIT-estimation of signal parameters via rotational invariance
techniques," IEEE Transactions on Acoustics, Speech and Signal Processing, vol. 37, no. 7,
pp. 984-995, Jul. 1989.
[19] G. Xu, S. D. Silverstein, R. Roy, and T. Kailath, "Beamspace ESPRIT," IEEE Transactions
on Signal Processing, vol. 42, p. 349–356, 1994.
[20] E. Misra, Global Positioning System, Signal, Measurements and Performance. Lincoln:
Ganga-Jamuna Press, 2006.
[21] El-Rabbany, Introduction to GPS the global positioning system. Norwood: Artech House,
2002.
[22] (2013, Jun.) Official U.S. Government information about the Global Positioning System
(GPS) and related topics. [Online]. http://www.gps.gov/systems/gps/space/
[23] G. Xu, GPS Theory, Algorithms and Applications. Berlin: springer, 2007.
[24] U. Iqbal, "Multi-sensor data fusion for vehicular navigation applications," PhD Thesis,
Electrical and Computer Engineering, Queen's University, Kingston, 2012.
[25] (2013, Jun.) Official U.S. Government information about the Global Positioning System
(GPS) and related topics. [Online]. http://www.gps.gov/systems/gps/space/
[26] J. A. Farrell, The Global Positioning System & Inertial Navigation. New York, NY:
McGraw-Hill, 1998.
[27] D. Doberstein, Fundamentals of GPS Receivers-A Hardware Approach. Nipomo, CA, USA:
Springer, 2012.
[28] B. W. Parkinson and J. J. Spiker, Global Positioning System: Theory and Applications.
AIAA, 1996, vol. 2.
157
[29] K. Borre, D. M. Akos, N. Bertelsen, P. Rinder, and S. Jensen, A Software-Defined GPS and
Galileo Receiver. Boston, USA: Birkhauser, 2007.
[30] J. Tsui, Fundamentals of Global Positioning Systems Receivers: A Software Approach. New
York: Wiley-Interscience, 2000.
[31] (2013, Jun.) Wikipedia. [Online]. http://en.wikipedia.org/wiki/GPS_signals
[32] Noureldin and Karamat, Fundamentals of Inertial Navigation,Satellite-based Positioning
and their Integration. Kingston: springer, 2013.
[33] C. O'Driscoll, "Performance analysis of the parallel acquisition of weak GPS signals," PhD
thesis, Electrical and Electronic Engineering, National University of Ireland, Cork, Ireland,
2007.
[34] Novatel Inc. www.novatel.com. [Online].
http://www.novatel.com/assets/Documents/Papers/GPS701_702GG.pdf
[35] T. Pany, Navigation Signal Processing for GNSS Software Receivers. Norwood, MA: Artech
House, 2010.
[36] O. Isoz, "Interference detection and localization in the GPS L1 frequency band," Lulea
University of Technology Licentiate Thesis, 2012.
[37] J. A. Farrell, Aided Navigation: GPS with High Rate Sensors. McGraw-Hill Professional,
2008.
[38] J. Nolan and e. al, "Geodetic processing using full dual band observables," in ION GPS-92,
Albuquerque, 1992, pp. 1033-1041.
[39] A. Balaei, "Detection, characterization and mitigation of interference in receivers for global
navigation satellite systems," Ph.D. Thesis, School of Surveying and Spatial Information
Systems, The University of New South Wales, Sydney, 2007.
[40] E.D.Kaplan, Understanding GPS: Principles and Applications. Norwood, MA.: Artech
House Inc., 1996.
[41] F. Dovis and L. Musumeci, "Use of the wavelet transform for interference detection and
mitigation in global navigation satellite systems," International Journal of Navigation and
Observation, vol. 2014, p. 14, 2014, Article ID 262186.
[42] D. Borio, C. O'Driscoll, and J. Fortuny, "GNSS jammers: effects and countermeasures," in
ESA Workshop on Satellite Navigation Technologies and European Workshop on GNSS
158
Signals and Signal Processing, Noordwijk, 2012.
[43] M. Jones, "The civilian battlefield: protecting GNSS receivers from interference and
jamming," Inside GNSS, Apr. 2011.
[44] H. Kuusniemi, E. Airos, M. Z. H. Bhuiyan, and T. Kröger, "GNSS Jammers: how vulnerable
areconsumer grade satellite navigation receivers?," European Journal of Navigation, vol. 10,
no. 2, pp. 14-21, 2012.
[45] M. J. Marcus, "Growing consumer interest in jamming: spectrum policy implications
[Spectrum Policy and Regulatory Issues]," IEEE wireless communications, p. 4, Feb. 2014.
[46] R. H. Mitch, et al., "Signal characteristics of civil GPS jammers," GPS world, Jan. 2012.
[47] S. Pullen, G. Gao, C. Tedeschi, and J. Warburton, "The impact of uninformed RF
interference on GBAS and potential mitigations," in International Technical Meeting of The
Institute of Navigation, Newport Beach, CA, 2012, pp. 780-789.
