DETECTION, CHARACTERIZATION AND MITIGATION OF INTERFERENCE IN RECEIVERS FOR GLOBAL NAVIGATION SATELLITE SYSTEMS

Transcription

1 DETECTION, CHARACTERIZATION AND MITIGATION OF INTERFERENCE IN RECEIVERS FOR GLOBAL NAVIGATION SATELLITE SYSTEMS ASGHAR TABATABAEI BALAEI DOCTOR OF PHILOSOPHY THE UNIVERSITY OF NEW SOUTH WALES 2007

2 DETECTION, CHARACTERIZATION AND MITIGATION OF INTERFERENCE IN RECEIVERS FOR GLOBAL NAVIGATION SATELLITE SYSTEMS By ASGHAR TABATABAEI BALAEI B.Sc., Sharif University of Technology, Iran, 1997 M.Sc., Sharif University of Technology, Iran, 2000 A thesis submitted to The University of New South Wales in partial fulfillment of the requirements for the degree of Doctor of Philosophy DOCTOR OF PHILOSOPHY School of Surveying and Spatial Information Systems The University of New South Wales Sydney NSW 2052, Australia November, 2007

3 ABSTRACT GPS has become very popular in recent years. It is used in wide range of applications including aircraft navigation, search and rescue, space borne attitude and position determination and cellular network synchronization. Each application places demands on GPS for various levels of accuracy, integrity, system availability and continuity of service. Radio frequency interference (RFI) which results from many sources such as TV/FM harmonics, radar or mobile satellite systems, presents a challenge to the use of GPS. It can affect all the service performance indices mentioned above. To improve the accuracy of GPS positioning, a continuously operating reference station (CORS) network can be used. A CORS network provides all the enabled GPS users in an area with corrections to the fundamental measurements, producing more precise positioning. A threat to these networks is a threat to all high-accuracy GPS users. It is therefore necessary to monitor the quality of the received signal with the objective of promptly detecting the presence of RFI and providing a timely warning of the degradation of system accuracy, thereby boosting the integrity of GPS. This research was focused on four main tasks: a) Detection. The focus here is on a power spectral density fluctuation detection technique, in which statistical inference is used to detect narrowband continuous-wave (CW) interference in the GPS signal band after being captured by the RF front-end. An optimal detector algorithm is proposed. At this optimal point, for a fixed Detection Threshold (DT), probability of false alarm becomes minimal and for a fixed probability of false alarm, we can achieve the minimum value for the detection threshold. Experiments show that at this point we have the minimum computational load. This theoretical result is supported by real experiments. Finally this algorithm is employed to detect a real GPS interference signal generated by a TV transmitter in Sydney. b) Characterization. In the characterization section, using the GNSS signal structure and the baseband signal processing inside the GNSS receiver, a closed formula is derived for the received signal quality in terms of effective carrier to noise ratio ( ( C / No) ). This formula is tested and proved by calculating the C/No using the eff I and Q data from a software GPS receiver. For pulsed CW, a similar analysis is done to characterize the effect of parameters such as pulse repetition period (PRP) and also i

4 Abstract duty cycle on the received signal quality. Considering this characterization and the commonality between the GPS C/A code and Galileo signal as a basis to build up a common term for satellite availability, the probability of satellite availability in the presence of CW interference is defined and for the two currently available satellite navigation systems (GPS L1 signal and Galileo signal (GIOVE-A BOC(1, 1) in the E1/L1 band)) it is shown that they can be considered as alternatives to each other in the presence of different RFI frequencies as their availability in the presence of CW RFI is different in terms of RFI frequency. c) Mitigation. The last section of the research presents a new concept of Satellite Exclusion Zone. In this technique, using our previously developed characterization techniques, and considering the fact that RFI has different effects on different satellite signals at different times depending on satellite Doppler frequency, the idea of excluding the most vulnerable satellite signal from positioning calculations is proposed. Using real data and real interference, the effectiveness of this technique is proven and its performance analyzed. d) Hardware implementation. The above detection technique is implemented using the UNSW FPGA receiver board called NAMURU. ii

5 ACKNOWLEDGMENT This research was carried out from December 2004 through to November 2007 under the supervision of Associate Professor Andrew Graham Dempster and Professor Chris Rizos. I am sincerely grateful to them for their encouragement, invaluable advice and patient guidance throughout the course of this study. I would like to thank Professor Letizia Lo Presti from University of Politecnico di Torino for her helpful suggestions. I also wish to thank all members of the Satellite Navigation and Positioning (SNAP) group, Dr. Yong Li, Dr. Ravindra Babu, Dr. Joel Barnes, Dr. Steve Hewitson, Dr. Jinling Wang, Dr. Linlin Ge, Dr. Jack Wang, Dr. Binghao Li, Dr. Craig Roberts, Mr. Peter Mumford, Mr. Thomas Yan, Mr. Fabrizio Tappero, Mr. Bilal Amin, Mr. Usman Iqbal, Mr. Eamonn Glennon, Mr. Kevin Parkinson, Mr. Michael Chang, Ms. Jinghui Wu and Ms Beatrice Motella for their support in a variety of ways during this study. A small part of this work was performed in the NAVSAS group at the Politecnico di Torino and Instituto Superiore Mario Boella (ISMB) in Turin, Italy. I am grateful to Professor Letizia Lo Presti, Associate Professor Fabio Dovis and all the researchers in that group for their warm hospitality and support. Great thanks to Chris Baker and Mark Knight of the DSTO (Defense Science and Technology Organization), Australia for providing the testing and measuring environment for part of the experiments. I would also like to sincerely acknowledge the Cooperative Research Center for Spatial Information, Australia, for awarding me the postgraduate research scholarship to pursue my PhD. studies at the UNSW. Much gratitude to the U.S. Institute of Navigation (ION) for awarding me a student competition winner scholarship to support my attendance at the 19 th International Technical Meeting of the ION Satellite Division, held in Fort Worth, USA, September Thanks to European Institute of Navigation (ENC) for awarding me a best student paper award to partially support my attendance at the 11 th meeting of the ENC-GNSS, held in Geneva, Switzerland, 29 May 1 June In addition, my sincere appreciation to the Graduate Research School at the UNSW for awarding me a travel scholarship to support my attendance to the 20 th International Technical Meeting of the ION Satellite Division, held in Fort Worth, USA, September iii

6 Acknowledgment Last but far from least, I would like to extend my deepest appreciation to my family, especially my mother Fatemeh Sharifi, my father Hassan Tabatabaei Balaei and my wife Hadis Nosratighods for their love, encouragement and understanding during my PhD studies. iv

7 TABLE OF CONTENTS ABSTRACT...i ACKNOWLEDGMENT... iii TABLE OF CONTENTS...v LIST OF FIGURES...ix LIST OF TABLES...xiv ABREVIATIONS...xv Chapter INTRODUCTION Motivation Big Picture Contribution Structure of This Dissertation List of Publications...6 Chapter BACKGROUND Introduction GNSS System Control and Space Segment Receiver Block Diagram GNSS Acquisition and Tracking GNSS Signal Structure and Characterization Interference Threat Definitions Types of Interference Potential Sources of Interference (Landry, 1997) Characteristics of Interference Spatial-Directional Characteristics of RFI Temporal-Spectral Characteristics of RFI GNSS Receiver and Interference Interference and GNSS RF front end System Vulnerability to Interference...27 v

8 Table of contents GPS Receiver Interference Threshold CORS Networks Relationship Between Received Signal Quality and Positioning Quality User-to-Satellite Geometry Receiver Autonomous Integrity Monitoring (RAIM) Hypothesis Testing and Statistical Definitions Definitions Inference Based on Two Data Populations Previous Interference Detection, Characterization and Mitigation Techniques Detection Characterization of the Interference Effects Mitigation Localization...53 Chapter REAL GNSS SIGNAL INTERFERENCE Introduction Source and Effect Characterization In-band RFI Detection Summary...64 Chapter A STATISTICAL INFERENCE TECHNIQUE FOR GNSS INTERFERENCE DETECTION Introduction Problem Definition Mean and Variance of Test Statistic Mean Variance Experiments Discussion Real RFI Detection The RFI Source Discussion (Real Interference)...80 vi

9 Table of contents 4.6 Summary and Future Work...85 Chapter CHARACTERIZATION OF THE EFFECTS OF CW AND PULSE CW INTERFERENCE ON THE GPS SIGNAL QUALITY Introduction Analysis of the C/No after the Tracking Loop Experiments and Discussions The Effect of Integration Time on C/No in the Presence of CW RFI The Effect of Pulse CW RFI on the GPS Signal Quality Summary Chapter A PREVENTATIVE APPROACH TO MITIGATING CW INTERFERENCE IN GPS RECEIVERS Introduction The Effect of CW RFI on the GPS Signal Quality Positioning Quality Elements Satellite geometry and satellite signal quality, mathematical approach Mitigation Algorithm Experiment Discussion and Results The Impact of the HDOP Exclusion Zone RAIM Comparison Summary Chapter GNSS SATELLITE AVAILABILITY IN THE PRESENCE OF CW RFI Introduction Commonality of Signal Structure of different GNSS Systems Worst Spectral Line for GPS PRN1 and Galileo GIOVE-A Tracking Loop design Probability of Satellite Availability in the Presence of CW RFI Comparing the Effect of CW RFI on the GPS and Galileo Available Signals 143 vii

10 Table of contents RFI Frequency RFI Power Summary Chapter SUMMARY AND FUTURE WORK Introduction Summary and Conclusions Future Work REFERENCES Appendix A HARDWARE IMPLEMENTATION A.1 Hardware A.2 Software A.3 Experiments viii

11 LIST OF FIGURES Figure 2-1 GPS constellation...10 Figure 2-2 Principle of satellite navigation...11 Figure 2-3 Four satellites to reveal the time error...11 Figure 2-4 GNSS receiver block diagram (Maxim-IC, 2005)...14 Figure 2-5 Code and carrier tracking loop in a GNSS receiver...15 Figure 2-6 Phase Lock Loop (Viterbi, 1966)...16 Figure 2-7 Spreading the data over the C/A code spectrum bandwidth...18 Figure 2-8 Interference and background noise is added to make the final GPS signal received at the antenna...18 Figure 2-9 Code is despread by being multiplied by the receiver code replica and interference is spread...19 Figure 2-10 Tracking loop low pass filter: extracts the data and the interference which is outside the filter bandwidth...19 Figure 2-11 Galileo BOC(1,1) modulation scheme...20 Figure 2-12 GIOVE-A Signal generation for E1 band BOC (1,1)...21 Figure 2-13 Tail probability and p-value...39 Figure 3-1 The distance between Sydney city centre and Hampden Road, Artarmon, where the data collection took place from Google Earth...55 Figure 3-2TV Tower in Hampden Road - Artarmon (left) and the detection of the TV transmission (at MHz) by means of the spectrum analyzer (right)...56 Figure 3-3 The signal transmitted by the TV broadcaster antenna at the GPS L1 frequency (peaks indicate the interference)...57 Figure 3-4 Spectrum of the received signal at different times (at 1 15, 1 50 and 2 55 from the beginning of the data collection respectively)...58 Figure 3-5 Height of the spikes w.r.t. the noise level vs time (above) AGC gain vs time (below)...59 Figure 3-6 Spectrum average within a band of 1 MHz around the central frequency (above) AGC level versus time (below)...60 Figure 3-7 Interference-affected data collected in a location from where the main lobe of the transmission is received at a specific time (time = 129s into the collected data sequence)...61 ix

12 List of figures Figure 3-8 The hardware setup to characterize the behavior of AGC level of NordNav software receiver...62 Figure 3-9 Characterization of the behavior of AGC level w.r.t the received in-band power in the NordNav software receiver...62 Figure 3-10 Normal histogram of the IF data when the AGC is in its linear operation...63 Figure 3-11 Histogram of the IF data when the AGC is saturated by interference...63 Figure 3-12 AGC level in the strongly interfered area versus time...64 Figure 4-1 Process of generating the samples for each frequency bin of the two populations...66 Figure 4-2 Parameters affecting the significance of difference between the two samples...69 Figure 4-3 The points in which each of the four experiments are done...72 Figure 4-4 Computational load for experiment one...74 Figure 4-5 Computational load for experiment two...75 Figure 4-6 Computational load for experiment three...77 Figure 4-7 Trade off zone where the lower power detectable interference can be achieved with the price of processing load...79 Figure 4-8 p-value across the bandwidth in the presence of strong interference in a specific time...80 Figure 4-9 Power Spectral Density of the Received Signal at time...80 Figure 4-10 log10 (p-value) versus time and frequency...81 Figure 4-11 P-value across the L1 bandwidth for time 23 sec...82 Figure 4-12 p-value across time in frequency MHz (for DB = 128*1024 and SS = 8)...82 Figure 4-13 p-value across time in frequency MHz (for DB=32*1024 and SS=32)...83 Figure 4-14 p-value across time in frequency MHz (for DB = 1024 and SS = 1024)...84 Figure RF down-conversion and band-pass sampling diagram...85 Figure 5-1 Correlator (Code and Carrier Tracking Loops)...88 Figure 5-2 C/No calculated using the mathematical expression (5-14) for satellite 1 with Doppler frequency changing from 0 khz to 10 khz and CW interference at 14 khz away from the band center at GHz...94 x

13 List of figures Figure 5-3 Hardware setup for the experiments from left: Spectrum analyzer, RF signal generator, NordNav front-end, GPS signal generator Figure 5-4 C/No calculated using the power ratio technique for satellite 1 with Doppler frequency changing linearly with time from 0 to 9 khz and CW interference at 14 khz away from the band center at GHz...96 Figure 5-5 One of the troughs where RFI coincides with a code spectral line in the actual measurement...97 Figure 5-6 One of the troughs where RFI coincides with a code spectral line in the theoretical measurement...98 Figure 5-7 C/No calculated using the power ratio technique for satellite 1 with Doppler frequency changing from -4 khz to 4 khz and CW interference at the band center at GHz...99 Figure 5-8 C/No plots calculated using both the parametric method and power ratio method Figure 5-9 The difference between the estimated C/No and the theoretically calculated one Figure 5-10 (a,b) C/No of PRN1 when interference is at 4.5 khz for a) T d = 16ms and b) T d = 8ms Figure 5-11Sinc functions associated with the pulse interference (black) and the integrator block of the tracking loop (red) Figure 5-12 Time domain representation of code despreading and integrating blocks Figure 5-13 Frequency domain representation of code despreading and integrating blocks Figure 5-14 Sinc functions associated with the pulse interference (black) and the integrator block of the tracking loop (red) Figure 5-15 C/N 0 calculated using the mathematical expression for satellite 1 with Doppler frequency changing from 0 khz to 1 khz and pulsed CW interference at 3 khz away from the band center at GHz with different duty cycles Figure 5-16 Actual C/No measured by the receiver for different duty cycles Figure 5-17 Sinc functions associated with the pulse interference (black) and the integrator block of the tracking loop (red) xi

14 List of figures Figure 5-18 C/No calculated using the mathematical expression for satellite 1 with Doppler frequency changing from 0 khz to 1 khz and pulsed CW interference with DC= 20% at 3 khz away from the band center at GHz with varying pulse lengths Figure 5-19 Actual C/No measured by the receiver for different pulse periods Figure 5-20 C/No calculated using the mathematical expression for satellite 1 with Doppler frequency changing from 0 khz to 1 khz and pulsed CW interference with Tp = 0.1s at 3 khz away from the band center at GHz Figure 5-21 Actual C/No measured by the receiver for different duty cycles Figure 5-22 Sinc functions associated with the pulse interference (wide) and the integrator block of the tracking loop (narrow) Figure 5-23 C/No calculated using the mathematical expression for satellite 1 with Doppler frequency changing from 0 khz to 1 khz and pulsed CW interference with Tp = s at 3 khz away from the band center at GHz Figure 5-24 Actual C/No measured by the receiver for different duty cycles Figure 6-1 C/No calculated using the mathematical expression for satellite 1 with Doppler frequency changing from 0 khz to 10 khz and CW interference at 14 khz away from L1 frequency Figure 6-2 Variation of Doppler frequencies for the visible satellites over 24 hours. Exclusion zones are indicated at multiples of 1 khz Figure 6-3 Relation between trough depths in C/No calculated theoretically and the corresponding satellite exclusion zone Figure 6-4 Actual C/No calculated using the I and Q samples (red) and Theoretical C/No calculated using 1 (blue) for PRN 1 and interference at 4 khz away from L1 121 Figure 6-5 C/No for four different values of RFI power (magenta, black, blue and red for -82, -85, -88, -91 dbm respectively) 40 db-hz chosen as threshold Figure 6-6 Position error vs. pseudorange error for the 4 and 5 satellite configuration at a particular time for satellite 11 which is important to the geometry of the 5-satellite constellation Figure 6-7 Position error vs pseudorange error for the 4 and 5 satellite configuration for satellite 25 which contributes little to constellation geometry Figure 6-8 Algorithm description flow chart xii

15 List of figures Figure 6-9 Position error vs. time (a) and scatter diagram (b). The comparison is between the 5 satellites configuration (blue line) and the 4 satellite one (red line) when the interference affects the satellite 1 (kept out in the 4 sat configuration) Figure 6-10 C/No for the PRN 11 (a) HDOP for the 5 and 4 satellites configuration (b) Figure 6-11 Comparison between the positioning error using 5 and 4 satellites in presence of interference (a) Comparison between the positioning error using the 5 satellites constellation with and without interference (b) Figure 6-12 (a) Doppler frequency for PRN 1 (b) Exclusion zone for PRN Figure 6-13 Position error for 6 satellites, one of which is affected by interference scatter plot (a) Positioning error before and after the exclusion zone (b) Figure 6-14 Maximum distance between positioning from different satellite configuration for the 5 satellite set, one affected by interference Figure 7-1 GPS signal generation for L1 band C/A code Figure 7-2 Galileo BOC(1,1) modulation scheme a) Simulating pure satellite signal b) signal plus noise and interference Figure 7-3 PRN spectrum and their worst lines. GPS PRN1 spectrum (a) and its 42 khz frequency component (b); Galileo C-PRN (Square wave modulated with C-code) spectrum (c) and its 771 khz frequency component (d) Figure 7-4 Definition of probability of satellite availability Figure 7-5 GPS and GIOVE-A probability of availability versus frequency of CW RFI RFI Figure 7-6 GPS and GIOVE-A probability of availability versus power of CW RFI Figure A-1 Namuru circuit board Figure A-2. RF down-conversion and band-pass sampling diagram Figure A-3. Detailed hardware block diagram Figure A-4. Output of 512 bin FFT engine with L1 center frequency input Figure A-5. Frequency sweep around MHz near the Artarmon transmitter tower Figure A-6. Example of statistical result output to a console window Figure A-7. Example of zoom processing to console window xiii

16 LIST OF TABLES Table 2-1. Various types of RF interference...23 Table 2-2 Mobile and stationary transmitters...26 Table 2-3 GPS Susceptibility Threshold...29 Table 2-4 Interference threshold versus interference bandwidth for GPS receivers and for SBAS and GBAS air navigation receivers in track mode...30 Table 2-5 Level of significance for a directional and non-directional test with degree of freedom (df) = Table 4-1 Minimum detectable interference for experiment one...73 Table 4-2 Mean of samples for experiment one...73 Table 4-3 Standard deviation of samples for experiment one...73 Table 4-4 Minimum detectable interference for experiment two...74 Table 4-5 Mean of samples for experiment two...75 Table 4-6 Standard deviation of samples for experiment two...75 Table 4-7 Minimum detectable interference for experiment three...76 Table 4-8 Mean of samples for experiment three...76 Table 4-9 Standard deviation of samples for experiment three...76 Table 4-10 Simulation and the theoretically calculated p-values...78 Table 6-1 Exclusion zone widths for four different RFI powers for two consecutive C/A code spectral lines Table 6-2 HDOP for the 5 and 4 satellite configuration Table 6-3 Maximum position error before and after applying the mitigation technique Table 7-1 Tracking Loop setting for different data Table A-1 Detection of -100dBm zoom Table A-2. Detection of -110dBm zoom Table A-3. Detection of -110dBm zoom Table A-4. Frequency resolution at zoom xiv

17 ABREVIATIONS ABA Australian Broadcasting Authority ADC Analog-to-Digital Converter AGC Automatic Gain Control AOA Angle of Arrival ATC Air Traffic Control AWGN Additive White Gaussian Noise bps bits per second BPSK Binary Phase Shift Keying C/A Course Acquisition C/A code Course Acquisition code CAT Category cdf Cumulative Distribution Function CDMA Code Division Multiple Access CLT Central Limit Theorem C/No Carrier to Noise Density Ratio CORS Continuously Operating Reference Station CW Continuous Wave dbw decibels relative to one Watt db decibels (logarithmic measurement of power or gain ratios) dbm decibels relative to one milliwatt deg Degree DGPS Differential Global Positioning System DoD Department of Defense DOP Dilution of Precision DSP Digital Signal Processing DT Detection Threshold ENC European Navigation Conference FAA Federal Aviation Administration FFT Fast Fourier Transform FM Frequency Modulation FPGA Field Programmable Gate Array GDOP Geometric Dilution of Precision xv

18 Abreviations GNSS Global Navigation Satellite System GPS Global Positioning System H 0 Null Hypothesis H A Alternative Hypothesis Hz Hertz (cycle per second) IEEE Institute of Electrical and Electronics Engineers, Inc. IF Intermediate Frequency IID Independent Identically Distributed ILS Instrument landing System INS Inertial Navigation System ION Institute of Navigation L-band All frequencies between 1 and 2 GHz L1 first L-band GPS frequency: MHz (154x10.23 MHz) LAAS Local Area Augmentation System LNA Low Noise Amplifier LOS Line of Sight LPF Low Pass Filter m meter MDB Minimum Detectable Bias MHz megahertz (millions of cycles per second) MLS Microwave Landing System MOPS Minimum Operational Performance Standards NBP Narrow Band Power PAL Phase Alternation Line P Code Precise Code pdf Probability Density Function P fa Probability of False Alarm P MD Probability of Miss-Detection PPS Pulse per Second PR Pseudorange PRN Pseudo Random Noise PSD Power Spectral Density QM Quality Monitoring xvi

19 Abreviations RF Radio Frequency RFI Radio Frequency Interference RV Random Variable SA Selective Availability SBS Special Broadcasting Service SDR Software Defined Radio SNR Signal to Noise Ratio SPS Standard Position Service SQM Signal Quality Monitoring SV Space Vehicle TCXO Temperature Control Crystal Oscillator TDOA Time Difference of Arrival UHF Ultra High Frequency VHF Very High Frequency WAAS Wide Area Augmentation System WBP Wide Band Power WSS Wide Sense Stationary xvii

20 Chapter 1 INTRODUCTION 1.1 Motivation In recent years there has been a rapid growth in the utilization of Global Navigation Satellite Systems (GNSS) across a diverse range of application areas. However, coupled with this expansion has been the expansion in awareness that GNSS is potentially vulnerable to sources of interference. This has caused particular concern in the domain of transportation, as indicated for example by the US Department of Transport s undertaking of a report entitled Vulnerability Assessment of the Transportation Infrastructure Relying on GPS (Volpe report, 2001). Whilst this report dealt specifically with the issue of GPS usage within the transport infrastructure of the US, the fact that other applications share the same reliance on GPS, means that the vulnerability remains equally valid for all of them. Mobile communication networks based on the cdmaone and cdma2000 standards require that the base-stations of their radio access networks be synchronized (Schneuwly, 2002). All the base-stations of the network need a 1 Pulse Per Second (lpps) phase reference with an accuracy of 3 μs (Schneuwly, 2002). The Global Positioning System (GPS) is currently the only practical way of implementing this type of synchronization and thus these networks are also vulnerable to GPS interference (Khan, 2007a; Khan, 2007b). GPS has a variety of applications on land, at sea and in the air. Basically, GPS is usable everywhere except where it is difficult to receive the signal, such as inside most buildings, in caves and other subterranean locations, and underwater. The most common airborne applications are for navigation by general aviation and commercial aircraft. At sea, GPS is also typically used for navigation by recreational boaters, commercial fishermen, and professional mariners. Land-based applications are more diverse. The scientific community uses GPS for its precision timing capability and position information (Parkinson, 1996a; Bullock, 1997). Surveyors use GPS for an increasing portion of their work. GPS offers cost savings by drastically reducing setup time at the survey site and providing the required accuracy. Basic survey units, costing hundreds of dollars, can offer 1

21 Chapter 1 Introduction accuracies down to one meter. More expensive systems are available that can provide accuracies to within a centimeter (Uren, 2006). Recreational uses of GPS are almost as varied as the number of recreational sports. GPS is popular among hikers, hunters, snow-mobilers, mountain bikers, and cross-country skiers, just to name a few. Anyone who needs to keep track of where he or she is, to find his or her way to a specified location, or know what direction and how fast he or she is going can utilize the benefits of the global positioning system. It is now commonplace in automobiles as well. Some basic systems are in place and provide emergency roadside assistance at the push of a button (by transmitting your current position to a dispatch center). More sophisticated systems that show your position on a street map are also available. Currently these systems allow a driver to keep track of where he or she is and suggest the best route to follow to reach a designated location (Navman; Tomtom). In all of the applications of the satellite navigation systems, accuracy and also reliability in the sense of integrity, availability and continuity is essential. Interference affects all of these performance parameters. The Volpe report made some recommendations on how to continue the ever increasing reliance on GPS safely. These recommendations effectively tend to suggest increases in these performance parameters. They include continuing the GPS program (i) with higher broadcast power and (ii) with the eventual availability of other civil frequencies. It also recommends installing systems to monitor, report and locate interference, to assess the applicability of military anti-jamming technology and to work with industry to make the technology available for civil users. 1.2 Big Picture GPS Continuously Operating Reference Station (CORS) networks now form the backbone of national and geodetic infrastructure in most developed countries. Geoscience Australia has operated the Australian Regional GPS Network, which comprises 15 permanent GPS receivers for almost 10 years. This continent-wide network forms the legal basis of all geodetic infrastructure in Australia. The output of this research contributes to the project which will enhance national and regional infrastructure of networks of CORS receivers. This will support the operation of the entire Australian spatial information industry at local, regional and national levels into 2

22 Chapter 1 Introduction the next decade. This project aims to target problems that presently exist in CORS network site installation and the quality of the raw data derived from CORS networks concentrating on signal interference and multipath. In this thesis, the problem of detection and characterization of the GNSS signal interference source is addressed. GNSS signal interference affects each one of the four system performance metrics of the GNSS system (accuracy, integrity, availability and continuity). These metrics will be defined in the introductory chapter. In order to improve these metrics in the presence of interference there are different approaches. Detection of interference improves integrity. Mitigation of interference improves accuracy. Source localization of interference improves availability of the system. The two main researches that have each addressed one of these approaches are Marti (2004), who addressed the detection approach and Gromov (2002), who addressed the source localization approach. Also Ndili (1998) has extensively investigated the characterization of the effects of different types of interference on different parts of the GNSS receivers. It was shown in Ndili (1998) that Automatic Gain Control in the RF front-end can be affected by interference no matter what the type of the interference is. So it could be used to help detection of interference (Ward, 2007). Characterization of the effects can help mitigate them too. Not all the different types of interference are shown to have the same sort of effect. Continuous Wave interference was shown to have the strongest effect on the operation of the receiver. Particularly if it is narrow enough to pass entirely from the tracking loop filters in the receiver. These researches have all been conducted for the application of GNSS system in the aviation industry. For the first time in this thesis, the issue of GNSS signal interference has been addressed for the application of CORS networks. In the CORS network application, the level of accuracy which is expected from the operation of the GNSS system is very high. However, unlike the aviation industry, CORS network application has far less dynamic when it comes to positioning which allows having longer Time to Alert (TTA) in the detection approach. This specification also allows us to mitigate the effect of the interference by extracting its power and frequency information and predict its effect on the GNSS received signal quality. 1.3 Contribution 3

23 Chapter 1 Introduction The contribution of this research can be classified in three domains. The first is related to interference detection using Intermediate Frequency (IF) data blocks received by a software defined radio receiver design. It is first shown that based on how the data block is broken into sub-blocks, detection threshold changes with a fixed probability of false alarm. It is then shown that there is an optimal way to break this data window to achieve the lowest detection threshold for a fixed value of probability of false alarm in which the computational load is also minimal. Then the optimal detector algorithm is proposed. It is shown that only in this optimal point, for a fixed Detection Threshold (DT), probability of false alarm becomes minimal and for a fixed probability of false alarm, we can achieve the minimum value for the detection threshold and experiment shows that in this point we have the minimum computational load. In the second domain, which is characterization of the effect of interference on the received signal quality, by considering a very narrowband RFI compared to the reciprocal of the correlation integration, an expression for the effective C/No is derived. This new expression which has been driven after the correlation, shows that the value of the effective C/No depends on the accuracy of the operation of the code tracking loop on the one hand and on the difference between the frequency of the RFI and the carrier frequency on the other hand. It also depends on the integration duration time of the carrier tracking loop. This result is approved by calculating the C/No using the I and Q data from a software GPS receiver. For the pulse CW, a similar analysis is done to characterize the effect of parameters such as pulse repetition period (PRP) and also duty cycle on the received signal quality. It is specifically shown that for equal interference power level, in the cases where the PRP is far less than the pseudorandom noise code period, then the signal degradation increases with increasing the duty cycle whereas it does not change if the two periods are equal or the PRP is far bigger than the code period. Considering this characterization and the commonality between the GPS C/A code and Galileo signal as a basis to build up a common term for satellite availability, the probability of satellite availability in the presence of CW interference is defined and for the two currently available satellite navigation systems (GPS L1 signal and Galileo signal (GIOVE-A BOC(1, 1) in the E1/L1 band)) it is shown that they can be considered as 4

24 Chapter 1 Introduction alternatives to each other in the presence of different RFI frequencies as their availability in the presence of CW RFI is different in terms of RFI frequency. In the third domain, which is the RFI mitigation domain, it is first explained how the characterization of the effect of CW interference can be used in introducing a new algorithm to mitigate the effects of CW RFI on the positioning evaluation. The new concept of exclusion zone is defined and analyzed for each satellite. This exclusion zone is shown to be predictable for each satellite as a function of time. Using this prediction, the CW interference effect on the positioning quality of the receiver is mitigated by cancelling the affected satellite within the exclusion zone during the position evaluation. Then the decision of canceling a particular satellite (the threshold beyond which a satellite should be excluded), is worked out by studying the mutual effect of the geometry and the signal quality of that satellite on the positioning quality. 1.4 Structure of This Dissertation This report is organized in eight chapters. Chapter 1 is an introduction and briefly introduces motivation, pig picture, contributions and publications resulting from this research. Chapter 2 introduces the concept of GPS positioning, discusses the various positioning techniques and lists some of the sources of error encountered in GPS measurements. Interference as one of those threats is defined. Also different approaches in dealing with interference are presented. Literature review and the necessary means and tools used in this dissertation are also presented in this chapter. In Chapter 3, a real GPS L1 band signal interference generated by a TV transmitter in Sydney, Australia is introduced. Chapter 4 focuses on the interference detection scheme which is based on statistical inference. In Chapter 5, the effect of CW and pulsed CW interference on received GPS satellite signal quality is characterized. Chapter 6 is dedicated to a present a preventative mitigation algorithm which is based on the characterization concepts introduced in Chapter 5. 5

25 Chapter 1 Introduction In Chapter 7, extending the characterization of the effect of CW RFI on the received signal quality done in Chapter 5, the availability of the GPS and Galileo satellite signal are compared in the presence of CW RFI. The conclusions are drawn in Chapter 8 and a few recommendations are made for the future studies. The detection algorithms presented in Chapter 4 are implemented in UNSW FPGA receiver and the results are presented in Appendix A. 1.5 List of Publications The results of this research have been published or submitted for publication as 4 journal articles and 11 conference papers. 1. Tabatabaei Balaei, A., Motella, B., Dempster, G.A., (2007d) A Preventative Approach to Mitigating CW Interference in GPS Receivers accepted for publication in GPS Solution October In this paper of which the results are presented in Chapter 6, all the ideas and the background theory are developed by the author and Beatrice Motella helped to conduct some of the experiments. 2. Tabatabaei Balaei, A, Dempster, G.A., (2006e) A Statistical Inference Technique for GPS Interference Detection Submitted to the IEEE Transaction on Aerospace and Electronic systems, November We have received the reviewers comments for this paper and sent the updated version for the journal. 3. Tabatabaei Balaei, A., Dempster, G.A., Lo Presti, L., (2007a) Characterization of the effects of CW and pulse CW interference on the GPS signal quality submitted to IEEE Transaction on Aerospace and Electronic systems March 4. Motella, B., Tabatabaei Balaei, A., Lo Presti, L., Leonardi, M., Dempster, G.A. (2006) Characterization of Radar Interference Sources on Galileo E6 Band Submitted to the IEEE Transaction on Aerospace and Electronic systems, November In this paper the contribution of Tabatabaei is to characterize the effect of chirp type interference on the received signal quality and the rest of the work has been done by Beatrice Motella 6

26 Chapter 1 Introduction 5. Tabatabaei Balaei, A., Wu, J., Dempster G.A., (2007f) Comparison Between GPS and Galileo Satellite Availability in the Presence of CW Interference Proceedings of IGNSS, Australia, December In this paper all the ideas and the background theory is developed by the author and Jinghui Wu has helped to conduct the experiments. 6. Tabatabaei Balaei A., Motella, B., Dempster, G.A., (2007e) GPS Interference Detected in Sydney- Australia Proceedings of IGNSS, Australia, December In this paper, Beatrice Motella helped collecting data and also investigated the effects of interference on the AGC. 7. Mumford, P., Tabatabaei Balaei, A., Dempster, G.A., (2007) UNSW GNSS Interference Detection Unit Proceedings of IGNSS, Australia, December In this paper Tabatabaei helped with designing all the detection algorithms and some of the implementation algorithms and Peter Mumford did all the implementation. 8. Dempster, G.A., Mumford, P., Parkinson, K., Qaisar, S., Wu, J., Tabatabaei Balaei, A., Glennon, E., (2007) NAMURU, a platform for GNSS research, Location April In this paper the author helped with designing all the detection algorithms and some of the implementation algorithms and Peter Mumford did all the implementation. 9. Tabatabaei Balaei, A., Motella, B. (2007b) Satellite exclusion zone in the presence of CW interference Experimental results ENC-GNSS Geneva, Switzerland, May-June In this paper of which the results are presented in Chapter 6, all the ideas and the background theory is developed by Tabatabaei and Beatrice Motella helped to conduct some of the experiments. This paper won the student paper prize in the 11 th conference of the ENC-GNSS, held in Geneva, Switzerland, 29 May 1 June Tabatabaei Balaei, A., Motella, B., Dempster, G.A. (2007c) Mutual effects of satellite signal quality and satellite geometry on positioning quality ION/GNSS September, Fort Worth, USA 7

27 Chapter 1 Introduction 11. Tabatabaei Balaei, A., Motella, B., Dempster, G.A. (2006d) Exclusion zones for GNSS signals when reconfiguring receiver hardware in the presence of narrowband RFI. IAIN / GNSS October, Korea In the last two papers, of which the results are presented in Chapter 6, all the ideas and the background theory is developed by the author and Beatrice Motella helped to conduct some of the experiments. 12. Tabatabaei Balaei, A. (2006c) Statistical Inference Technique in Pre- Correlation Interference Detection in GPS Receivers. ION/GNSS September, Fort Worth, USA. This paper won the student paper prize in the ION Satellite Division, held in Fort Worth, USA, September Tabatabaei Balaei, A., Dempster, G.A., Barnes, J. (2006a) A novel approach in detection and characterization of CW interference of GPS signal using receiver estimation of CNo PLANS( ION IEEE)- April 14. Tabatabaei Balaei, A., Dempster, G.A., Barnes J. (2006b) Application of Post-Correlation Interference Detection and Characterization of the GPS Receivers in the Receiver Reconfigurability IGNSS July 15. Tabatabaei Balaei, A., Barnes, J., Dempster, A.G., (2005) Characterization of interference effects on GPS signal carrier phase error SSC Melbourne 8

28 Chapter 2 BACKGROUND 2.1 Introduction The purpose of this chapter is to present all the necessary technical concepts, which are later used for the development of the techniques and algorithms introduced in this dissertation and the literature relevant to the subject of this dissertation. The structure of this chapter is as follows. Section 2.1 covers the GNSS system components and signal characterization and receiver block diagram. In section 2.3, the background definition regarding interference types and sources is introduced. The threat of signal interference for the GNSS receiver is investigated in section 2.4. In section 2.5 CORS networks are defined. The relationship between positioning quality and received signal quality is discussed in section 2.6. Analysis of user-to-satellite geometry is presented in section 2.7. In section 2.8 the concept of receiver autonomous integrity monitoring (RAIM) is presented as we later compare its performance in mitigating CW RFI with the preventative CW RFI mitigation technique of Chapter 6. Section 2.10 is dedicated to introducing a background for statistical inference techniques, which have been used for interference detection in Chapter 4 and finally this chapter will finish with covering all the algorithms and techniques in the literature in the four main approaches in dealing with GNSS signal interference. 2.2 GNSS System The Global Navigation Satellite System (GNSS) works on the principle that if you know your distance from several locations, then you can calculate your location. In the US developed GNSS system, the Global Positioning System (GPS), the known locations are the 32 satellites located in six orbital planes at an altitude of 20,200Km (see Figure 2-1). These satellites circle the Earth every 12 hours and broadcast a data stream at the primary frequency L1 of 1.575GHz which carries the coarse-acquisition (C/A) encoded signal to the ground. The GPS receiver measures the time of arrival of the C/A code to a fraction of a millisecond, and thus determines the distance to the satellite. 9

29 Chapter 2 Background Figure 2-1 GPS constellation A GNSS consists of three segments. There are the satellites that transmit the position information (the space segment), there are the ground stations that are used to control the satellites and update the information (the control segment), and finally there is the receiver (the user segment). It is the receiver that collects data from the satellites and computes its location anywhere in the world based on information it gets from the satellites. The GNSS receiver uses an elaboration of a technique that is tried and true and used by navigators and surveyors for centuries. Basically it uses a known set of locations to compute its current location by taking fixes on the known sites. It measures its distance from the satellites and uses this information to compute a fix. To measure the distance, it measures the length of time the signal takes to arrive. The signal moves at the speed of light so it can compute the distance based on the travel time. However, these transmitter sites are moving. The solution to this problem is to have the satellite itself send enough information to calculate its current location relative to the receiver. Now, armed with the satellite location and the distance from the satellite we can expect that we are somewhere on a sphere that is described by the radius (distance) and centered at the satellite location. By acquiring the same information from a second satellite a second sphere that intersects the first in a circle can be computed. If the same information from a third satellite is acquired the new sphere would intersect the existing circle at only two 10

30 Chapter 2 Background points. If the approximate position of the receiver is known, then one of those points can be cancelled and we are left with the exact fix location in 3D space (Figure 2-2). Figure 2-2 Principle of satellite navigation In practice the information from a fourth satellite is acquired and this eliminates the incorrect point. XYZT Figure 2-3 Four satellites to reveal the time error The satellite sends the current time along with the message so the receiver can subtract its knowledge of the current time from the satellite time in the message (which is the time that the signal started its descent) and use this to compute the difference. For this to work the time in the receiver must be accurate to ns precision. 11

31 Chapter 2 Background The satellite itself has an atomic clock to keep the time very precisely, but the receiver unit is neither big enough nor expensive enough to have an atomic clock built in, so its clock is likely to be in error. For this reason the above assumptions about the distance calculation are likely to have considerable error. A fourth satellite is used to solve for this local clock error (see Figure 2-3). We will then have a good position fix and as a side effect we will also have the correct time to about 100 nanoseconds or so. One of the applications of GNSS technology is to provide the correct time even when position is not an issue (timing receiver). In addition to the data already mentioned, the receiver uses almanac data to compute the approximate positions of all the satellites. It is the ephemeris data downloaded directly from the satellite that is used to compute its exact position in the sky. Similar to the geometry problem in traditional survey systems of taking bearings on fixed sites, the satellite geometry has a significant effect on the accuracy of our final position. A unit-less number representing this geometry is called Dilution of Precision (DOP). It is the ratio of the eventual position error to the error in the pseudorange measurements and is used by the receiver in determining which of the satellites available are the best to use. The smaller the number, the better the geometry Control and Space Segment In addition to the receiver, for the satellite navigation system to operate, a set of satellites in the sky and a method of updating the data from each satellite are also needed. There are full time land based sites that monitor the various satellites, which are often referred to as Space Vehicles, SVs. These land based sites check the health of the SVs, measure their orbits, calculate the clock accuracy, and send models of these as needed. The land based sites are located at well-distributed positions so that they can verify the operation of the satellites. The satellites are traveling 11,000 nautical miles above the earth in carefully modeled orbits twice a day. Each orbit takes 11 hours and 58 minutes, so in comparison to the sun they will seem to drift 4 minutes a day (they are in fact fixed with respect to the fixed stars). The complete constellation consists of a minimum of 21 SV's and 3 working spares. Currently there are 30 total satellites in the sky. There are 6 orbits with multiple satellites in each orbit as depicted in Figure 2. Each orbit is inclined at 55 degrees and thus there are no orbits that go directly over the poles, but 12