[48] S. M. Deshpande, "Study of interference effects on GPS signal acquisition," Ph.D Thesis,
Geomatics Engineering, University of Calgary, Calgary, 2004.
[49] A. Ndili and P. Enge, "GPS receiver autonomous interference detection," in IEEE Position,
Location and Navigation Symposium, Palm Springs, California, 1998.
[50] Y.-R. Chien, "Design of GPS anti-jamming systems using adaptive notch filters," IEEE
Systems Journal, vol. PP, no. 99, pp. 1-10, Oct. 2013.
[51] L. A. Davis, P. K. Enge, and G. X. Gao, Eds., Global Navigation Satellite Systems: Report of
a Joint Workshop of the National Academy of Engineering and the Chinese Academy of
Engineering. Washington, DC: The National Academies Press, 2012.
[52] R. Collins. (2013, Sep.) Integrated GPS Anti-Jam System (IGAS). [Online].
http://www.rockwellcollins.com
[53] B. D. Van Veen and K. M. Buckley, "Beamforming: a versatile approach to spatial filtering,"
IEEE ASSP Magazine, pp. 4-24, 1988.
[54] J. Li, "GPS interference mitigation for small UAV applications," M.Sc. Thesis, Electrical
and Electronic Engineering, The University of Adelaide, Adelaide, South Australia, 2009.
[55] M. Trinkle and D. Gray, "GPS interference mitigation; overview and experimental results,"
in 5th International Symposium on Satellite Navigation Technology & Applications,
Canberra, 2001, pp. 1-14.
159
[56] L. Scott, "Adaptive antenna arrays, multi-GNSS tropospheric monitoring, and high-dynamic
receivers," Inside GNSS, Apr. 2006.
[57] N. Inc. GAJT GPS Anti-Jam Technology. [Online]. http://www.novatel.com/gajt/overview/
[58] W. Lijun and M. Huanping, "The system model and analysis of all-digital temporal and
spatial anti-jamming for GPS receivers," in IET International Conference on Wireless,
Mobile and Multimedia Networks IET International Conference on, Beijing, 2010.
[59] S. Daneshmand, "GNSS interference mitigation using antenna array processing," Ph.D.
Thesis, Geomatics Engineering, University of Calgary, Calgary, 2013.
[60] J.Foutz and e. al, Narrowband Direction of Arrival Estimation for Antenna Arrays,
C.A.Balanis, Ed. Morgan & Claypool, 2008.
[61] C. F. Prades, "Advanced signal processing techniques for global navigation satellite systems
receivers," PhD Thesis, Signal Theory and Communications, Universitat Politecnica de
Catalunya, Barcelona, 2005.
[62] A. Massoud and A. Noureldin, "Direction of arrival estimation in passive sonar systems,"
Ph.D Thesis, Electrical and Computer Engineering, Queen's University, Kingston, 2012.
[63] L. C. Godara, Smart Antennas. Canberra: CRC press, 2004.
[64] P. Strobach, "Fast recursive subspace adaptive ESPRIT algorithms," IEEE transactions on
Signal Processing, vol. 46, pp. 2413-2430, 1998.
[65] L. Montloin, "Impact of interference mitigation techniques on a GNSS receiver,"
M.Sc.Thesis, Ecole Nationale de l'Aviation Civile, Toulouse, France, 2010.
[66] A. Brown, D. Reynolds, D. Roberts, and S. Serie, "Jammer and interference location system
– design and initial test results," in ION GPS, Nashville, TN, 1999, pp. 137-142.
[67] A. T. Balaei, A. G. Dempster, and J. Barnes, "A novel approach in detection and
characterization of CW interference of GPS signal using receiver estimation of C/No," in
Position, Location, And Navigation Symposium, 2006 IEEE/ION, 2006, pp. 1120-1126.
[68] D. Borio, "GNSS acquisition in the presence of continuous wave interference," IEEE
Transactions on Aerospace and Electronic Systems, vol. 46, no. 1, pp. 47-60, 2010.
[69] F. Nunes and F. Sousa, "Jamming detection in GNSS signals using the sample covariance
matrix," in Satellite Navigation Technologies and European Workshop on GNSS Signals and
Signal Processing, Noordwijk, 2012.
160
[70] A. T. Balaei and A. G. Dempster, "A statistical inference technique for GPS interference
detection," Aerospace and Electronic Systems, IEEE Transactions on, vol. 45, no. 5, pp.
1499-1509, 2009.
[71] Y.-R. Chien, "Hybrid successive continuous wave interference cancellation scheme for
global positioning system receivers," The Journal of Engineering, May 2013.
[72] A. Tani and R. Fantacci, "Performance evaluation of a precorrelation interference detection
algorithm for the GNSS based on nonparametrical spectral estimation," IEEE Systems
Journal, vol. II, pp. 20-26, 2008.