32 Chapter 2 Background certainly several satellites can be seen from the poles or anywhere else on the earth. The goal of the system is always to provide at least 4 satellites visible somewhere in the sky. In practice there are usually many more than this, sometimes as many as 12. Each satellite maintains an essentially circular orbit around the earth. It contains a receiver to accept update information, a transmitter to send information to the GPS receiver, an antenna array to provide good gain in the direction of the earth, several atomic clocks to keep accurate time, control hardware, and photoelectric cells to power all the other subsystems Receiver Block Diagram In Figure 2-4 the block diagram of a GPS receiver is shown. The signals transmitted from the GPS satellites are received from the antenna. Through the radio frequency (RF) chain the input signal is amplified and the frequency is converted to a desired output frequency. This process uses filtering and limiting, a Low Noise Amplifier (LNA), a frequency synthesizer and down-converter. An analog-to-digital converter (ADC) is used to digitize the output signal. Where multi-bit quantization is used, a pre-correlation Automatic Gain Control (AGC) (or ADC threshold control which has a similar effect) is required. It is used to increase dynamic range, control the quantization level and suppress pulse interference (Parkinson, 1996a). In Chang (1982), the effects of sampling and quantization performance on digital filters is analyzed and it is shown that the minimum quantization degradation happens for a particular ratio of maximum quantization threshold to RMS noise level. The antenna, RF chain, and ADC are the hardware used in the analogue part of the receiver. After the signal is digitized, either hardware or software can be used to process it. In this base-band multi-channel signal processing stage, signal acquisition and tracking is performed. Acquisition consists of finding the signal of a certain satellite. The tracking process is used to maintain lock to both the spreading code and the carrier, with the output being the phase transitions of the navigation data. In a conventional receiver, acquisition and tracking are both performed in hardware. In such receivers, in several channels of which the operations are similar, carrier phase and code phase are derived using Phase Lock Loop (PLL) and a Delay Lock Loop (DLL) which operate inter-connectedly. The pseudoranges and navigation data bit-stream are the outputs of this section. From the navigation data phase transition the sub-frames and navigation data can be obtained. The ephemeris data are used to obtain the satellite 13

33 Chapter 2 Background positions. Finally, the user position can be calculated from the satellite positions and the pseudoranges. Figure 2-4 GNSS receiver block diagram (Maxim-IC, 2005) GNSS Acquisition and Tracking To demodulate the received signal in order to extract the navigation data, the receiver needs first to determine which satellites are visible, then determine the carrier frequency and code phase of the signal and finally remove the carrier and the code from the signal. This process is done in two stages: acquisition and tracking. During acquisition, carrier frequency and code phase are determined coarsely using a search technique. The coarse information is then used to initiate a coupled phase-delay lock loop (PLL-DLL) to track precisely the code and carrier phases and strip them from the signal to extract the navigation data and calculate the range. In an ordinary receiver, acquisition is usually performed in an application specific integrated circuit (ASIC). In a software receiver, it is implemented in software. Three main search methods for the two parameters are (Borre, 2007): 1. Serial search acquisition in which all the different combinations of possible carrier frequency and the code phase are tried serially to find the correct values. 14

34 Chapter 2 Background 2. Parallel frequency space search acquisition in which using an FFT algorithm, carrier frequency is found and then code phase. 3. Parallel code phase search acquisition in which a circular cross correlation between the input signal and the locally generated PRN without any phase shift is used. In Fantino (2005), the acquisition performance with new modulation schemes is analysed. In Borio (2006) the impact of the acquisition search strategy on the detection and false alarm probabilities in a CDMA receiver is studied. The next stage after acquisition is tracking. Tracking loops give the carrier phase and pseudorange measurements. Also the precise values of the code and carrier phases are removed from the signal to obtain the navigation data. This process is done by using a phase lock loop (PLL) and a delay lock loop which work coupled together. This is shown in Figure 2-5 (Tsui, 2000). Figure 2-5 Code and carrier tracking loop in a GNSS receiver In the DLL, the carrier frequency from the PLL is used to strip the carrier. The result is correlated with a locally generated code in three different phase shifts (early, prompt and late) to calculate the autocorrelation. Correlation here involves integrating (summing using and integrate and dump filter) the output of the product of the signal code and the locally generated code, the delay of which is controlled by the loop. A discriminator compares (usually) the early and late correlations and this 15

35 Chapter 2 Background output drives the oscillator so the prompt correlation is maximized, and its output provides the exact code phase delay. The BOC modulation uses different discriminators to avoid the false peaks in the ACF, such as BOC-PRN (Dovis, 2005) and BOC-PRN E plus L (Wu, 2007). In the PLL, the carrier phase is tracked. In this loop the code from the DLL is assumed to be removed from the signal. For strong signals a normal PLL (see Figure 2-6) may be used, but more typically, to avoid the carrier phase transitions, which happen because of the navigation data, a Costas loop is used. Similar to the DLL there is a choice discriminator to provide the error signal. There are number of different techniques to detect the lock (Parkinson, 1996a). Carrier to noise density ratio is among these methods which can also be a good indicator of the received signal quality (Parkinson, 1996a). In Chapter 5 a complete analysis of the characterization of the effect of CW RFI on this indicator of signal quality is presented. Figure 2-6 Phase Lock Loop (Viterbi, 1966) A PLL exhibits various tracking phenomena when subject to an input consisting of the sum of two sinusoids (signal and interference). For a first order PLL Tillman (2003) and Oie (1994) show that the behavior of the loop depends on the frequency difference between the signal and the interference, the ratio between the amplitude of the signal and the interference and the loop gain. According to Tillman (2003), there is a bifurcation point in which the loop stops tracking the signal and might start tracking the interference. In Stensby (2004) it is shown that the interference will always cause a periodic phase error, the period of which is related to the frequency difference between the signal and the interference. This result is also proved for a second order PLL in Karsi (2000) where the analysis is done in the presence of additive Gaussian noise. Another result which is important to be reiterated in Karsi (2000) is that the vulnerability of the loop to the presence of interference 16

36 Chapter 2 Background increases if the interferer and the desired signal spectral locations are on opposite sides of the loop s VCO (Voltage Control Oscillator) quiescent frequency. If we consider these analyses of the behaviour of PLL in the presence of CW RFI as microscopic analysis, then the characterization of the effect of CW RFI that we have presented in Chapter 5 can be considered as macroscopic analysis. In that approach the loop is considered as a block whose quality of operation is determined by the C/No of the output signal. However the parameter used for measuring the range in GNSS receivers is usually the code phase, which ultimately is measured in the DLL. Sometimes ranging also uses the carrier phase. In the next section the relationship between the received signal quality and the code and carrier phase jitter is reviewed GNSS Signal Structure and Characterization Choosing better signal structure and characterization is one of the nonreceiver-based techniques to make GNSS operation more robust in the presence of interference. It will be explained in the following sections of this chapter that the effect of interference depends not only on the characteristics of the interference source but also on the signal spectrum. The objective of this section is to investigate and characterize the current GNSS, i.e. GPS L1, signal structure. Frequency assignment and modulation format and the method of multiple access for different users are important in the context of interference. Autocorrelation and cross-correlation of the signals can affect the code tracking errors and multipath effects. GPS uses Binary Phase Shift Keying (BPSK), where the RF carrier is either transmitted as is or with a 180 degree phase shift depending on whether digital 0 or 1 is conveyed. Direct sequence spread spectrum (DSSS) extends further the use of BPSK by GPS (Kaplan, 2005). DSSS uses a pseudo-random noise code (PRN), which has a much higher symbol (known as a chip ) rate than the navigation data. The PRN is unique to each satellite and is used for multiple access. It also provides more precision in ranging and some immunity to narrowband interference and multipath. This method of multiplexing in contrast to frequency (FDMA) or time (TDMA) division multiple access, uses the PRN codes for multiple use of the same channel, hence the name CDMA (code division multiple access). Immunity against interference is highly dependent on the processing gain of the system. The higher the processing gain, the wider the GPS signal is spread. By spreading the signal over a wide bandwidth, only a fraction of the desired signal is near to the narrow-band 17

37 Chapter 2 Background interference and can be corrupted. After the despreading process, the narrow-band interference is spread, by the same processing gain factor. Processing gain is defined as (Plausinaitis, 2006): Processing gain = 10 log (chip rate/data rate) = 43dB where chip rate is 1.023Mcps and data rate is 50bps in the case of the GPS CA code L1 signal. The quality of the despread GPS signal partially determines the precision of the GPS receiver and can be quantified by the resulting bit-error rate (BER). The GPS C/A code is a Gold code with a relatively short 1-ms period (i.e., the PRN sequence repeats every 1 ms). Therefore, the C/A code (neglecting the navigation data) has a line spectrum with lines 1 khz apart (Kaplan, 2005). In Figure 2-7, the spectrum of the navigation data is incorporated into the code spectrum. Then noise and interference are added to the signal to achieve the final GPS signal received at the antenna. In Figure 2-8 this additive effect is shown in the frequency domain. Figure 2-7 Spreading the data over the C/A code spectrum bandwidth Figure 2-8 Interference and background noise is added to make the final GPS signal received at the antenna In this initial analysis, interference is assumed to be CW Constant Amplitude (CA) [15] at a frequency f i away from the band center of the code spectrum and away from the nearest Dirac line in the spectrum. This line is also assumed to be the j th line of the spectrum. In other words this line is j khz away from the band center as all the lines are 1 khz apart from each other. Now; in the base-band processing of the GPS receiver, this signal is passed through the correlator to be acquired and tracked. Δ f i 18

38 Chapter 2 Background In the correlator, first the code is despread (Figure 2-9) and then the navigation data is filtered out (Figure 2-10). The despreading function allows the information to be extracted from the noise. Nevertheless, this function may be highly disturbed by the presence of interference. Figure 2-9 Code is despread by being multiplied by the receiver code replica and interference is spread J_a Figure 2-10 Tracking loop low pass filter: extracts the data and the interference which is outside the filter bandwidth The code is despread as it is multiplied by the replica code in the receiver. Similarly interference is spread over the frequency bandwidth of the original signal in the same way that the data is spread in the original signal (Figure 2-9). A low pass filter is used in the tracking loops (the integrate and dump block). Only interference that is within the bandwidth of this filter remains (Figure 2-10). In Figure 2-8, the amplitude of the interference is shown to be J b. In passing the RFI through the correlator and the filter, the value of the amplitude of the remaining interfering signal is J a. The aim of these figures is to show that the value J a is determined by the strength of the nearest line to the interference (in this example j th line). Now if we have an RFI with fixed frequency and a GPS signal with Doppler frequency that varies with satellite motion, over time the RFI coincides with several different consecutive lines in the spectrum. Each of these lines has its own unique effect on the remaining interference in the output of the loops. We will examine the effect of a series of lines, and thus calculate the frequency of the RFI. The quantity that can best reflect this 19

39 Chapter 2 Background effect is the correlator output power. Carrier to noise ratio, which quantifies the quality of the signal is the parameter that is used for this purpose. GIOVE-A is the first and only prototype Galileo satellite launched. It started to transmit satellite signals modulated by the navigation message, the E1 carrier (which is the same as L1) and the Binary Offset Carrier BOC(1,1) spreading code on 12 January BOC(1,1) can be regarded as the modulation product of a PRN and square wave sub-carrier, which has the same frequency as the chipping rate of the PRN (shown in Figure 2-11). Figure 2-11 Galileo BOC(1,1) modulation scheme The signal, GIOVE-A BOC(1,1) at E1 band has two separate channels: a data channel modulated with a 1.023Mchip/sec PRN, the B-code, with 4ms period and 250chip/sec navigation data, and a pilot channel modulated with a 1.023Mchip/sec Primary PRN code, the C-code, with 8ms period and a 125chip/sec secondary PRN code, the S-code with 200ms period (Borre, 2007). The signal generation scheme is shown in Figure As the navigation data period is the same as one full length of 4ms of B-code, the maximum non-coherent integration time for the data channel will be 4ms. A similar situation exists on the pilot channel where the S-code has a code period of that of one full C-code length. Therefore, the S-code in the pilot channel can be thought of in a similar way to the navigation data in the data channel, with the exception that it is predictable. 20

40 Chapter 2 Background Figure 2-12 GIOVE-A Signal generation for E1 band BOC (1,1) In Chapter 7 the characterization of the effect of CW interference on a Galileo signal is investigated. The use of this signal is free of charge and is open service (OS). It is transmitted in L1 frequency so it is called L1 OS signal. The modulation and multiplexing techniques used in Galileo are more complex than those of GPS. Modernized GPS signals and Galileo signals have improved performance over the current GPS signal. Binary offset carrier (Betz, 2002) improves the signal acquisition and tracking schemes. The BPSK spectrum is split into two symmetrical components with no power at the carrier frequency (Martin, 2003). The autocorrelation function (ACF) is narrower, which improves the tracking accuracy. However, it also has two side peaks, which have to be avoided while tracking. The time domain signal is shown in Figure The vulnerability to narrowband interference and probability of satellite availability is different from that for GPS. This is also investigated in Chapter Interference Threat Because GNSS receivers rely on external radiofrequency signals, they are vulnerable to Radio Frequency Interference (RFI). In this section, after some basic definitions, the nature of the sources of RFI and other features of vulnerability of the GNSS receivers to RFI are analyzed Definitions Noise. An unwanted signal, which can be broken out into three basic categories: background, modulated and interference noise. 21

41 Chapter 2 Background Background or Thermal Noise. Background noise is the inherent noise of the system. Thermal noise is generated primarily in the first amplifier of a well-designed receiver. Modulated Noise. Modulated noise is an undesirable signal that enters a system and rides on a signal using the system power producing undesirable side effects in signal quality. Interference Noise. Interference noise is noise that comes in at the same frequency (ies) as the wanted signal and can mask (or overwhelm) parts of the desired signal. Interference Source. An interference source is an electrical apparatus, which emits electromagnetic interference. Unintentional Interferences. Interference received from a source not intended to interfere. Any RF transmitter is a potential source of GPS interference, if it has sufficient power. Intentional Interferences. Various formats of interference that are tactically deployed against GPS, which includes jamming, spoofing and meaconing. Jamming. Jamming is the deliberate act of transmitting in the GPS band. Spoofing. Involves the transmission of legitimate-appearing false signals in an attempt to confuse or deceive a GPS receiver and can be categorized into two groups: 1) Denial Spoofing. A GPS similar waveform is transmitted in an attempt to prevent a GPS receiver from tracking real GPS satellite signals. 2) Deceptive Spoofing. Involves transmitting similar waveforms, but it attempts to deceive the GPS receiver into believing the spoof signals are the actual GPS satellite signals Meaconing. Meaconing is a system of receiving radio beacon signals and rebroadcasting them on the same frequency to confuse navigation. The meaconing stations cause inaccurate bearings to be obtained by aircraft or ground stations Types of Interference 22

42 Chapter 2 Background Table 2-1 briefly summarizes various types of RF interference and their corresponding sources. Table 2-1. Various types of RF interference Type Typical Sources Wideband Gaussian Intentional noise jammers Wideband phase/frequency modulation Television transmitters harmonics or near-band microwave link transmitters overcoming front end filter of GPS receiver Wideband spread-spectrum Intentional spread spectrum jammers or near-field of pseudolites Wideband-pulse Radar transmitters Narrowband phase/frequency modulation AM stations transmitters harmonics or CB transmitters harmonics Narrowband swept continues wave Intentional CW jammers or FM stations transmitters harmonics Narrowband continues wave Intentional CW jammers or near band unmodulated transmitters carriers In terms of bandwidth, interference can be categorized into three main groups. In the following, the terms used in Table 2-1 are explained in detail. Narrow Band Interference. Interference sources with discrete frequency spectra (e.g. high-frequency generators and microprocessor systems) emit narrowband interferences. Wide Band Interference. Switchgear and electric motors in household appliances, however, spread their interference energy over broad frequency bands and are considered to belong to the group of interference sources having wideband frequency spectrum. Ultra Wide Band Interference. One definition of UWB signals describes the spectral emissions as having an instantaneous bandwidth of at least 25% of the center frequency. Other names for UWB, or terms associated with it, include: impulse radio, impulse radar, carrier-less emission, time-domain processed signal, and others. Also, in terms of spectral separation or overlap between the interference and the useful signal it can be categorized into three different groups. In Band Interference. In Band Interference, for a receiver, means those emissions from a transmitter that are permitted under the conditions of the license under which the receiver operates. 23

43 Chapter 2 Background Out of Band Interference. If the receiver is operated in the space of a spectrum license, affected by emissions at frequencies outside the frequency band of the spectrum license. Intermodulation Interference. Intermodulation products are unwanted frequency components resulting from the interaction of two or more spectral components passing through a device with nonlinear behavior such as a mixer, an amplifier, output stage of a transmitter or input stage of a receiver (Landry 1998). The unwanted components are related to the fundamental components by sums and differences of the fundamentals and various harmonics. For two signals, example of the intermodulation products includes: f ± f ;2 f ± f ;2 f ± f ;3f ± 2f etc. Intermodulation products can occur in both transmitter and receiver Potential Sources of Interference (Landry, 1997) 1. Interference due to VHFCOM harmonics. The ATC (Air Traffic Control) communication mode uses the [ MHz] emission band, which corresponds to the VHFCOM band. It contains 760 channels spaced at 25 KHz. The VHFCOM A/G(Air to Ground) emitted power is 14dBW and 17dBW for the G/A (Ground to Air) mode. Their harmonics are considered as CWI which contain the maximum energy of the VHFCOM signal. 2. Interference due to ACARS harmonics. The channels from the ACARS (Aircraft Communication Addressing and Reporting System) generate interference into the GPS and very near the GLONASS bands, which may penetrate the in-band filter and hence the GNSS receiver filter may not provide any rejection to this signal. 3. Interference due to MODE-S IMP. The Mode-S (Mode Select Beacon System) is a radio navigation system using two fixed frequencies. The interrogator pulsed signal is at 1030MHz and the reply signal is at 1090MHz. The maximum transmitted power is 52.5 dbw for the interrogator and 27dBW for the reply signal from the aircraft. The Mode-S is considered as a potential pulsed interferer 4. Interference due to MODE-S side lobe power. The Mode-S standard A specifies power limits at various frequency offsets from the carrier. Within 24

44 Chapter 2 Background the GPS band, the standard requires a minimum power of 60dB down from the carrier for the interrogator and the reply signals. 5. Interference due to the SATCOM emitters. The SATCOM Communications use the frequency band [ MHz]. The channel bandwidth is 20 KHz and they are frequency spaced at 0.75MHz. The mean EIRP is 18dBW and the minimal rejection is 100dB in the L1 band. The SATCOM emitters generate many intermodulation products which can fall inside the GPS band. 6. Interference due to TV harmonics. There are 6 TV channels generating harmonics of an order smaller than 10 which cause interference problems to GPS receivers. In Chapter 3 of this thesis, it is shown how the third harmonic of a TV broadcaster in Sydney, Australia interferes with GPS L1 signal. 7. Interference due to FM harmonics. Many small frequency bands inside the FM band [ MHz] have their harmonics in the GPS band. 8. Amateur radio harmonic interferences. The American amateur radio band [ MHz] has 7 harmonics of 7 orders directly inside the GPS band and many others are near this. The emitted power may reach 500W in the United States. The Amateur Radio emitters may not reject their harmonics sufficiently for the GPS applications Characteristics of Interference Spatial-Directional Characteristics of RFI Transmitters may be either mobile or stationary and they may be on the Earth s surface, airborne, in orbit or in deep space. Some examples are shown in Table 2-2. The proximity of a surface RFI source to the receiver will determine the level of its effect on the observation; the earth s curvature or local geography may obscure a transmitter from the receiver s view. Observations will be affected over a larger radius by airborne and orbital transmitters because line-of-sight signal reception is possible for greater distances. For example a telescope operating in the 2 30 GHz band that has an aperture greater than 100 wavelengths will have a gain equal to that of an isotropic radiator at 19 from the axis of the main beam. For such a telescope satellite signals, of which there are many, are difficult RFI sources to avoid. 25

45 Chapter 2 Background Table 2-2 Mobile and stationary transmitters Surface GSM, taxi services TV, radio broadcasts Airborne RADAR, aero plane communications In orbit GPS, satellite-based geo-synchronous RADAR satellites Deep space deep-space probes Temporal-Spectral Characteristics of RFI A signal may be either continuous or intermittent. An example of a continuous signal is a public broadcast signal such as TV or radio. Many continuous communications signals have some degree of periodicity that is associated with either the carrier or the information being transmitted. The signal s amplitude may be constant, as in FM modulation, or varying as in AM modulation. Examples of intermittent signals are radar, where the signal will be received only when the transmitter is pointing in the direction of the telescope, and analog mobile telephones that only transmit when the user is speaking. Intermittent voice communication has the same signal characteristics, during transmission, as it would if it were continuous. On the other hand, radar signals and interferences due to current switching may be impulsive. 2.4 GNSS Receiver and Interference Interference and GNSS RF front end The first stage of a radio receiver is usually a low noise amplifier (LNA). Typically, the designer will put filters at both sides of the GPS LNA. A filter in front of the LNA helps to reject out-of-band signals and prevent the LNA from being saturated. This filter should have a very low insertion loss. Putting a high insertion loss filter before the LNA increases the system's noise figure. A receiver's dynamic range is the input signal power range over which the receiver is capable of detecting and making a meaningful estimate of the original transmitted signal degraded by noise or other detrimental channel impairments (Chen, 1997). When the received signal is very weak, the system noise figure dominates receiver signal-to-noise ratio (S/N) performance. When the received signal is very strong the intermodulation noise due to the nonlinearity of the front-end dominates. 26

46 Chapter 2 Background In the case of GNSS receivers, where the power of the signal is lower than the power of the thermal noise, the highest gain mode of the receiver is when the automatic gain control (AGC) operates at the receiver s thermal noise level (Kaplan, 2005). This means that the input of the ADC is always kept constant and this is essentially the gain objective in the front end design in the GNSS receivers. When an interfering signal is received, it may well exceed the noise level, and the AGC will attenuate the received GPS signals. This is discussed in more detail in Chapter 3 when detection of a real interference is presented System Vulnerability to Interference GPS has some inherent features that increase its resistance to interference. The Direct Sequence-Spread Spectrum (DSSS) modulation process was chosen for GPS partly because of its resistance to jamming. To ensure that satellite signals do not interfere with terrestrial line-of-sight microwave systems the ITU has set a flux density regulation on the power that can be 2 transmitted. The respective flux density limit has been set to -154 dbmw / m (Parkinson, 1996a). With this low signal power level, receivers will still be clearly susceptible until the modernization program is completed. Modernization involves the development of additional civilian GPS signals in separate bands with greater power that are better equipped to overcome the effects of unintentional interference. From the vulnerability assessment of the US national transportation infrastructure that relies on GPS, there is a recommendation to ensure the integrity and availability of civil SPS signals (Volpe report, 2001). To protect integrity, we must be able to quickly detect the presence of interference and to protect availability, we must be able to locate and deactivate an interference source GPS Receiver Interference Threshold The susceptibility thresholds for GPS systems are well defined and consistent between various standards and regulations. The ITU Radiocommunication Sector (ITU-R) is one of the three sectors (divisions or units) of the International telecommunication Union (ITU) and is responsible for radio communication. Its role is to manage the international radiofrequency spectrum and satellite orbit resources and to develop standards for radio- 27

47 Chapter 2 Background communications systems with the objective of ensuring the effective use of the spectrum. A representative set of data, taken from ITU-R M.1477, is summarized below. On June 2, 1994 the termination of the Microwave Landing System (MLS) was announced in the United States for Category II and III landings. Research focus was on satellite based landing systems. Standalone GPS does not satisfy the stringent aircraft approach and landing requirements. Four fundamental parameters provide the bases for allocation of specific requirements for navigation systems. These metrics are each affected by signal interference. Accuracy. Accuracy is the measure of the navigation output deviation from truth under fault-free conditions often specified in terms of 95% performance. Integrity. Integrity is the ability of a system to provide timely warnings to users when the system should not be used for navigation. Integrity risk is the probability of an undetected navigation system error or failure that results in hazardously misleading information (HMI) onboard the aircraft. Continuity. Continuity is the likelihood that the navigation signal-in-space supports accuracy and integrity requirements for the duration of the intended operation. Continuity risk is the probability of a detected but unscheduled navigation function interruption after an approach has been initiated. Availability. Availability is the fraction of time the navigation function is usable (as determined by its compliance with the accuracy, integrity, and continuity requirements) before the approach is initiated. The task of detecting any anomaly in a GPS signal requires a statistical decision process to draw conclusions about the existence of the anomaly. This requires that any signal parameters need to be associated with detection threshold (DT), which is driven by the probability of false alarm (P fa ), the probability of missed detection (P md ) and the minimum detectable bias (MDB) as specified by the requirements in FAA. There are three types of GPS air navigation systems in which receivers are relatively well developed. 1. Satellite Based Augmentation System (SBAS). This system is designed for category I (FAA 1984) precision approach. Wide Area Augmentation 28

48 Chapter 2 Background System (GPS/WAAS) and European Geostationary Navigation Overlay Service (EGNOS) are examples of this system. 2. Ground Based Augmentation System (GBAS). This system uses ground based pseudolite emitting signal having similar characteristics of GPS. An example is GPS/LAAS (FAA 1995) designed for category II/III (FAA 1984) precision approach. 3. Semi-Codeless Receiver. Receivers of this type are typically ground based and are more sensitive to interference. Receiver susceptibility thresholds defined in ITU-R M.1477 are summarized in Table 2-3 (see Nguyen 2002). ITU-R M.1477 also provides additional data concerning the behavior of the susceptibility threshold as a function of interference signal bandwidth. The narrow band acquisition mode data shows that CW, for signals with bandwidth up to 700 Hz, is the most severe threat with the lowest thresholds of dbm. The same threshold holds for both SBAS and GBAS air navigation receivers. The susceptibility threshold is monotonically higher for larger interference bandwidth. The interference threshold versus bandwidth for SBAS and GBAS air navigation receivers in track mode is similar to the data presented in Table 2-3 (Nguyen 2002). The same trend is also used for receivers in acquisition mode, with the interference threshold 6dB lower. In Table 2-3 a narrow band signal is defined as having bandwidth less than or equal to 700Hz, and a wideband signal as having interference bandwidth in the range 100kHz to 1MHz. Narrow-band Track mode Narrow-band Acquisition mode Wide-band Track mode Wide-band Acquisition mode Table 2-3 GPS Susceptibility Threshold SBAS Receiver GBAS Receiver Semi- Codeless Receiver dBm dBm dBm dBm dBm dBm dBm /MHz dBm /MHz dBm /MHz dBm /MHz dBm /MHz dBm /MHz 29

49 Chapter 2 Background Table 2-4 Interference threshold versus interference bandwidth for GPS receivers and for SBAS and GBAS air navigation receivers in track mode. Bandwidth Receiver Interference Threshold 0 BW dbm 700 BW 10kHz Linearly increasing from -120 dbm to dbm 10kHz BW 100kHz Linearly increasing from dbm to dbm 100kHz BW 1 MHz dbm 1 MHz BW 20 MHz Linearly increasing from dbm to dbm 20MHz BW 30MHz Linearly increasing from dbm to dbm 30MHz BW 40MHz Linearly increasing from dbm to dbm 40MHz BW dbm The reasons why the threshold increases with respect to the bandwidth of the interference will be explained in detail in the next chapters. The output of this research is to be used in a Continuously Operating Reference Station (CORS) network. This type of network is explained in the following section. The type of GPS receiver used in this network is considered equivalent to the semi-codeless receiver (Table 2-3) used for air navigation GPS. 2.5 CORS Networks Even though satellite systems have revolutionized the art of positioning and navigation, there are still several setbacks to the accuracies that can be attained due to intentional and unintentional sources of error which the system and satellite signals are susceptible to. These include anti-spoofing (AS), receiver clock errors, satellite clock errors, satellite orbit errors, satellite signal interference, receiver noise, ionospheric error, tropospheric error and multipath. The accuracy of the unassisted GPS single point positioning signal is adequate for most applications such as recreation, automobile navigation and fleet tracking, but many other applications require greater accuracy. The use of GPS for geodetic survey applications has resulted in a critical need for development of acceptable accuracy standards and GPS survey 30

50 Chapter 2 Background specifications for control surveys performed by relative positioning techniques. Satellite positioning with accuracy better than 5m requires the use of a reference station. As a Trimble editorial once put it, surveyors have been using GPS for precise surveys for years fixing points to an accuracy of millimeters. This requires the use of multiple receivers-so involved that only trained geodesists can do it (Ogonda, 2003). Precise positioning using the Standard Positioning Service (SPS) requires the use of GPS carrier phase observables and relative positioning. Carrier phase observables are susceptible to ambiguity, carrier phase multipath and interference. For dual frequency receivers, ambiguity resolution is more easily resolved but the quality of the carrier phase multipath calibration depends on the ability to separate the multipath effects from other errors, especially phase centre variation, ionospheric effects and interference. Reference stations provide error correction information, which is useful to the Rover receivers in removing the remaining tropospheric, ionospheric and orbit biases. They are mainly used in GPS positioning for the following three purposes (Ogonda, 2003); Harmonization of the different geodetic reference systems around the world and more specifically, transforming the WGS84 reference system to user defined ellipsoid/datum, such as Clarke 1866 or the Geodetic Reference System of 1980 (GRS80) on which ETRS89 is based. Detection of malfunctioning and failure of other reference stations Attenuation of the satellite, receiver and signal propagation biases. Methods have been, and are still being, developed to reduce the effects of errors and enhance the accuracies attainable with GPS. This concept is commonly known as Differential GPS (DGPS) positioning. The differential mode eliminates most of the errors except multipath and other receiver errors, which are local and usually depend on the environment surrounding the station (receiver). The main advantage of satellite positioning since its inception in the early 1980s is that it permits the determination of the position of one receiver relative to another reference receiver without the requirement of station inter-visibility unlike the earlier conventional techniques such as triangulation, trilateration and traversing using theodolites and electromagnetic distance measuring equipment. There are two types 31

51 Chapter 2 Background of differential mode GPS: real time and post-processed DGPS. Selective availability was turned off by the U.S.A government in May 2000, which reduced the advantage of DGPS, which could eliminate Selective Availability (SA), over single point GPS. Accuracies of 5 to 7 meters in the horizontal and 8 to 9 meters in the vertical at 95% confidence have been guaranteed. The process of reference station establishment takes into consideration all the possible sources of such errors and tries to eliminate or reduce them. For the required accuracies to be achieved in positioning, two main factors have to be taken into consideration. The position of the reference station should be determined at the highest accuracy possible by eliminating or reducing all the errors affecting the station position estimate. The rover and reference stations should be relatively closely spaced, since in relative/differential positioning, the same conditions are assumed to exist at the receivers at either ends of the baseline. Three different options for corrections are as follows. Can be performed in real-time or in post processing May be obtained from one or more reference stations Can be from a permanent service or service specific to the project Continuously operating reference stations (CORS) have been deployed to support very high accuracy geodetic applications since the 1980s (Evans et al., 2002). Geodetic techniques are by their very nature multi-station, taking advantage of the geometric strength, reference datum stability (and redundancy) afforded by networkbased positioning. Such CORS networks have been deployed globally, as well as in geodynamic hot spots like Japan and Southern California where there is significant tectonic motion (Rizos, 2004). As GPS base station data are crucial requirements for high-accuracy systems, providers of Continuously Operating Reference System (CORS) data must be able to assure the quality of their data. This thesis is part of the project which will develop techniques and algorithms that minimize the impact of multipath and interference on CORS data, and identify local base station deformation. This project is part of a bigger project (enhancing Australia s core geodetic infrastructure) which is defined by 32

52 Chapter 2 Background the Cooperative Research Center for Spatial Information (CRC-SI), Australia. The aims of the project are as follows (Stewart, 2004). Develop devices and software to detect and evaluate GPS signal interference and identify interference sources. Construct algorithms and models to identify and mitigate CORS sitedependent GPS multipath errors and hence allow site assessment before hardware installation. Develop standards and protocols for a seamless integration of regional and local CORS GPS networks. Create improved geodetic models to enhance the definition of the national reference frame. In this thesis we concentrate on the first aim. 2.6 Relationship Between Received Signal Quality and Positioning Quality In the previous chapter, it was shown that RFI affects many components and observables in the GNSS receiver including pseudorange and position. Many receivers report received signal quality as carrier power-to-noise density ratio, denoted C/No (Parkinson, 1996a). The objective of this section is to review the works that show that there is a close relationship between this quantity and the code phase measurement error variance and thus ultimately with the positioning quality of the receiver. This objective is the motivation for the extensive work which has been done in Chapter 5 in characterizing the effect of CW RFI on the C/No. Testing reported in Ross (2001) found that different receivers report different C/No when presented with same non-white interference. Those differences could have been the result of different front end bandwidth, discriminator design or other receiver characteristics. Betz (2001) by theoretical prediction of the response of different interference spectra started an effort to explain those curious differences in receiver reports of C/No. In Kaplan (2005), the effects of RFI on carrier and code tracking were investigated separately. In these expressions the Doppler frequency of the signal is essentially ignored. The reason is that the RFI bandwidth is assumed to be wide enough so that the spectral separation coefficient and the receiver jamming resistance coefficient are both Doppler frequency invariant. 33

53 Chapter 2 Background Evaluating the effect of RF interference on correlator output signal to noiseplus-interference ratio (SNIR) provides the basis for assessing the effect of this interference on acquisition, tracking and data modulation. When the interference can be modeled as statistically stationary, and when the spectra of either the interference or the desired signal (or both) are well approximated by a constant over a bandwidth that is the reciprocal of the integration time used in the correlation, the correlator output SNIR is as follows (Betz, 2000): jθ j2πfτ 2 2TCs / N0[ R{ e Ss ( f ) HT ( f ) H R ( f ) e df }] ρ c ( τ, θ ) = (2-1) 2 2 H R ( f ) Ss ( f ) df + Ci / N0 H R ( f ) Si ( f ) Ss ( f ) df where τ is the delay of the locally generated replica code relative to the true TOA of the received signal in space; θ is the carrier phase of the replica carrier signal relative to the received power of the desired signal phase of the received signal; T is the integration time of the correlator; Cs is the received power of the desired signal (in watts); N0 is the power spectral density of the white noise (in W/Hz); R{.} denotes the real part of the enclosed function; S s ( f ) is the power spectral density of the signal, normalized to unit area over infinite bandwidth; H T ( f ) is the transfer function of the SV signal transmitter; H R ( f ) is the transfer function of the received filter; Ci is the power of the received interference signal (in watts); and S i ( f ) is the power spectral density of the aggregate interference, normalized to unit area over infinite bandwidth. This equation shows that any nonwhite interference must be accounted for as well as white noise in calculating the carrier to noise density ratio which is considered to be an indication of the received signal quality. A new parameter is defined to be the white noise equivalent of the nonwhite interference in terms of its effect on the signal quality. Based on this definition, the expression for the effective carrier power to noise density ratio can be shown in (2-2). 34

54 Chapter 2 Background ( C / N ) s 0 eff 1 = (2-2) 1 Ci / Cs + ( C / N ) QR s 0 c Where C s / N0 is the unjammed carrier to noise power ratio of the received signal inside the receiver and the dimensionless Q is defined in Kaplan 1996 to be the receiver jamming resistance quality factor and code generator in chips per second. Q can be calculated from (2-3). R c is the spreading code rate of the H Q = R ( f ) R κ c 2 S ( f ) df is s (2-3) where kis is called the Spectral Separation Coefficient (SSC) (Betz, 2002), which is defined as 2 is H R ( f ) Si ( f ) Ss ( f ) κ = df (2-4) which has units of seconds and depends on both the spectrum of the desired signal and the spectrum of the interference. 2.7 User-to-Satellite Geometry The accuracy of positioning, using the measured pseudoranges from the receiver to each of the satellites depends on different factors. The final stage in the position evaluation in the GPS receiver is the estimation of four quantities (x, y, z and time offset in the local clock) using four or more measured distances from the receiver to the satellites with known positions. So we have the following nonlinear estimation problem (Parkinson, 1996a). ρ = r r + c. b + ε (2-5) i i u u ρ i where r i is the satellite position at transmit time; r u is the receiver position at receive time; b u is the bias in the receiver clock, c is the speed of light and ε ρ is the i 35

55 Chapter 2 Background composite of errors. Parkinson (1996a) provides an error budget for ε ρ under various conditions. The states to be estimated are r u and b u. The linearized version of the above equation about a nominal point ( rˆ u, bˆ u ) is as follows: Δ Δr G. (2-6) u ρi = + Δε ρi c. Δb u Where 1ˆ r rˆ i u i =, u u u ri rˆ u Δr = rˆ r, Δ bu = b ˆ u b, u ε ρ = ε i ρ ε i ρ i Δ ˆ and G [ 1ˆT 1] =. i Or briefly: Δρ = G Δx + Δε. ρ Δ ε ρ is assumed to be zero mean, so that the least squares solution to the set of normal equations is given by T 1 T Δ xˆ = ( G G ) G Δ ρ (2-7) It is easy to see in this equation that the position accuracy is decided by two factors, the measurement quality and the user-to-satellite geometry reflected in matrix G (columns of G are the unit vectors pointing at the satellites). Each of these factors is separately and extensively discussed in Parkinson (1996a). In this section we will make a quantitative comparison of the effect of each one of these factors on the positioning accuracy. The aim is to find out if Δ ε p of the measured pseudorange of a i satellite is large because of the poor signal quality (low C/No), under which circumstances we can achieve better position accuracy by eliminating that satellite, noting the fact that eliminating the satellite will affect the geometry. This is investigated by applying error to different satellite pseudoranges in a real data set in Chapter Receiver Autonomous Integrity Monitoring (RAIM) It will be shown in Chapter 5 that degradation in satellite signal quality (C/No) due to interference can prevent one or more satellites being available at a given time. 36

56 Chapter 2 Background In Fante (2000) the probability of availability of N satellites in the presence of interference in terms of C/No is investigated. Due to the nature of this impact of CW RFI on the GPS satellite signals, one can reasonably think of using a RAIM algorithm to mitigate this effect. Position domain errors are assessed with the use of traditional least-squares estimation (Kim, 2006) and divergence estimation (Yun, 2006) in the presence of interference, and the RFI is mitigated using integrity monitoring techniques. In Oshman (2006), using the Interacting Multiple Model (IMM) estimator, each satellite is assigned a probabilistic measure of its health, and the solution is obtained by probabilistically weighting all available data. Thus, the implied decision on the health status of each satellite is not a hard, binary one and is not achieved by comparing a statistic against an externally set threshold. Rather, the (implicit) decision is soft and is achieved by comparing each satellite s posterior probability against the probabilities assigned to the other satellites. This yields an improved identification capability and enables an extended operational envelope. System integrity refers to the ability of a system to provide timely warnings to users as to when the system should not be used (Parkinson, 1996b). If for any reason a satellite signal is not available or is defective, the receiver needs to produce an alarm to disregard this signal and to use other satellites instead. This task can be done by receiver autonomous integrity monitoring algorithms inside the receiver. The first requirement for such an algorithm is obviously redundancy. RAIM needs at least 5 satellites to detect the signal anomaly and 6 satellites to identify which satellite is defective, satellite geometry permitting. There are a number of different approaches to RAIM which can be put into two categories; the snapshot scheme in which only the current redundant measurements are used in a self-consistency check (Lee, 1986) and a scheme where both past and present measurements along with a priori assumptions with regard to the vehicle motion are used in the RAIM decision (Brown, 1986). The focus in Chapter 6 will be on the first scheme. The range comparison method, least-squares-residuals method, parity method and maximum separation of solutions method are the major approaches in the snapshot scheme. We use the final one of these techniques. In the maximum separation of solutions method, if there are n satellites in view, one can consider the n subset solutions obtained by omitting one satellite at a time from the full set. If a failure exists, the failed satellite is omitted from one of the subsets, and the solution thus obtained is declared a "good" solution. All other subsets 37

57 Chapter 2 Background contain the failed satellite, and they are in error to various degrees. If the pseudorange error in the failed satellite gradually increases with time, one would expect the subset solutions to begin to spread apart with time, and the maximum separation observed among the n solutions can be used as a measure of the solutions spread. The "good" solution remains near truth, because it does not contain the failed satellite. If there is no failure present, the solutions should remain grouped around the true position. Thus, the maximum observed solution separation in the horizontal plane can be used as a scalar and nonnegative test statistic. The "no-failure" decision is separated from the "failure" decision with a given detection probability by the threshold that can be set by utilizing the relationship between the radial error and the radial-error-protection level. The detection probability varies with satellite geometry. 2.9 Hypothesis Testing and Statistical Definitions Definitions A hypothesis is a statement about an unknown population parameter. In hypothesis testing we are interested in testing between two mutually exclusive hypotheses, called the null hypothesis (denoted H 0 ) and the alternative hypothesis (denoted H 1 ). H 0 and H 1 are complementary hypotheses in the following sense: If the parameter being hypothesized about is θ, and the parameter space (i.e., possible values for θ ) is Θ, then the null and alternative hypotheses form a partition of Θ : H 0 :θ Θ 0 Θ H Θ c 1 :θ 0 Θ c Θ 0 is the set of all test statistic values for which H 0 will be rejected. This region is called rejection region. A test statistic, similar to an estimator, is just some real-valued function Tn T ( X 1,..., X n ) of the data sample X 1,..., Xn. Clearly, a test statistic is a random variable. A test is a function mapping values of test statistic into {0, 1}, where: hypothesis H 1. 0 implies that you accept the null hypothesis H 0 reject the alternative 38