[73] Jiang, Z.; Ma, C. and Lachapelle , G., "Mitigatin of narrow-band interference on software
receivers based on spectrum analysis," in ION GNSS 2004, Long Beach, CA,, Sep.21-24,
2004.
[74] D. R. McGaughey, M. J. Korenberg, K. M. Adeney, S. D. Collins, and G. Aitken, "Using the
fast orthogonal search with first term reselection to find subharmonic terms in spectral
analysis," Annals of Biomedical Engineering, vol. 31, pp. 741-75, June 2003, v 31.
[75] M. J. Korenberg, "Identifying nonlinear difference equation and functional expansion
representations: The fast orthogonal algorithm," Annals of Biomedical Engineering, vol. 16,
no. 1, pp. 123-142, 1988.
[76] M. J. Korenberg, C. J. H. Brenan, and I. Hunter, "Raman spectral estimation via fast
orthogonal search," The Analyst, vol. 122, no. 1, pp. 879-882, 1997.
[77] A. Osman, A. Noureldin, N. El-Sheimy, J. Theriault, and S. Campbell, "Improved target
detection and bearing estimation using fast orthogonal search for real-time spectral analysis,"
Measurement Science and Technology, vol. 20, no. 6, p. 14, 2009.
[78] M. J. Korenberg and K. M. Adeney, "Iterative fast orthogonal search for modeling by a sum
of exponentials or sinusoids," Annals of bomedical engineering, vol. 26, pp. 315-327, Mar.
1998.
[79] E. Ifeachor and B. Jervis, Digital Signal Processing: A Practical Approach, 2nd ed. Essex,
England: Prentice Hall, 2001.
[80] M. J. Korenberg, "A Robust orthogonal algorithm for system indentification and time series
analysis," Biological cybernetics, vol. 60, pp. 267-276, 1989.
[81] A.Osman, "Coherent array processing of GPS sonobuoys," Ph.D Thesis, Department of
161
Geomatics Engineering, University of Calagry, Calgary, 2010.
[82] K. M. Adeney and M. J. Korenberg, "Fast orthogonal search for direction finding," IEE
Electronic letters, vol. 28, no. 25, pp. 2268-2269, 1992.
[83] Novatel Inc., "FireHose test setup and interface guide," Novatel Inc. Equimpent Manual,
2012.
[84] X. Wang, M. G. Amin, F. Ahmad, and E. Aboutanios, "Coarray based optimum geometries
for DOA estimation with multiple CRPA GPS arrays," in ION GNSS+, Tampa, FL, 2014.
[85] Q. Wang, L. Wang, K. An, Z. Shou, and H. Zhang, "DOA estimation of smart antenna signal
based on MUSIC algorithm," Journal of Networks, vol. Vol 9, no. 5, pp. 1309-1316, 2014.
[86] S. Applebaum, "Adaptive arrays," IEEE Transancations Antennas Propagation, vol. 24, pp.
585-598, 1976.
[87] V.L.Vasyiyshyn, "Antenna array signal processing with high-resolution by modified
beamspace music algorithm," in 6th International Conference on Antenna Theory and
Techniques, Sevastopol, 2007, pp. 455-457.
[88] K. M. Adeney and M. J. Korenberg, "Fast orthogonal search for array processing and
spectrum estimation," IEE Proceedings - Vision, Image and Signal Processing, vol. 141, no.
1, pp. 13-18, 1994.
[89] P. J. Bevelacqua, "Antenna arrays: performance limits and geometry optimization," Ph.D
Thesis, Electrical, Computer and Energy Engineering , Arizona State University, Tempe,
Arizona, 2008.
[90] K. G. Gromov, "Generalized interference detection and localization system," PhD Thesis,
2002.
[91] W. L. Stutzman, "Shaped-beam synthesis of nonuniformly spaced linear arrays," IEEE
Transactions on Antennas and Propagation, 1972.
[92] M. Gavish and A. J. Weiss, "Array geometry for ambiguity resolution in direction finding,"
IEEE Transactions on Antennas and Propagation, vol. 44, pp. 889-895, 1996.
[93] C. A. Balanis, Antenna Theory: Analysis and Design. New York: Wiley, 2005, 2005.
[94] N. H. Noordin, et al., "Uniform circular arrays for phased array antenna," in Loughborough
Antennas and Propagation Conference, Loughborough, 2011.
[95] A. Dalli, L. Zenkouar, E. F. Adiba, and S. Bri, "Circular array with central element for smart
162
antenna," Electrical and Electronic Engineering, vol. 3, no. 3, pp. 86-95, 2013.
[96] L. C. Godara and A. Cantoni, "Uniqueness and linear independence of steering vectors in
array space," The Journal of the Acoustical Society of America, vol. 70, no. 2, pp. 467-475,
1981.
[97] S. C. Clark, A. W. Morrison, and M. D. Guyse, "Practice safer GPS navigation – get
protection," 2005.