58 Chapter 2 Background 1 implies that you reject the null hypothesis H 0 accept the alternative hypothesis H 1. In this chapter, we refer to a test as a combination of both (i) a test statistic; and (ii) the mapping from realizations of the test statistic to {0, 1}. Normally, we start with the research hypothesis and set up the null hypothesis to be directly counter to what we hope to show. We then try to show that, in the light of the collected data, the null hypothesis is false. We do this by calculating the probability of the data if the null hypothesis is true. A p-value is a measure of how much evidence we have against the null hypotheses. The smaller the p-value, the more evidence we have against H 0. It is also a measure of how likely we are to get a certain sample result or a result more extreme, assuming H 0 is true. The type of hypothesis (directional or non-directional: see Figure 2-13 and Table 2-5) will determine what more extreme means. When using some types of inferential statistics the alternative hypothesis may be directional or non-directional. A directional hypothesis (or one-sided hypothesis) is used when either only positive or negative differences are of interest in an experimental study. For example, when an alternative hypothesis predicts that the mean of one sample would be greater (but not less) than another, then a directional alternative would be used. This type of statistical procedure is known as a one-tail test. A non-directional (or two-sided) hypothesis would be used when both positive and negative differences are of equal importance in providing evidence with which to test the null hypothesis. Probability Density function S 1 = P value/ 2 S 2 = α / 2 t 0.3 t α / 2 t α / S 2 S Test statistic (t) t Rejection Region Acceptance Region Rejection Region Figure 2-13 Tail probability and p-value 39

59 Chapter 2 Background In Figure 2-13 the dashed blue area (S 1 ) is the p-value of the observation t. Also t α / 2 and t α / 2 are the values of test statistic beyond which the null hypothesis is rejected. For an α level significance test the p-value is the area S 2. The p-value measures consistency by calculating the probability of observing the results from your sample of data or a sample with results more extreme, assuming the null hypothesis is true. The smaller the p-value is, the greater is the inconsistency. A test with significance level α is one for which the probability of false alarm (P fa ) or the probability of rejecting H 0 when it is actually true, is controlled at a specified level (Devore, 1995). In Table 2-5, for some typical values of α, the corresponding values for test statistic are given for both directional and nondirectional test. Table 2-5 Level of significance for a directional and non-directional test with degree of freedom (df) = 28 Level of significance for a directional test Level of significance for a nondirectional test t=1.7 t=2.05 t=2.47 t=2.76 t=3.67 In parameter estimation, an interval of plausible values for the parameter being estimated is called the confidence interval (Wackerly, 1996). Usually, we use the term confidence interval (CI) to refer to a combination of an interval estimate, along with a measure of confidence (such as the confidence coefficient). Hence, a confidence interval is a statement like θ is between 1.5 and 2.8 with probability 80%. This interval is found using a pivotal quantity (Wackerly, 1996) given a confidence coefficient. This quantity is a random variable which is a function of the parameter in question and the random variables X,..., X 1 n but whose distribution is independent of that parameter. When we create CI's by inverting tests, the relevant pivotal quantity is the test statistic. 40

60 Chapter 2 Background Inference Based on Two Data Populations The inferences discussed in this section, concern a difference μ1 μ2 between the means of two different independent population distributions. In our case for example, we wish to test the difference between true average powers of two different signals at the same frequency bin. One such hypothesis would state that μ 1 = μ2. Alternatively, it might be appropriate to estimate μ1 μ 2 computing a 99.9% CI. In real problems, it is virtually always the case that the values of the population variances are unknown. For large sample sizes, the sample variance is used in place of population variance in the test procedure. The assumption of large sample size is made to use the properties of the central limit theorem (CLT). In fact the CLT allows us to use these test methods even if the two populations of interest are not normal (Devore, 1995). The approximation that the CLT leads to normality is quantified by the Berry Essen theorem in Feller (1971). It is shown that this approximation has an upper bound which is determined by the number of samples. In doing a large sample T-test, for the two population X,..., X 1 n and m Y 1,...,Y with corresponding sample means x, y and true means μ 1, μ 2 and a common sample variance S p, the null hypothesis, the test statistic, the alternative hypothesis and the rejection region for a specific significance level of test will be as follows. Null hypothesis: H 0 : μ1 μ2 = Δ 0 where in this case Δ 0 = 0. Test statistic value: t = x y Δ S p n m (2-8) which has a t distribution with m+n-2 degrees of freedom (df), where 2 pooled estimator of the common varianceσ (Wackerly, 1996). Alternative hypothesis: H : μ1 μ 2 Δ 0 a with rejection regions for levelα test: t tα 2, 2 or t 2, 2 / m+ n t α / m+ n 2 S p is the For the cases where the variances of the two populations are not equal, the following procedure (called the Smith-Satterthwaite test (Wackerly, 1996)) is known to be an approximately level α test 41

61 Chapter 2 Background t = Test statistic value: x y Δ s1 s2 + n m where s 1 and s 2 are sample variance of the populations. (2-9) 2.10 Previous Interference Detection, Characterization and Mitigation Techniques Detection An interfering signal at GPS frequencies may cause GPS performance in the vicinity to be degraded or lost entirely. Such a signal may affect both GPS-based positioning and GPS-based time services, and the performance loss may take place regardless of whether the interfering signal is due to intentional jamming by a hostile entity or if the signal is transmitted unintentionally (for example as a consequence of equipment malfunction). Either way, it is important to know if the GPS signal is disturbed (or jammed), in order to determine whether or not the GPS data is reliable. The thermal noise power of the receiver with 2 MHz bandwidth is -110dBm. The GPS signal power is -130 dbm which is 20 db less than the power of noise. Therefore interference signal power even lower than the thermal noise power can affect normal operation of the receiver. Parkinson (1996a) notes that the CW signal is the most disturbing when the interference is coherent with the GPS carrier frequency. In this case the CW passes through the carrier tracking loop filter and seriously affects the correlator output. A GPS jammer/interference detector is used to ensure that the integrity of the GPS signals is preserved. It warns the operator if an interfering signal is detected at GPS frequencies. Therefore, by using this product an operator gets a confirmation that the GPS signal is (or is not) undisturbed, and thus that his/her other GPS-based equipment is (or is not) operating under the best possible conditions. The task of detection can be performed either in the RF front end or in the base-band block. In the RF front end, it affects many of the components and these effects can be exploited for the purpose of RFI detection. These effects have already been examined in the previous sections. The Automatic Gain Control (AGC) for instance has been specifically used for the purpose of RFI detection (Bastide, 2003; 42

62 Chapter 2 Background Ward, 2007). Automatic Gain Control (AGC) is a very important component in a Global Navigation Satellite System (GNSS) receiver. Such functionality is required any time multi-bit quantization is implemented so as to minimize quantization losses. In GNSS receivers, where the signal power is below that of the thermal noise floor, the AGC is usually driven by the ambient noise environment rather than the signal power, and hence is quite stable. As a result, AGC can be a valuable tool for detecting changes in the operating environment of a GNSS receiver. AGC and its functionality are investigated throughout this thesis. Behavior of this system, both in the nominal thermal noise environment as well as in the presence of various types of interference, are examined. Its usefulness as an interference detection and estimation tool is highlighted in several tests using simulated and real signals. Moreover, the frequency assignment of the proposed GPS L5 signal is expected to face a significant interference environment as a result of other aviation systems operating in or near that band. A specific AGC and analog-to-digital converter (ADC) design performing digital pulse blanking and dedicated to L5 receivers is also studied. In Bastide (2003) it was shown that the AGC system is an accurate indicator of the noise environment of the receiver. Its gain varies with respect to the interference power present and so is a valuable tool to detect it. Moreover the ADC bins distribution may be also used to detect interference by using a simple Chisquare test detecting distribution changes. Adaptive ADC was used in Amoroso (1983) to mitigate the effect of RFI. RFI also affects the components within the base-band block. It affects the acquisition process and the tracking loops. Nidli (1998) and Luo (2003) extensively investigate these effects and use a statistical approach for characterization. Types of interference considered include CW and broadband, pulsed and continuous. We study the effects of different types of interference on GPS receiver sub-units, including the ADC, correlator measurements, the PLL and the DLL. From analysis and simulation we present interference detection algorithms based on the observable effects of the various types of interference on the GPS receiver raw measurements. Interference detection is based on a combination of the following test statistics - correlator power output, variance of correlator power output, carrier phase vacillation, and AGC control loop gains (Ndili, 1998). Observing quantities like correlator output power and carrier phase jitter, which are driven by in-phase and quadrature phase samples of the output from the 43

63 Chapter 2 Background correlators, is satellite/receiver channel specific. By contrast, the AGC is driven by the signal digitization process and is not a channel specific observable for detecting interference. Signal anomalies like multipath or intra-system and cross-correlation type interference cannot be detected by observing the AGC. Also the AGC behaviour is dependent on the environment temperature. Because the AGC uses noise to set its levels it is not possible to detect interference with power less than the thermal noise power using the AGC. The advantage of assessment of interference using the AGC in addition to it being a fast method is that it exploits an existing receiver component and it is not channel- or satellite-specific. In other words the satellite signal does not need to be detected first before the interference is detected. This can be a drawback in detecting interference using post-correlation observables. Detecting RFI using these observables is often not fast enough to meet the Time To Alert (TTA) requirement which is specified in (FAA, 1984). Detection of interference based on observing different observables may lead to identification of the type of interference, as different types produce different behavior. While AGC is very sensitive to pulse RFI with high peak power (Ndili, 1998), correlator output processing is very sensitive to CW type RFI (Parkinson, 1996a). In Yun (2006) the presence of RFI is determined by monitoring a metric which is called divergence estimation. This metric is a time difference of pseudorange minus carrier phase which is effectively an estimation of the receiver positioning error. The low sensitivity of this technique to the type of RFI is shown in Yun (2006). This detection scheme is then used for a mitigation algorithm. It is shown that this technique is less sensitive to the type of interference received. RFI power levels of less than -95 dbm cannot be detected using the specific type of the receiver used in that work. This technique was clearly improved upon by the -135 dbm achieved by the techniques presented in Chapter 4 of this thesis. Overall the earlier stages of the receiver are faster in detecting interference but are more sensitive to the power of the interference. The later stages are less sensitive to the power and more robust for different type of interference. The subject of signal detection and estimation deals with the processing of information-bearing signals in order to make inferences about the information that they contain. Although this field clearly traces its provenance to the classical work of Bayes (1763), Gauss (1823), Fisher (1922) and Neyman (1933) on statistical inference, it was not until after the mathematical stage was set in the 1930 s by the 44

64 Chapter 2 Background pioneers of stochastic processes - Kolmogorov, L evy, Wiener, and others that the field took hold as a recognizable discipline (Kailath, 1998). The scope of the field of signal detection is very large, with several useful textbooks available, including Helstrom (1995), Poor (1994) and Kay (1998). For signal detection when the a priori probability distribution function is not known for the signal, further statistical concepts must be introduced, e.g., locally optimal tests, etc. (Poor, 1994). Also locally optimal approaches are suitable for detecting weak signals. Processing the received GNSS signal - from the raw IF stage to the higher levels of base-band processing - has become more versatile with the advent and developments of software receivers (Akos, 1997, Sharawi, 2007; Tabatabaei, 2006). It was on this basis that Marti (2004) for the first time exploited the idea of signal detection in the context of RFI detection. That work focused on CW RFI and the IF data at the output of the RF front end and introduced a test statistic with which the detection could be carried out without a priory information about the interference probability distribution. The only problem was that the components inside the RF front end can affect the signal and ultimately the RFI detection schemes (Marti, 2003). Front-end induced errors are discussed in Marti (2003) and the statistics of the signal parameter estimators that are affected by the front-end are derived. It was shown that amplification and frequency mixing do not change the statistical properties of the received signal. However, band pass filtering introduces correlation into the data that leads to estimation errors. In Marti (2004) it is shown that longer observation windows allow detection of lower power RFI. However, acquiring more information may make the decision easier and the detection confidence greater, but the processing load also increases. One question is if given a fixed data window size, is there an optimal way of making the decision about the presence of interference using the proposed inference technique in Marti (2004). We later show that based on how the data block is broken into subblocks, the detection threshold changes with a fixed probability of false alarm. Then it is shown that there is an optimal way to break up this data window to achieve the lowest detection threshold for a fixed value of probability of false alarm in which the computational load is also minimal. Then an optimal detector algorithm is proposed. At this optimal point, for a fixed Detection Threshold (DT), probability of false alarm becomes minimal and for a fixed probability of false alarm, we can achieve the 45

65 Chapter 2 Background minimum value for the detection threshold. Experiments show that at this point we have the minimum computational load Characterization of the Interference Effects If the operation of the GPS receiver is not yet stopped because of the presence of RFI, then degrading the accuracy of positioning is the highest level of effect that the RFI can have on the functionality of the receiver. An operating GPS receiver must be frequency-locked onto the modulated and Doppler shifted carrier frequency, delay-locked onto the C/A code, and phase-locked onto the carrier. Thus at a bare minimum, an operating GPS receiver is frequency-, delay-, and phase-locked to a GPS signal. Two common testing methodologies are used to measure the effect of interference on receiver operation or locking (Hoffman, 2001). 1. Break Lock (BL). This point is defined to be a minimum amount of interference that causes a receiver to lose lock. 2. Reacquisition Time (RQT). This point is defined to be a maximum amount of interference that takes a receiver tracking a GPS signal to reacquire the signal after it has been momentarily removed in a specific time. These two points bracket a region of GPS receiver performance degradation. The RQT point sets the lower bound where the interference begins to have a detrimental effect on the operation of the receiver. The BL point sets the upper bound where operation is impossible. Apart from the receiver operational effect which set two boundaries for the amount of interference, RFI has characterizable or measurable effects on many of the components in the GNSS receiver. Here we start from the lowest level components, which are the nearest components to the antenna to the high level components (or observables), which are close to the positioning evaluation blocks with the focus on the works, which led us to make our contribution on the characterization of the effect of RFI on the GNSS receiver. A categorization in this trend is necessary at this stage. Recently the idea of software defined radio has attracted the GNSS receiver designers (Akos, 1997). The idea in this type of receiver is to move the analogue to digital converter as close as possible to the antenna. The advantage is to be able to enjoy the possibilities and flexibilities that digital signal processing provides us. The Nyquist sampling theorem allows bandpass sampling, i.e. sampling at twice the signal bandwidth, so the ADC can follow a bandpass filter 46

66 Chapter 2 Background immediately after the antenna. This technology is not practically used very often in receivers because of the high Q required of that bandpass filter (required to eliminate all the aliases that appear in the baseband) and the high ratio of input bandwidth to sampling rate. The processing before the ADC on the signal is analogue signal processing and after the ADC it is digital signal processing. The assumption here is that the RF signal after being received by the antenna, and filtered, is down-converted to an IF and then passed through the ADC. In these analogue stages interference may saturate the filters in the RF front end and cause intermodulation products (Landry, 1998). Nonlinearity in LNA can even generate in-band RFI from out of band RFI. There are techniques to linearize the LNA (Vitzilaios, 2006). Slow AGCs will be severely affected by pulse interference. These AGCs are slow to respond and will incorrectly determine the quantization levels which will result in improper sampling (Heppe, 2003). The characterization of the effect of narrow band RFI on the operation of AGC is investigated in Bastide (2003) and Shi (2007). An ADC with a small number of bits will also be easily saturated by the jamming signal (Amoroso, 1983). In Amoroso (1983), a technique is proposed and investigated to use an adaptive ADC converter, to mitigate the effect of CW RFI in direct sequence pseudo-noise (DSPN) spread spectrum communication receivers. Right after the IF data and before it passes through the ADC, a quantity called effective carrier to noise density ratio ( C / No) is defined and calculated in the presence of narrow band CW interference in Betz (2001). In this work calculation is in terms of receiver front end bandwidth and RFI and GPS signal spectrum. This expression of effective C/No is compared with the experimentally measured C/No in Betz (2000) and it was shown that the theory and experiment match for narrowband RFI. Later in Betz (2002), another concept introduced was spectral separation coefficient (SSC), which distinguished the effect of interference power from the effect of particular spectral shape of the interference. Although slightly different expressions result from the use of coherent processing (where the carrier phase of the reference signal is matched to that of the received signal) and non-coherent processing (where the reference is not phase-matched), the resulting expressions for SSCs and C/No are the same (Betz, 2004). This will be discussed in more detail in the next chapter. The C/No of course is not a physical phenomenon. It is a concept introduced to give a measure of the received signal quality. The point is to evaluate the position by eff 47

67 Chapter 2 Background examining code phase and if C/No was going to be a good measure of the signal quality, it should show that lower ( C / No) meant higher code tracking error. Betz (2000) compares different techniques to estimate the code tracking error and also compares these results with the actual measured code tracking error. These steps were all taken before the acquisition block along with the other works that introduced techniques to estimate the C/No mainly using I and Q samples after the acquisition block (Sharawi, 2007). Also the advent of multicorrelator receivers widened the range of post-correlation methods that can be considered to tackle the problem of interference characterization. In particular, this enabled the characterization of interference effects on the tracking loops through the analysis of the shape of the correlation peak (Bastide, 2001). At this stage there was a need to take the results from Betz (2001) further from the case in which the bandwidth of the RFI cannot be less than the reciprocal of the correlator integration time, to very narrow bandwidth to see the effect of Doppler frequency on the SSC, ( / N and eff C 0 ) eff the receiver jamming resistance quality factor. This will be explained in detail in Chapter 5. This happened in two parallel works (Tabatabaei, 2006a; Borio, 2006). In Borio (2006) SSC (introduced first in Betz (2002)) was evaluated after digitization. In Tabatabaei (2006a), we came up with a closed formula for the value of effective C/No in the presence of CW interference. It is shown in that work that ( / N has to do C 0 ) eff with the frequency difference between the interference and the carrier or any integer multiple of reciprocal of the code period from the carrier which was consistent with the result in Karsi (2000) who chose a stochastic approach to look at the effect of CW interference on the phase lock loop (PLL). It also depends on the phase difference between the RFI and the carrier and the receiver integration time. In this work it is shown that the result from the C/No calculation using the proposed closed formula is exactly the same result achieved by using a C/No estimation scheme introduced in Dierendonck (1996). The prediction scheme is proposed in the analogue domain. C/No was estimated after the ADC considering the effect of quantization on the estimation introduced in the very recent work of Sharawi (2007). It was shown in this work that the quantization effect can be negligible (2.2 db-hz in the worst case when hard limiter used as ADC). Lower C/No means lower quality of the received signal. Galileo satellites will soon be in place operating together with GPS satellites. Considering the designer s 48

68 Chapter 2 Background intention of maintaining interoperability between different satellite navigation systems, it is reasonable to seek a quantified comparison between the different systems in terms of vulnerability to CW interference. In the next step considering each one s signal structures, the characterization of the effect of CW interference on the C/No for GPS and Galileo is investigated and compared. In Chapter 7 it is shown that for the available Galileo signal (GIOVE-A BOC(1, 1) in the E1/L1 band), the worst spectral line happens far from the L1 frequency. A frequency is selected which is midway between the GPS and Galileo worst spectral lines, and for the same power of RFI, GPS is shown to be more vulnerable to interference. Also the probability of availability of one GPS satellite is compared with that of one Galileo satellite in terms of interference power and frequency. In Chapter 7 it is also shown that these two systems can be considered as alternatives to each other in the presence of different RFI frequencies as their availability in the presence of CW RFI is different in terms of RFI frequency ( L5, L2C, E5 and E6 are not considered). In this work, the navigation data is not there. This might have an effect depending on the integration duration time in the tracking loops. Performance degradation of an operating GPS receiver is also evaluated through its other observables which include: pseudorange, carrier phase (Ndili, 1998; Enge, 2006), Doppler frequency-shift, clock-offset, and carrier cycle-slippage. There is significant opportunity for future investigation in this area Mitigation In section it was explained that post-correlation detection techniques have been recently investigated due to the advent of the software receiver and also development of the FPGA (Engel, 2004) and DSP (Benson, 2005) receiver test-bed platforms which gave the opportunity to researchers in this area to characterize the effect of interference on the components operating in the base-band block of the GNSS receiver in the first stage and come up with the ideas to mitigate it. Here we introduce a categorization for the RFI mitigation techniques. The mitigation can be generally considered in one of two main categories; non-receiver-based and receiverbased techniques. First we focus on non-receiver based methods. These techniques obviously consist of all the approaches to lower the vulnerability of a GNSS receiver to interference, which are performed independent of 49

69 Chapter 2 Background the receiver. Out of band augmentation, which can be considered as a kind of backup planning or system-level redundancy, is one of these techniques. Enhanced Loran, or eloran, is independent of GNSS but fully compatible in its positioning and timing information and its failure modes are very different. eloran is based on the existing low frequency Loran-C infrastructures that exist today in the United States, Europe, the Far East and throughout much of the northern hemisphere. It is an internationally recognized positioning and timing service. eloran operates on a 100 khz (low frequency) carrier at very high power levels. It is virtually impossible to jam and at that frequency propagates indoors far better than GNSS signals. Those attributes make it attractive as a backup and extension to GNSS for most applications (Jacobson, 2007). Another backup plan can be to use navigation alternatives (e.g. inertial measurement units and GPS/INS integration). This focuses on a U.S. Department of Transportation (DOT) action plan released on March 7, 2002, that seeks "to maintain the adequacy of backup systems for each area of operation in which the GPS is being used for critical transportation applications." This action plan follows the GPS vulnerability report (Volpe report, 2001) that identifies susceptibility of GPS to unintentional interference caused by atmospheric effects, signal blockages from buildings, communication equipment, and potential intentional jamming. Finally, by better signal characterization and transmission (e.g. wider bandwidth and higher transmission power), it is possible to have a GNSS receiver more robust in the presence of interference. The spreading of the GPS signal by the PRN code not only distinguishes each signal from the others, but also gives a level of protection against interference. Immunity against interference is highly dependent on the processing gain of the system. The higher the processing gain, the wider the GPS signal is spread. By spreading the signal over a wide bandwidth, only a fraction of the desired signal can be corrupted by narrow-band interference. The despreading process in the receiver despreadds the wanted signal and spreads narrow-band interference. The wider it spreads the less effective it will be on the operation of the correlator (Parkinson, 1996a). In the receiver-based approaches, the receiver design assumes the existence of RFI. The RFI suppression philosophy is to lower the effect of RFI on the positioning quality as much as possible. But as was explained earlier, there are different types of interference with different characteristics and with different effects on the various 50

70 Chapter 2 Background parts of the receiver. Thus the idea of stopping the interference from getting into the receiver at as earlier a stage as possible may not very well serve the RFI suppression philosophy explained earlier. As an example, the spectral separation coefficient between the RFI and the GNSS signal introduced in Betz (2002) indicates that maybe the adverse effect of a particular RFI on positioning is less than the adverse side effect of applying a suppressing technique to eliminate that RFI. It is because of this important consideration that RFI mitigation should be carried out with extreme care. The other limitation in this field is the fact that the proposed technique should be applicable for a civilian GNSS receiver aiming to serve a huge mass market. This will restrict us to very cheap and yet clever ideas to mitigate the RFI, which cannot be achieved without considering signal and interference source and effect characterization. The other stage to be used to suppress the RFI is the low noise amplifier. Typically, the designers put filters at both sides of the GPS LNA. A filter in front of the LNA helps to reject the out-of-band signal and prevent the LNA from being saturated. This filter should have a very low insertion loss. Putting a high insertion loss filter before the LNA should be avoided because this will increase the system's noise figure. According to Friis equation (Spiegel, 2003), the total noise figure is dominated by the noise figure or loss of the first stage. A second filter at the back of the LNA can be used to further improve the out-of-band rejection to prevent the later stage from being overloaded (Spiegel 2003, Chow 2006). The fact that interference sources are usually distributed on the ground distinguishes the interference from the GNSS satellite signal in terms of the direction of arrival to the receiver antenna. This is a basis of all the adaptive array antenna designs to either null the interference (Wang, 2007) or beam the satellites (Brown, 2000) or both to improve the signal to interference ratio which will be shown to have significant effects on the effective carrier to noise ratio. Conventional spatial techniques can adaptively null the interference, but they are incapable of cancelling multiple narrowband interferences due to the limited number of degrees of freedom. In Soderstrand (1997), a real-time adaptive notch filter approach is presented to detect and suppress multiple sinusoidal RFI. Space time adaptive processing (STAP) was the next stage, however computational complexity is an issue in these techniques. The reduced dimension technique proposed by Myrick (2001) has made a significant improvement in this regard (Lu, 2006). All of these algorithms have some effects on 51

71 Chapter 2 Background the desired signal as well. Research is undertaken to minimize this effect using adaptive filtering techniques (Soderstrand, 2006). In Bucco (2004) the FPGA (Field-Programmable Gate Array) implementation of an adaptive filter for narrow band interference excision in Global Positioning Systems is described. The algorithm implemented is a delayed LMS (Least Mean Squares) adaptive algorithm. Some implementation issues about the filter weights are addressed and discussed in that work. As the last stage in the RF front end, the analogue to digital converter (ADC) has sometimes been used to mitigate the interference. In (Amoroso, 1983) an adaptive ADC approach is proposed for that purpose. The problem with that technique was that it was suitable for strong CW RFI and for the cases in which AWGN was dominant the technique was ineffective (Goiser, 1989). After the ADC, we have the acquisition and tracking blocks. In the previous section it was explained how the characterization of the effect of CW interference was extended up to this stage and here it is explained how it is used in introducing a new algorithm to mitigate the effects of CW RFI on the positioning evaluation. First, the new concept of exclusion zone is defined, characterized and analysed for each satellite. This exclusion zone is shown to be predictable for each satellite as a function of time. Using this prediction, the CW interference effect on the positioning quality of the receiver is mitigated by cancelling the affected satellite within the exclusion zone during the position evaluation. Then the decision of cancelling a particular satellite (the threshold beyond which a satellite should be excluded), is worked out by studying the mutual effect of the geometry and the signal quality of that satellite on the positioning quality. This algorithm does not have the computational complexity that the array signal processing and also the adaptive filtering have. Neither does it need any hardware complexities like array antenna. It also does not have any adverse effect on the desired signal after suppressing the effect of RFI as techniques like notch filtering have. Receiver Autonomous Integrity Monitoring (RAIM) is another mitigation approach (Kim, 2006; Yun, 2007). Specifically for the CW RFI which can affect one satellite at a time, this technique proves to be effective. However, our ability to predict the impact of CW RFI on the signal quality of the GPS receiver can achieve better results than simply applying RAIM algorithms. Specifically, in the static applications such as Continuously Operating Reference Stations (CORS) network base stations where the environmental characteristics such as multipath 52

72 Chapter 2 Background (Weiss, 2005) and interference model do not change, this proactive approach to monitoring the satellite signal quality and availability is shown to have advantages Localization Besides the existing suppression and mitigation techniques developed in the receivers, it would be if once RF interference is detected, one could invoke a fast location and removal response, because of the high military and civilian reliance on GPS. Many works have investigated this. Leung (1997) has done an investigation on using array signal processing in finding the number of interference emitters and their directions. A very general and comprehensive review of the developed techniques in array signal processing can be studied in Krim (1996). In the same year, Geyer (1997), described the data processing algorithms embedded in the Aircraft RFI Localization and Avoidance System (ARLAS), which was being developed at that time by the U.S. Department of Transport (DOT) Volpe National Transportation System Center (Volpe Center), under sponsorship of Federal Aviation Administration (FAA). In this technique, an RFI source was located using a GPS receiver, gyros and processor in the aircraft. Back in 1999, finding the geo-location of RF interference source was most noted among the other issues in GPS receiver jamming and RF interference. Wohlfiel (1999), described a project that had been directed by U.S. congress starting in 1997 to detect and locate sources of collateral interference and intentional jamming, and to assess the technical feasibility and utility of this concept on an airborne platform. This system consists of a short baseline three antenna array; a high sensitivity, three channel fast scanning receiver; three interferometric direction finding processors; a state-of-the-art signal processor and a DAT tape drive. It was stated in this paper that this airborne concept can also be adapted for a static implementation around airfields and ports for a continuous monitoring/surveillance capability. This project was called Location of GPS Interference (LOCO GPSI). In Brown (1999), the problem of locating large numbers of low power jammers is addressed by proposing a network centric jammer location system and using three types of different sensors (C/No, AOA and TDOA jammer location sensor). No performance figures for this system were given. Gromov (1999) also presented a hardware platform to measure the signal propagation delay along the baseline between two antennas to estimate the direction of the signal. An expanded version of 53

73 Chapter 2 Background this system was introduced in Gromov (2000). This prototype is called Generalized Interference Detection and Localization System (GIDL). It includes four antennas and RF sections slaved to a common clock to allow three-dimensional interference location. Influence of multipath and also low signal power on the accuracy limitation of using adaptive antenna technology in RFI direction finding was noted in Bond (2000). Finally Jan (2001), considers two methods to estimate jammer location. One method is based on jammer bearing measurement at different points along the sensor trajectory, or from the sensor array. Another method is based on the Doppler shift of the jammer frequency caused by the relative motion between the sensor and the jammer. In all the above works, signal direction finding, signal time difference of arrival or received signal strength have been used to estimate the location of RFI. Geyer (1997), Wohlfiel (1999), Brown (1999), Bond (2000) and Jan (2001) have proposed an airborne framework to implement these techniques. Leung (1997) and Gromov (2000) have introduced a centralized static framework for this purpose. In this thesis, the approach of network-based wireless location is introduced to localize the RFI sources. Accurate and low-cost sensor cooperative localization is a critical requirement for the deployment of wireless sensor networks in a wide variety of applications. Sensors must be inexpensive and energy efficient devices and sensor networks must be scalable from dozens to millions of devices (Patwari, 2005). Wireless location refers to the geographic coordinates of a mobile subscriber in cellular or wireless local area network (WLAN) environments. Wireless location finding has emerged as an essential public safety feature of cellular systems in response to an order issued by the Federal Communications Commissions (FCC) in 1996 (Sayed, 2005). The idea of network-based wireless location has been improved and deployed for many localization purposes. 54

74 Chapter 3 REAL GNSS SIGNAL INTERFERENCE 3.1 Introduction Several authors have already reported GPS interference in different countries (Clynch, 2003; Hutchinson, 1994; Buch, 1995; Borza, 1999; Berggraf, 1995; Scaramuzza, 1995). This is the first time, however, that such disturbances are reported from Australia. There are several TV channels generating harmonics of order smaller than 10 which cause interference problems for GPS receivers. By studying the Draft License Area Plan for Sydney Analog Television (ABA, 2002), we found that the third harmonic of a transmitter site located in Artarmon, with Easting and Northing , can be received within the GPS receiver bandwidth. Beatrice Motella has helped the author collecting and analyzing this data. This work is published in Tabatabaei (2007e). The data sets were collected in Ampden Road near the TV transmitter tower at about 6 km from the city centre (see Figure 3-1). Figure 3-1 The distance between Sydney city centre and Hampden Road, Artarmon, where the data collection took place from Google Earth The tower hosts the TV transmitter of Channel 28 (SBS), which has the carrier frequency at MHz in the band IV of Ultra High Frequencies (UHF) (ABA, 2002). The TV transmitter power and the detection of the TV channel 28 transmission by means of the spectrum analyzer are shown in Figure

75 Chapter 3 Real GNSS signal interference Figure 3-2TV Tower in Hampden Road - Artarmon (left) and the detection of the TV transmission (at MHz) by means of the spectrum analyzer (right) The interferer is an analogue TV channel (Phase Alternation Line - PAL), which means that it is characterized by a vestigial sideband 8 MHz wide. The lower band edge is at MHz from the carrier frequency and the upper band edge is at +6 MHz. The carrier frequency of this transmitter is fc = MHz. This means that its 3 rd harmonic is at MHz. Consequently the lower frequency of the 3 rd harmonic is at 3*1.75 MHz = 5.25 MHz from MHz. Its distance from L1 central frequency is only 1 MHz ( MHz), which means that it falls within the front-end filter bandwidth of the receiver used in this test: The data were collected on 12th July 2006 at am. For the data collection the NordNav R-30 GPS software receiver has been used. The NordNav R-30 is a software receiver that allows us to save the raw data and post-process it. The data is stored at the output of the AD converter using a sample frequency of MHz. The Intermediate Frequency is at MHz and the front end -3dB bandwidth is 3.78 MHz. During the experiment a patch antenna was used. The antenna height of this transmitter is 216m and it transmits at all angles of azimuth with the maximum ERP of 850 kw. The NordNav R30 software receiver and a spectrum analyzer were simultaneously used to capture the data both in the L1 GPS frequency (Figure 3-3) and at the carrier frequency of the transmitter (Figure 3-2). 56

76 Chapter 3 Real GNSS signal interference Figure 3-3 The signal transmitted by the TV broadcaster antenna at the GPS L1 frequency (peaks indicate the interference) As can be seen in Figure 3-3, a harmonic of the transmitted signal from this TV antenna lies within the GPS L1 frequency bandwidth. In fact this interfering signal is strong enough to be detected easily by the spectrum analyzer. The start and the stop frequency of the display is set on MHz and MHz. The strong peak occurs almost in the L1 band center. In Figure 3-2, the central frequency of the display of the spectrum analyzer is set to 527 MHz. It can be seen that the carrier frequency is at MHz. 3.2 Source and Effect Characterization In this section the effect of the interference on the NordNav R-30 GPS receiver is analysed. The first step in detecting the presence of undesired electromagnetic sources is the evaluation of the signal spectrum. By analysing the spectrum of the signal it was possible to detect the presence of several spikes that distort the shape of the signal spectrum at the front end output. The spectra were evaluated using a 4096 sample Fast Fourier Transform (FFT) and 256 windows for the Welch technique (Proakis, 1994). Because the sampling frequency is MHz, the resolution bandwidth is equal to 4 khz. Starting from the data collection start time, the spectrum has been evaluated versus time every 5. Figure 3-4 shows three different snapshots for the spectrum evaluated respectively at 1 15, 1 50, and 2 55 with respect to the start time of the data collection. The spectrum is estimated using the data extracted at the output of the 57

77 Chapter 3 Real GNSS signal interference Analogue to Digital Converter (ADC) block. It is possible to recognize the shape of the front end filter that has a 3dB bandwidth of about 3.8 MHz. Figure 3-4 Spectrum of the received signal at different times (at 1 15, 1 50 and 2 55 from the beginning of the data collection respectively) Over time several peaks were detected. The strongest and most frequent are at the frequencies: MHz; MHz; MHz. Further investigation into the impact of these peaks on the operation of the receiver was done. Figure 3-5 examines the relationship between the height of the spikes and the behaviour of the Automatic Gain Control (AGC) level. The height of the peaks was measured with respect to time and the noise floor level for about 3 minutes. The spike at MHz (in pink) is the most frequent and it is present for almost the whole time interval; its height varies up to 11 db above the noise floor. It is strongest in the time interval between 42 and 64 seconds. The spike at MHz (in red) is less frequent and weaker. It is present in the time interval seconds and it is has a maximum of 6 db at around sec 45. The spike at MHz (in green) is the least frequent, but strongest. Its presence is limited to around second 80, but it emerges from the noise up to 13 db. 58

78 Chapter 3 Real GNSS signal interference Figure 3-5 Height of the spikes w.r.t. the noise level vs time (above) AGC gain vs time (below) Comparing the behavior of the height of the peaks with the AGC level versus time (Figure 3-5), it is easy to observe how the AGC gain is affected by the presence of interference sources within the received signal. In order to allow the ADC to work properly, the aim of the AGC is to maintain the incoming signal within a certain amplitude range. To do that, when the interferer is present, the AGC modifies its gain, compensating for the increasing/ decreasing power level of the received signal. In detail, it is possible to observe how the AGC changes its gain in the time interval sec (when the peaks at and MHz occur), around second 80 (when the peak at MHz is on) and in the interval sec (when the peak at MHz is again present). In the absence of interference, the AGC would normally not change value in the period observed in this experiment. Since the automatic gain control has the aim of maintaining the power of the incoming signal, the presence of undesired sources makes its behavior also affect the power of the useful signal, i.e. when interference is strong, the wanted signal is attenuated. In order to have a quantitative measure of such a loss, the average of the spectrum within 1 MHz of bandwidth has been evaluated. This is the portion of band 59

79 Chapter 3 Real GNSS signal interference where the GPS signal has the strongest frequency components. Figure 3-5 shows an example between sec 70 and sec 100. It indicates the relationship between the average of the spectrum (evaluated within 1 MHz around the central frequency) and the behavior of the AGC. Within this time interval, due to the presence of the strong spike at MHz, the AGC adjusts its gain, attenuating the received signal. The consequence is a loss of up to 7 db (from -62 to -69 db) within the frequency band where the GPS signal is present. This effect is shown in Figure 3-5 (seconds 80-85). -55 Spectrum Av in band [dbw] time [sec] -4-6 AGC Level [db] time [sec] Figure 3-6 Spectrum average within a band of 1 MHz around the central frequency (above) AGC level versus time (below) 3.3 In-band RFI Detection Depending on where the interfered data is collected, different levels of interference can be received by the GPS receiver antenna. If the receiver antenna is in the location where it can receive the main lobe of the transmitted signal, then the received power can be enough to saturate the RF front-end of the receiver and possibly stop the normal operation of acquisition and tracking of the receiver. The power spectral density of the received signal in this situation is shown in Figure 3-7. It can be seen that the interferer is spread across the band and the reason it looks stronger in the band center may be because of the NordNav receiver front end bandwidth. 60

80 Chapter 3 Real GNSS signal interference -30 time = 129 s mag [dbw] freq [MHz] Figure 3-7 Interference-affected data collected in a location from where the main lobe of the transmission is received at a specific time (time = 129s into the collected data sequence) To get an idea of how much power is received by the receiver antenna, the Automatic Gain Control (AGC) level of the receiver was recorded. The AGC is the first component in the receiver that reacts to the input signal power. If the interference power is higher than the environmental thermal noise power, then the AGC, by changing the gain, will try to keep the signal to noise ratio at its optimum value (Parkinson, 1996a) at the output of the RF front end. It will continue decreasing the gain with the increasing input signal power till it saturates. This also attenuates the GPS signal. The characterization of the behavior of this component in the NordNav receiver RF front-end in the presence of wideband noise within the L1 band is investigated here. This characterization allows us to estimate the amount of interference power, received at each moment. To do this experiment the hardware setup of Figure 3-8 was used. 61

81 Chapter 3 Real GNSS signal interference Figure 3-8 The hardware setup to characterize the behavior of AGC level of NordNav software receiver In Figure 3-8, to generate wideband noise, a three stage amplifier with each stage having a gain of 27 db providing in total a gain of 81 db is used. At the output of this amplifier there is a REACTEL filter with a bandwidth of 20 MHz and a central band of MHz. This amplified and filtered signal is passed through a programmable attenuator to control the input noise power. This signal is finally fed into the GPS receiver. A DC power supply provides power for these components. 10 Receiver AGC Characterization 5 AGC Level Wideband Noise (dbm) Figure 3-9 Characterization of the behavior of AGC level w.r.t the received in-band power in the NordNav software receiver 62

82 Chapter 3 Real GNSS signal interference This experiment is performed in a few steps where the input noise power changes by adjusting the programmable attenuator. Figure 3-9 shows the level of the AGC versus the input power. The figure shows that the AGC decreases the gain once the input power is high. Figure 3-10 and Figure 3-11 show the histograms of the IF data when the input power is less than the thermal noise and also in the situation when the AGC is saturated because of the high amount of received signal. 18 x Histogram of IF data Receiver front-end quantization level Figure 3-10 Normal histogram of the IF data when the AGC is in its linear operation 12 x Histogram of IF data Receiver front-end quantization level Figure 3-11 Histogram of the IF data when the AGC is saturated by interference The intermediate frequency (IF) data from the receiver radio frequency (RF) front-end is sampled by a four-bit ADC, giving 16 levels of quantization. Figure 3-10 shows that the histogram in the situation in which the AGC is working in its linear normal operation looks like a Gaussian distribution, as can be expected given the 63

83 Chapter 3 Real GNSS signal interference signal appears simply to be noise. Alternatively if the interference or the noise power received by the receiver is high it can saturate the AGC in a way such that it can no longer keep the histogram in a Gaussian shape. In an extreme situation the histogram can become like that shown in Figure This histogram is for real data suffering from interference collected from the TV transmitter antenna where the received power of interference is very strong. In Figure 3-12, the AGC level of the NordNav receiver front end is plotted versus time in the position where a relative strong signal interference is received by the receiver antenna.. From the AGC level (Figure 3-12) and the AGC characterization, it is clear that the interference received is strong. This interference actually eventually stopped the normal operation of the NordNav software receiver and no GPS satellite signal was acquired or tracked in those locations at the time when interference was present. -6 Receiver AGC in Artarmon-Sydney AGC level Time (sec) x 10 5 Figure 3-12 AGC level in the strongly interfered area versus time 3.4 Summary An interference generated by a TV transmitter tower within the bandwidth of GPS (L1) is reported in this work. The location is Artarmon in Sydney, Australia. This interference in some points was strong enough to saturate the AGC completely and stop the software receiver from tracking the satellites. The effects of this interference on AGC were analysed. In the following chapter, a statistical inference technique is used to detect this interference source. The future plan is to repeat this measurement in other areas and to generate a radio map from Sydney in terms of radio frequency interference within the GPS bandwidth. 64

84 Chapter 4 A STATISTICAL INFERENCE TECHNIQUE FOR GNSS INTERFERENCE DETECTION 4.1 Introduction Marti (2004) has shown that the use of long observation windows allows detection of low power RFI. However, while acquiring this extra information makes the decision easier and the detection confidence greater, the cost is an increase in the processing load. The question is; if given a fixed data window size, is there an optimal way of making the decision about the presence of interference using the proposed inference technique in Marti (2004). This chapter first shows that based on how the data block is broken into sub-blocks, the detection threshold changes with a fixed probability of false alarm. It is then shown that there is an optimal way to break up this data window to achieve the lowest detection threshold for a fixed value of probability of false alarm and such that the computational load is also minimal. An optimal detector algorithm is then proposed. At this optimal point, for a fixed Detection Threshold (DT), the probability of false alarm becomes minimal; or for a fixed probability of false alarm, we achieve the minimum value for the detection threshold. Experiments show that at this point computational load is minimized. The problem of interference detection is mathematically described in section 4.2. In section 4.3 it is shown theoretically that for each window size there is an optimum value for the data block size for which the minimum detectable power interference can be achieved. This result is then supported in section 4.4 by experiments. In section 4.5 these techniques are applied to real interfered data collected in Sydney, Australia and the results are discussed. The chapter concludes with section 4.6. The majority of this work has been published or submitted for publication in Tabatabaei (2006c and 2006e). 4.2 Problem Definition 65

85 Chapter 4 A statistical inference technique for GNSS interference detection In this section a technique of pre-correlation CW interference detection is described. The idea is to detect the CW interference in the GPS signal received by a wideband RF front-end by investigating the power spectral density fluctuation in the frequency bins under investigation (Marti, 2003). The GPS signal is approximately 20 db weaker than the thermal noise power in a 2MHz bandwidth receiver. If the environmental noise is n(k) and the GPS signal 2 2 is s(k) we have: [ s( k) ] E[ n( k) ] E <<. Therefore, it is justifiable to approximate s(k) to be a real, band pass, zero mean, WSS Gaussian process (Therrien, 1992) and the time samples are assumed to be sufficiently independent from each other because of the wide bandwidth of the receiver. The interference under study is considered to be a single tone at the band center frequency. Within the class of non-gaussian signals, a large group is formed by the constant amplitude (CA) signals (Marti, 2003). So the interference i(k) = 2J sin(2π fink + θin( k)) where J is the RFI power, f in and θin are the RFI frequency and phase. If s(k) is the received signal at the GPS antenna (buried under the background noise) when there is no interference, then the overall received signal at the GPS antenna in the presence of RFI will be: x ( k) = s( k) + i( k). The detection test is done based on a two sample T-test. The first population is from a GPS signal that is known to be interference free (assessment window). The second population is taken from the period in which the test is going to be performed (evaluation window). Each window is then divided into M data blocks (DB). Depending on the size of the window (WS, in samples), each DB has N samples where: N = WS/M. In other words, M is the number of statistical data samples and N is the number of frequency bins to be investigated, once an FFT is applied to the DB. Figure 4-1 Process of generating the samples for each frequency bin of the two populations 66

86 Chapter 4 A statistical inference technique for GNSS interference detection The window size (WS), sample number (M) and the data block size (N) are obviously related to the sampling rate of the front-end. For example if the sampling frequency is 16 MHz, for a window size of 300ms, 3 6 WS = 300*10 *16*10 = Also for a data block of size DB=1ms there will be (M=300/1=300) statistical samples for each frequency bin and the number of frequency bins to be investigated will be (N=WS/M=16000), therefore the size of each frequency bin will be ( 16*10 6 /16000 = 1000 ) or 1 khz. In this analysis each frequency bin is examined independent of the adjacent bins. In the rest of the chapter, the i th data block will be denoted by DB i and the j th frequency bin by f j. Also the data bock size (N) will be denoted by DB and the number of the samples (M) by SS. The spectral power density in f j of the assessment window should have the same mean over all of the data blocks of this window as the corresponding frequency bin of the evaluation window (in the absence of interference). These two populations are identically independent. The only assumption which is not theoretically met in this statistical inference problem is normality of the populations; however due to the large number of samples, the Central Limit Theorem (CLT) gives a good approximation to normality and we can still use the test procedure explained in section 2.9 of Chapter 2. As explained in that section, the null hypothesis in our test is that the two windows have the same sample mean. Since the true means of the windows are equal, then the test statistic for the test will be: ˆ μ ˆ assess μeval t = (4-1) ˆ σ 2 2 ˆ assess + σ eval where μˆ represents the sample mean andσˆ represents the sample variance of each population. In Marti (2004), data block size (DB) is considered constant and the effects of varying WS are analyzed. In the next section, it is theoretically shown that SS and DB both affect the minimum detectable power interference with a fixed significance level. The characterization of the effects of different WS, SS and DB on the minimum detectable interference is also investigated. Section 4.4 reports experiments supporting these results. 67

87 Chapter 4 A statistical inference technique for GNSS interference detection 4.3 Mean and Variance of Test Statistic The process of applying statistical inference to the signal is shown in Figure 4-1. At the end of process, we will have M samples for each frequency bin f j (SS=M). The number of frequency bins is N where N=WS/M. The experiments presented later use the NordNav R30 software receiver front-end which has sampling frequency MHz so for example for a data block size of 2ms, there are N=DB=8183 bins of width 2 KHz and for a window size of 256ms there are SS = M = 128 samples for each bin. In this section, we are going to look into the effects of different DB and SS on the GPS signal statistics in the presence of interference. The value to be examined is X ( f j 2 ). If the sample mean of this random variable is called Y j, the statistical inference is performed on Y j from the two groups of samples (assessment and evaluation). The type of the test here is directional (Devore 1995) as the signal power in a frequency bin for interfered signal can only be larger than that of the noninterfered signal. As mentioned in the last section, the distribution of 2 X ( f j ) is unknown. However, the distribution of Y i for a large enough number of samples can be considered to be normal as in Figure 4-2. In Marti (2004), the difference between this distribution and a normal distribution is explained using the Berry-Essen bound. The two groups are significantly separated to represent the null and alternative hypotheses. As can be interpreted from the figure, the two independent parameters affecting the significance of difference between the two hypotheses are the mean and the variances of the samples in the two populations. These two parameters are affected by DB and SS in different ways. 68

88 Chapter 4 A statistical inference technique for GNSS interference detection Detection Threshold 0.1 PDF Null Var 0 Alternative 0.04 Var A Statistical Parameter under Decision Figure 4-2 Parameters affecting the significance of difference between the two samples Mean The relation between DB and the mean of Y i as a random variable is analyzed here. Each data block is assumed to contain N time samples. So the FFT for DB i is an N-point FFT: S( f j N = ) k = 1 0 s ( k) e i i2πjk N so S ( f i j ) 2 = N 1 k = 0 s 2 i ( k) + N 1 p= 0 N 1 q= 0 i2( p q) π si ( p) si ( q) cos( ) N (4-2) The received signal samples are assumed to be sufficiently independent. As was mentioned in the introduction and is explicitly investigated in Marti (2003), the limited front end bandwidth causes correlation between the samples but the assumption of a wide bandwidth front end will allow us to assume that s(p) and s(q) 2 are independent of each other. Consequently, we have E( Si ( f j ) ) = N( E( s ( k))). This shows that the mean is proportional to the number of FFT points. Hence if DB is doubled, the mean of the signal will double. However in the case of interference, this does not hold. For the interference i(k) described in the last section, we know (Proakis, 1996) that the amplitude of the spectral line of such a signal is proportional to N (this is true if the FFT is taken over an integer number of periods of the signal). If this condition is not met, it is due to spectral leakage which is an effect of 2 69

89 Chapter 4 A statistical inference technique for GNSS interference detection windowing. In other words if f j is the frequency of interference, we 2 ) 2 have E( I( f j ) N. This means that if DB is doubled, the E ( I( f ) ) will be multiplied by 4. So, as far as the mean power of the signal in a frequency bin is concerned, using a larger DB will let us distinguish the interference more significantly than when we use a smaller DB. Using a larger value for SS on the other hand, will not affect the mean. j Variance The other parameter affecting the significance of the test is the variance of the parameter on which the statistical inference is done. The purpose of this section is to investigate the effect of DB and SS on the variance of the test statistic. For this purpose the following two questions are addressed here: 1) What will happen to the variance of Y n if SS doubles? 2) What will happen to the variance of Y n if DB doubles? These questions can be considered both for the first sample group (assessment window) and for the second one (evaluation window). To approach the first question it is sufficient to review the central limit theorem (CLT) which says ify then 2 1 Y = 1 M i X k M k = 1 σ = 2 X M σ (where σ 2 X is the variance of all X k ) (Devore 1995). So if SS doubles, the variance is halved. This theorem can also be used to answer the second question. Instead of looking at the variance of Y i we can look at the variance of X k which in this case is var( S ( f ) ), as i i j 2 S i ( f j ) 2 is calculated in the previous section. Noting the fact that the time samples s i (q) and s i (p) are independent, we will have: var( S i i ( f j ) 2 ) N 2 so var( 2 Y i ) N which leads to the answer of the second question. The variance of Y i is multiplied by four if DB doubles. The only remaining point to note is that var( Y + a) = var( Y ), if a is a deterministic value. Based on this argument, as the interference is considered to be a deterministic single tone, the variance of the second population (evaluation window) in the presence of interference at the same frequency bin of the interference, obeys the rules explained above. There is a small chance of a difference between the variances due to spectral leakage, 70

90 Chapter 4 A statistical inference technique for GNSS interference detection (Proakis, 1996) which occurs because of rectangular windowing and results in variations in the amplitude of the received signal spectral line in a frequency bin. Therefore to find the optimum DB for a specific WS, or to find a pair of ( Ha H 0, ) for which a specific probability of false alarm gives us the lowest minimum detectable RFI, we first (as an example) chose the WS to be 256ms. Then the sample mean and variance ( μ 0,σ 0 ) of the data is calculated for the pair (DB = 128*1024 and SS =32). These values are chosen because for SS less than 32, the Berry-Essen bound grows dramatically (Marti, 2004). Then the pairs ( μ 0,σ0 ) and ( μ,σ 1 1 ) are calculated for other pairs of (DB, SS) shown in Figure 4-3 based on the relationships between the mean and variance and the value of DB and SS. Having the pair ( μ 0,σ 0 ), μ 1 is calculated knowing the RFI power. Having all these parameters, p-value for all pairs of ( H 0, Ha ) are calculated using equation (4-3). For this particular example, the p- value have been calculated for the six pairs of (DB, SS) shown in Figure 4-3 (experiments 3&4) and is listed in the last column of the Table It can be seen from that table that the pair (DB=32*1024, SS=128) is the optimum pair. This is supported by two different experiments in the next section. ( x μ ) 1 ( μ1) = exp( 2 μ σ 2 2σ 0 π 0 0 p value ) (4-3) Experiments To verify the results discussed in the previous section, four experiments were designed. The trajectories in which these four experiments are performed are visualized in Figure 4-3. The NordNav R30 software receiver was used to capture the IF data to be analyzed and an HP 8648B was used to generate CW interference which is combined with the GPS signal generated by a SPIRENT GSS6560 by adding the interference to the signal at the center of the receiver band. The environmental noise power is measured by an Anritsu MS2711D to be -110dBm. The NordNav RF frontend is used with -3dB input bandwidth of 3.78 MHz. This bandwidth is assumed to be large enough to consider the induced correlation between the samples to be neglected (Marti, 2003) 71

91 Chapter 4 A statistical inference technique for GNSS interference detection 1024 Sample Size (SS) 1 Exp. 3&4 Exp Exp Data Block Size (DB) (*1024) 256 Figure 4-3 The points in which each of the four experiments are done In the first experiment (DB fixed), for a specific value of DB (16*1024), SS is changed from 1024 to 32 and the minimum detectable power interference for a fixed significant level of test is measured for each SS. The p-value associated with the t- statistic in the experiments for a specific frequency bin to be identified as interference is chosen to be (detection threshold). This value is used for the first three experiments. DB and SS are equal for the assessment and evaluation windows. The six points at which this experiment is performed are shown in yellow stars in Figure 4-3. In Table 4-1 the minimum detectable RFI is recorded with and without spectral leakage. Spectral leakage causes energy (amplitude) from distinct spectral features to "leak" into adjacent frequency bins, giving rise to spurious components in the frequency spectrum of the signal. The only way to avoid such leakage entirely would be to arrange that all the frequency components of the signal being examined coincide exactly with frequency bins in the computed spectrum. This can be achieved if there is an integer number of periods of the sinusoid within the window which can be obtained by choosing the right window and data block size. Given that we are unlikely to know the exact frequency of the interference we are dealing with, this is an unrealistic expectation. Alternatively the maximum leakage happens when there is an integer number of periods plus a half period of signal within the window. This situation is equally as unlikely to occur so in real situations the result will be somewhere in between these extreme. It can be seen from Table 4-1 that the minimum 72

92 Chapter 4 A statistical inference technique for GNSS interference detection detectable interference decreases with the increase in SS. As the sample size increases, we expect the variance of these means to decrease. This can be observed from Table 4-2. The mean of the population in the frequency bin where the interference exists is fixed for the assessment window (see Table 4-2) while the computational load changes exponentially (Figure 4-4). In the following three tables, the columns correspond to the points of experiment 1 of Figure 4-3. Pair (DB=16*1024, SS=32) correspond to the left most column. Table 4-1 Minimum detectable interference for experiment one Minimum Detectable RFI (dbm) for DB = 16*1024 Without Spectral Leakage With Spectral Leakage SS Table 4-2 Mean of samples for experiment one Mean*1e05 for DB = 16*1024 Assessment Window Evaluation Window SS Table 4-3 Standard deviation of samples for experiment one Standard Deviation*1e04 for DB = 16*1024 Assessment Window Evaluation Window SS

93 Chapter 4 A statistical inference technique for GNSS interference detection Exp-1-Computational-Load (second) Sample Size(SS) Figure 4-4 Computational load for experiment one In the second experiment, for a specific value of SS (64), DB is changed from 4*1024 to 256*1024 and the minimum detectable power interference for a fixed significance level of test is measured for each DB and is shown in Table 4-4. The fact that the mean of the population changes by changing DB can be seen in Table 4-5. In fact the mean of both populations increases with increasing DB. In section 4, the rate of this increase has been calculated for both the assessment and the evaluation data in the presence of interference. It is because of this difference in the rates of increase that the minimum detectable interference decreases (Table 4-4). Population standard deviation and the test computational load are shown in Table 4-6 and Figure 4-5. In the following three tables, the columns correspond to the points of experiment 2 of Figure 4-3. Pair (DB=8*1024, SS=64) correspond to the left most column. Table 4-4 Minimum detectable interference for experiment two Minimum Detectable RFI (dbm) for SS = 64 Without Spectral Leakage With Spectral Leakage DB*

94 Chapter 4 A statistical inference technique for GNSS interference detection Table 4-5 Mean of samples for experiment two Mean*1e05 for SS = 64 Assessment Window Evaluation Window DB* Table 4-6 Standard deviation of samples for experiment two Standard Deviation*1e04 for SS = 64 Assessment Window Evaluation Window DB* Exp-2-Computational-Load (second) Data Block Size (DB) Figure 4-5 Computational load for experiment two In the third experiment, both SS and DB are changing but WS is kept constant. The idea that there is an optimum point for each specific WS (256ms in this example) is shown by this experiment. In Table 4-7 the fourth experiment point which corresponds to the pair (32*1024,128) for (DB, SS) is the optimum point for detecting minimum power interference. In Table 4-8, Table 4-9 and Figure 4-6 illustrates the means and standard deviations of the populations and also the computational load of 75

95 Chapter 4 A statistical inference technique for GNSS interference detection the tests. In the following three tables, the columns correspond to the points of experiment 3 of Figure 4-3. Table 4-7 Minimum detectable interference for experiment three Minimum Detectable RFI (dbm) Without Spectral Leakage With Spectral Leakage SS DB* Table 4-8 Mean of samples for experiment three Mean*1e05 Assessment Window Evaluation Window SS DB* Table 4-9 Standard deviation of samples for experiment three Standard Deviation*1e04 Assessment Window Evaluation Window SS DB* ` 76

96 Chapter 4 A statistical inference technique for GNSS interference detection Exp-3-Computational-Load log2(db/2) or -log2(ss/1024) Figure 4-6 Computational load for experiment three In the final experiment, unlike in experiment three, the power of interference is kept constant and the p-value (probability of false alarm) is calculated for a window size of 256 ms. In Table 4-10 each row corresponds to a point in Figure 4-3 (experiment four). Also based on the theoretical analysis, the p-value is calculated for all 6 points and the results are listed in Table The parameters in this table have been measured by experiment and we can see that for example μ 0 at point 2 is measured to be exactly twice the μ 0 of point 1. Also σ 0 at point 2 is 2 2 times that of point 1 which is consistent with section 4.3. The RFI power is considered to be fixed in this experiment for all points. Knowing this power we can calculate μ 1 at all points. With all these parameters, and the definition of p-value which is the probability of observing anything more extreme than what has been observed, the p- value for all points is calculated using equation (3-5). In the sixth column of Table 4-10 the actual p-value of the observation result of this experiment is shown. The ttest(x,y) command in MATLAB performs a paired t-test of the hypothesis that two matched (or paired) samples in the vectors x and y come from distributions with equal means. The difference x-y is assumed to come from a normal distribution with unknown variance and x and y must have the same length. In this experiment, for each frequency bin, the number of measurements for both assessment and evaluation windows is equal to the sample size (SS). The results of Table 4-10 relate only to the frequency bin in which the CW interference exists. 77

97 Chapter 4 A statistical inference technique for GNSS interference detection Table 4-10 Simulation and the theoretically calculated p values Point (Exp 4) σ 0 μ 0 σ 1 μ 1 Log10(pvalue)/ Exp. Log10(p-value) /theory 1 1.4e3 4.5e4 1.7e3 4.9e e3 9.0e4 5.2e3 1.3e e4 1.8e5 1.8e4 3.4e e4 3.6e5 6.3e4 9.8e e4 7.2e5 2.4e5 3.1e e5 1.4e6 8.1e5 8.4e Discussion It is clear from Table 4-10 that there is an optimum point (point number 4 in the both experiment and theory column for the specific window size) that gives us the minimum p-value. At this point we can detect the minimum detectable power interference using a 256ms window. This result is also consistent with what we saw in experiment three. These two experiments show that only at this optimal point, i) for a fixed probability of false alarm, we can achieve the minimum value for the detection threshold or minimum detectable RFI (experiment three) and ii) for a fixed Detection Threshold (DT), the probability of false alarm becomes minimal (experiment four). The experiments also show that at this point we have minimized the computational load. In Figure 4-7, the minimum detectable interference for the 256ms window is investigated in terms of computational load for the six points denoted in Figure 4-3. This figure characterizes the trade off between processing load and the minimum detectable interference. It indicates that before point 4 (which is shown to be the optimum point in terms of minimum detectable interference), lower power RFI can be detected at the cost of increased computation effort whereas after this point even though we are spending more in terms of processing, nothing is gained in terms of lower minimum detectable RFI. 78

98 Chapter 4 A statistical inference technique for GNSS interference detection -127 Exp-3-Minimum Detectable RFI (dbm) Trade off area Computational Load (second) Figure 4-7 Trade off zone where the lower power detectable interference can be achieved with the price of processing load. 4.5 Real RFI Detection The RFI Source As was mentioned in the previous sections, the interference detection technique introduced in this chapter is sensitive enough to detect CW interference when its power is lower than the power of environmental noise. In situations where the AGC is significantly affected by the power of interference, this will affect the statistical behavior of the signal in each frequency bin. This will consequently prevent the algorithm working properly. It is clear from Figure 3-7 that in these cases the RFI can be detected simply by analyzing the signal spectrum or by observing the AGC. Figure 4-8 shows the log 10 of p-value in the presence of interference (of Figure 3-7). It ia clear that no decision can be made based on this figure regarding the existence of RFI and its frequency. 79

99 Chapter 4 A statistical inference technique for GNSS interference detection 0 DB = 64*1024, SS = 64-5 log10(p-value) Frequency (250 Hz) Figure 4-8 p-value across the bandwidth in the presence of strong interference in a specific time Discussion (Real Interference) Based on the discussion in the previous section, data was collected at a point where interference is weak and therefore does not affect the AGC level. The histogram of the data for this case has a Gaussian shape. The power spectrum of this data is shown in Figure 4-9. From this power spectrum, no distinguishable anomaly is recognizable as interference. -55 time = 63 s mag [dbw] freq [MHz] Figure 4-9 Power Spectral Density of the Received Signal at time... 80

100 Chapter 4 A statistical inference technique for GNSS interference detection The result of applying the RFI detection technique to this data is shown in Figure In this figure, the log10 of p-value is shown across the 2MHz bandwidth for almost 40 seconds. Figure 4-10 log10 (p-value) versus time and frequency Each step of the time axis corresponds to 62.5 ms (the window used in the hypothesis test). The data block size is chosen to be 2*1024 samples of data resulting in each step of the frequency axis corresponding to 8 khz. This figure shows that there are two frequencies where CW interference is present within the 2 MHz bandwidth. There are moments like point 370 (23 sec) at which no interference is recognized from this figure. In fact interference is present at those moments but is not strong enough to be detected by a 62.5msec window size. Figure 4-11 shows that when using a 256 ms window size this interference is recognizable (the p-value for the frequency bin in which the interference exist goes below -10 which had been chosen to be the detection threshold). With a longer window size we can have both a larger SS (sample size) and a larger DB (data block) size. 81

101 Chapter 4 A statistical inference technique for GNSS interference detection 0 DB = 64*1024 SS = log10(p-value) Frequency (250 Hz) Figure 4-11 P-value across the L1 bandwidth for time 23 sec When a window size of 62.5 ms is used, based on the results of section 4.3, there is an optimal way of breaking this window into data blocks. Using the same approach as in Table 4-10, the optimum point for this window size is found to be 32*1024 samples per each data block. The sample size (SS) is 32 in this case. This is also shown practically using part of the real interference-affected data from time step 300 to 410 (from Figure 4-10). Figure 4-12, Figure 4-13 and Figure 4-14 are all the - log10 of p-value of the frequency bin in which the interference is present for this time period. DB = 128*1024, SS = log10(p-value) Time (62.5ms) Figure 4-12 p-value across time in frequency MHz (for DB = 128*1024 and SS = 8) 82

102 Chapter 4 A statistical inference technique for GNSS interference detection The difference between these figures is the data block size chosen for each case. As was discussed earlier, the computational load increases with increasing data block sizes. In terms of detection performance, if 10-8 is chosen as the detection threshold, the first experiment (Figure 4-12) indicates that no interference exists. The third experiment (Figure 4-14) despite using much more computational effort, doesn t reliably distinguish between the time when the interference exists (steps 20 to 60) and the time when interference has significantly lower power, as can be detected in (Figure 4-13) where the optimum value is used. 18 DB = 32*1024, SS = log10(p-value) Time (62.5 ms) Figure 4-13 p-value across time in frequency MHz (for DB=32*1024 and SS=32) 83

103 Chapter 4 A statistical inference technique for GNSS interference detection 40 DB = 1024, SS = log10(p-value) Time (62.5ms) Figure 4-14 p-value across time in frequency MHz (for DB = 1024 and SS = 1024) The implementation of the detecting algorithms and techniques developed in this chapter is done by UNSW hardware engineers. My role in this part was only an advisory role. At the heart of the design is an Altera FFT block, providing a 2048 bin, complex FFT with 8 bits of input precision. Input to this FFT block can come from two sources; direct from the incoming raw GPS IF data stream, or from a local oscillator mixed and accumulated version of the raw data. The second source provides the zoom functionality for determining the interference frequency to greater resolution. Figure 4-15 shows a simplified block diagram of the system. The RF front-end amplifies, filters, down-converts and band-pass samples the incoming signal. It passes the sampled intermediate frequency (IF) to the FPGA for digital processing as two-bit, sign and magnitude values. The complex output of the FFT is scaled and processed into a magnitude value, and then transferred to on-chip memory using an Altera direct memory access (DMA) block function. The data is then available for the Nios processor to access. At this point the software algorithms take over. An overview of the software and short summary of the results are provided in Appendix A. 84

104 Chapter 4 A statistical inference technique for GNSS interference detection Figure RF down-conversion and band-pass sampling diagram 4.6 Summary and Future Work This work presents a processing scheme using an existing GPS pre-correlation interference detection technique. In this technique, two windows of data are chosen; one from clean data and one from where we seek to detect the interference. Signal statistics in the two windows help us detect the presence of CW interference. Each window is divided into a number of data blocks whose size is denoted DB and the number of these data blocks within a window is referred to as sample size or SS (Window Size = DB*SS). We show that for each window size, there is an optimum solution for DB and SS that allows us to detect the minimum power of CW interference. There are two independent parameters which affect the significance of the difference between the two groups of data in the frequency bin where interference exists: mean and variance. It is theoretically shown that a larger DB means more significant difference in mean and for constant WS this means a smaller SS. A small SS on the other hand means a large variance which is equivalent to a less significant difference between the two groups of data. So there is a solution for each WS (DB, SS), in which we can achieve the most significant difference between the clean set of data and the interfered data. Also for all of the different choices of SS, DB and WS, computational load of the processor is measured. We present a procedure to determine the best operational point in terms of breaking the data samples into blocks for the interference detection algorithm based on the priority of processing load and the minimum detectable RFI. With this technique we were able to detect interference with power less than the background noise power. In this work detection of CW interference is studied. In the case of swept CW interference, averaging over a longer time does not lead to better results as the 85

105 Chapter 4 A statistical inference technique for GNSS interference detection frequency is changing. Other statistical approaches are required to detect lower power interference in such a case. Also modulated signals are not truly periodic so Fourier analysis may also not be applied directly. However; modulated signals have built-in periodic signals that can be extracted and analyzed using Fourier analysis. 86

106 Chapter 5 CHARACTERIZATION OF THE EFFECTS OF CW AND PULSE CW INTERFERENCE ON THE GPS SIGNAL QUALITY 5.1 Introduction Our knowledge of the effects that interference can have on the operation of different parts of the GNSS receiver can be useful in eliminating those effects. It is always possible to try and eliminate the interference from the received signal but this blind elimination can also cause loss of useful information. Pre-detection interference mitigation schemes which act directly on the received IF signal are among those techniques. In this chapter it is shown that even a strong CW RFI can be harmless on particular occasions for the operation of the receiver. In Kaplan, (2005) and (Parkinson, 1996a), the effect of CW RFI on the received signal quality in terms of effective carrier to noise ratio ( ( C / No) ) is analyzed. In Betz (2001) a closed eff formula for ( C / No) is derived. In that formula the assumption is that the bandwidth eff of the narrowband interference is wider than the reciprocal of the integration time in the correlator and therefore the effect of the Doppler frequency on the ( C / No) is not seen. In the investigation in this chapter, interference with a very narrow bandwidth (narrower than the reciprocal of the integration duration time of the correlator) and the GNSS signal structure studied in Chapter 2, a closed formula is derived for the ( C / No). This formula is tested and verified by calculating the C/No eff using the I and Q data from a software GPS receiver. For the pulse CW, a similar analysis is performed to characterize the effect of parameters such as pulse repetition period (PRP) and duty cycle on the received signal quality. In section 5.2, a mathematical expression for C/No is introduced based on the spectral analysis of both the signal and the RFI in passing through the correlator. The actual C/No is calculated using received I and Q data from a software GPS receiver and a characterization method is discussed in section 5.3. The same procedure is presented in section 5.5 for pulse CW interference. The chapter concludes with section 5.6. eff 87

107 Chapter 5 Characterization of the effects of CW and pulse CW interference on the GPS signal quality The majority of this work has been published or submitted for publication in Tabatabaei (2006a, 2005 and 2007a) and Motella (2007). 5.2 Analysis of the C/No after the Tracking Loop In this section, an expression for the post-correlation carrier to noise density (C/No) is derived. A high C/No implies that the code-tracking loop is operating on or near the correlation peak. We intend to show that this peak is affected by two factors: i) the non-coherency between the received code/ carrier phase and those generated locally in the receiver, and ii) the frequency difference between the continuous wave interference and the carrier. Figure 5-1 shows the concatenation of the two tracking loops (DLL for the code tracking and PLL for the carrier tracking) and the corresponding carrier and code wipe-off systems. At the PLL side the code is first stripped off by multiplying the incoming signal by the prompt code generated by the DLL. The resulting CW signal is passed through a dump integrator which is represented by the low pass filters in the diagram. In this way the carrier is tracked by the local oscillator, which is used to wipe off the carrier at the DLL input. At this point, taking into account the signal flow through the system of figure 5, it is possible to evaluate the value of the correlation peak, which is a key element in the carrier to noise expression. Figure 5-1 Correlator (Code and Carrier Tracking Loops) 88

108 Chapter 5 Characterization of the effects of CW and pulse CW interference on the GPS signal quality The received GPS CA code signal in the presence of interference can be modeled as (Kaplan,2005), S(t) = 2 P D(t)P (t) cos( 2πf t + θ (t)) + n (t) I(t) (5-1) s N c c T + where P s is the received signal power; D(t) is the data modulation at 50bit/s; f c is the carrier frequency; θ (t) is the carrier phase; P N (t) is the pseudorandom code c modulation; n T (t) is the thermal noise represented as a realization of a white and Gaussian Random Process and I(t) is the in band CW interference defined as I(t) = 2J cos(2π f t + θ ( t)) (5-2) i i where J is the interference power and f i and θ i are its carrier frequency and phase. The front end bandwidth is considered wide enough to pass essentially all of the signal and interference power. The thermal noise has a power spectral density N 0. We are now interested in evaluating the carrier-to-noise ratio, by examining the operations performed by the correlator inside the Code Tracking loop. It is known that this correlation requires a reference code (Figure 5-1) generated by the local code generator, which can be written as R PN (t) = P (t-τ) N (5-3) where τ is the code phase difference. The reference carrier (Figure 5-1), generated by the carrier tracking loop, can be written as R CAR (t) = cos ( 2πfˆ t + θˆ ) (5-4) c c 89

109 Chapter 5 Characterization of the effects of CW and pulse CW interference on the GPS signal quality where fˆ c and θˆ c are the receiver estimates of carrier frequency and phase. In most applications this reference carrier and the carrier of the signal are not considered to be coherent. After multiplying the signal by the reference code and carrier in the correlation process, the result is passed through an integrator with a duration of T d, which is often chosen to be the navigation data duration i.e. 1ms.The following analysis is to evaluate the signal, interference and noise after this process. To make the computation easier, we consider the signal component before the integration to be ~ a complex signal S ( t) = I ( t) jq ( t), where I s (t) and Q s (t) are the in-phase and s + s quadrature phase components due to the non-coherent carrier wipe off. At the point before the integration we have W (t) s = j( 2πΔfct+ Δθc ) 2 Ps PN (t)pn (t-τ)e (5-5) where Δ f c and Δθc are frequency and phase error. In the following equations because we are evaluating the absolute values, without loss of generality we assume that Δ θc is equal to zero. If we consider C i as the Fourier series coefficients of the periodic P N (t) function, we can write P ( t) = N i= C e i i j2π t Tc where T c is the code period, and W S (t) becomes W (t) S = 2 P P (t)p (t-τ)e s N N j( 2πΔfct) = l = i = C C i l ( i+ l ) l j2π [ +Δf c ] t j2π τ T T c c e e (5-6) Defining m=i+l and m l = m l l τ j2πl T c R ( τ ) = C C e, we have 90

110 Chapter 5 Characterization of the effects of CW and pulse CW interference on the GPS signal quality 91 = +Δ = m t f T m j m s s c c R P W (t) π τ 2 )e ( 2 (5-7) As shown in Figure 5-1, this signal is then passed through an integration over the period T d, giving + = + Δ Δ = ] [ 2 2 ) ( 1 )e ( ) ( 1 d d d d c c T T m T T t f T m j m d t f j N N d dt e R T dt t P t P T π π τ τ (5-8) Considering the equation = Δ ] [ 2 1 d d c T T t f T m j d dt e T π = + Δ + Δ } 2 ] [ 2sin{2 ] [ d c c c d T f T m f T m T π π } ] {[ Sinc d c c T f T m Δ + and substituting in (5-8), it is possible to write the integral of ) (t W s as = + + Δ = m d c c m s d T T s T f T m R P T W (t) dt d d )Sinc ( 2 2 / 2 / τ (5-9) Now if we assume that c c T f 1 << Δ then Δ dt (t-τt (t)p P P d d c T T t f j N N s 2 2 ) (2 e 2 π ) )Sinc( ( 2 0 d c d s T f R T P Δ τ (5-10)

111 Chapter 5 where R ( τ = 0 ) + l= C * l C e l l j2π τ Tc Characterization of the effects of CW and pulse CW interference on the GPS signal quality In the next stage, the contribution of the interference I (t) in the correlator output is analyzed. The interference is first multiplied by the reference carrier, and then is spread in the first correlation stage, where it is multiplied by the receiver generated code. The resulting signal is W (t) = I(t)P (t-τ)e I N j(2πfˆ ct) = + ( n+ l ) j2π [ +Δfi ] t Tc l j2π τ Tc J n C l e e (5-11) In this equation the assumption is that the interference frequency f i is n khz away from L1 plus Doppler frequency plus a residual term f ˆ + / c fi = Δfi n Tc, and Δ f i = f i - f c ˆ. After the integration we will have Δ fi, that is + Td / 2 Td / 2 W I(t) dt = T d l= n + l J nclsinc{[ + Δfi ] Td }e T c l j2π τ By definition, the residual term n + l Sinc [ + Δf Tc and otherwise n + l Sinc [ + Δf Tc i i ] T d ] T Therefore we have d 0 Sinc( Δf T i d ) T c Δ fi is lower than 1 / 2T then for n l c 92

112 Chapter 5 Characterization of the effects of CW and pulse CW interference on the GPS signal quality + Td / 2 n j2π τ * W T I ( t) dt = Td J ncnsinc( ΔfiTd )e (5-12) Td / 2 The effective C/No in the presence of interference may be expressed as (Parkinson, 1996a): Cs Ls C/No = (5-13) L N + I n 0 where C s is the received power of the signal; L s is the processing gain for the desired signal; N 0 is the thermal noise power spectral density; L n is the processing gain for the noise and I is the effective interference. Substituting (5-10) and (5-12) into (5-13) yields the expression for effective C/No: C/No ( 2P T R (τ)sinc( Δf T )) 2 s d 0 c d = (5-14) * 2 Ln N 0 + ( J ntdcnsinc( Td Δfi )) Previously in Chapter 2, an expression relating to C/No was introduced (Kaplan, 2005). The difference is that in (5-14) the interference bandwidth is assumed to be much smaller than the reciprocal of the integration period of the tracking loop. Therefore unlike Kaplan (2005) and Betz (2001), the effect of Doppler frequency and the integration duration time is considered in (5-14). In Figure 5-2, as an example, using (5-14) and assuming a specific environmental noise power, the C/No is shown for satellite 1 with Doppler frequency changing from 0 KHz to 10 KHz and CW interference at 14 khz away from the band center at GHz (i.e. at GHz). The width of the lines depend on the integration duration time of the receiver which in this example is 8ms. This is analyzed in more detail in the next section. The deep troughs in this graph correspond to the coincidence of CW RFI with the code spectral lines. This occurs at 1 khz spacing in the Doppler. As expected and explained in Chapter 2, there are different values for different lines, resulting in an apparently random variation in the depths of the troughs. This is due to the difference between the coefficients of different lines in the code spectrum. This 93

113 Chapter 5 Characterization of the effects of CW and pulse CW interference on the GPS signal quality particular line spectrum would be different for a different satellite code. Another point which is noticeable from this figure relates to the width of the troughs. The width of each trough is related to the integration period, as can be seen from (5-14). The longer the integration period is, the narrower the troughs will be. This is shown is section C / No (db-hz) Doppler frequency (Hz) Figure 5-2 C/No calculated using the mathematical expression (5-14) for satellite 1 with Doppler frequency changing from 0 khz to 10 khz and CW interference at 14 khz away from the band center at GHz. 5.3 Experiments and Discussions There are different techniques for estimating the carrier power to noise density, C/N 0. This estimate is important because it helps determine whether the code and carrier tracking loops are in lock, controlling the response of the receiver to low signal to noise environments, and determining the signal to noise in order to assess or predict receiver performance. Groves (2005) contains a review of different measurement techniques such as the narrow-to-wideband power ratio method, correlator comparison method, and non-normalized discriminator output statistics method. It was found in Groves (2005) that the power ratio method performs better than the others in terms of noise in low signal-to-noise environments. This widely known method is presented in Parkinson (1996a). The prompt I and Q samples, over the accumulation interval τ a are divided into M intervals. These samples are then used to calculate a narrowband power, P n over the whole accumulation interval and a 94

114 Chapter 5 Characterization of the effects of CW and pulse CW interference on the GPS signal quality wideband power, P W, over the intervalτ a / M estimates are, then summed overτ a. These power M M 2 2 P n = ( I Pi ) + ( QPi ) (5-15) i i M 2 2 P = ( I + Q ) W i Pi Pi (5-16) where I pi and Q pi are the observable I and Q components of the prompt, evaluated M times in the interval τ a. where The narrow-to-wide power ratio, P N/W, is simply the ratio of the two power measurements. However, to reduce the noise, the measurement is averaged over k blocks of I and Q samples. Thus, P N / M = 1 k k P P N, r r= 1 W, r (5-17) In (Parkinson, 1996a), it is shown that from the above equations it is possible to derive (5,18) E( P N / W ) M ( C / N τ + 1) M + C / N τ 0 a (5-18) 0 a where E() is the expectation operator. Rearranging this gives the measured carrier power-to-noise density as a function of the power ratio measurement: C / N 0 M τ a PN / W 1 M P N / W (5-19) 95

115 Chapter 5 Characterization of the effects of CW and pulse CW interference on the GPS signal quality In Parkinson (1996a) it is shown graphically that for C/No>23dB there is less than 1dB estimation error for an average time of 1 s (M=20 and k=50). For smaller values of C/No a longer averaging time is required. In Figure 5-3, the hardware setup to measure the actual C/No is shown. The NordNav software receiver is used to capture the IF data to be analyzed and post processed and an HP8648B is used to generate the CW interference which is combined with the GPS signal generated by a SPIRENT GSS6560. In Figure 5-4, the actual measured C/No using NordNav software GPS receiver is illustrated. Figure 5-3 Hardware setup for the experiments from left: Spectrum analyzer, RF signal generator, NordNav front-end, GPS signal generator C/No (dbh) Doppler Frequency (Hz) Figure 5-4 C/No calculated using the power ratio technique for satellite 1 with Doppler frequency changing linearly with time from 0 to 9 khz and CW interference at 14 khz away from the band center at GHz 96

116 Chapter 5 Characterization of the effects of CW and pulse CW interference on the GPS signal quality There are a number of points regarding this figure which are addressed as follows. At the beginning, the graph presents a large value (50 db), which is just an initial condition and gradually converges to the real actual value of C/No. Figure 5-5 and Figure 5-6 are focused on one of the troughs where RFI coincides with a code spectral line in the actual measurement and the theoretical calculation of C/No corresponding to that line. A small peak in the actual estimation right in the middle of the trough is noticeable. Around this peak, as interference frequency is very close to the carrier frequency, it is actually helping the C/No instead of degrading it. However, this phenomenon can not be helpful in our proposed detection-characterization method. By choosing appropriate M and K in the power ratio estimation of C/No (5-18), this peak can be averaged out. The only disadvantage of choosing K and M such that the peak is averaged out is that some of the information in C/No is also lost. This information fortunately is not essential to the proposed RFI characterization method here as only the peak information of the lines are used to work out the frequency of the CW interference Actual C/No (dbh) Time (second) Figure 5-5 One of the troughs where RFI coincides with a code spectral line in the actual measurement 97

117 Chapter 5 Characterization of the effects of CW and pulse CW interference on the GPS signal quality C/No (dbh) Doppler Frequency (Hz) Figure 5-6 One of the troughs where RFI coincides with a code spectral line in the theoretical measurement The results from section 2 and this section are compared here to extract the spectral information of the CW RFI. As was explained the peak information is used. The fact that each C/A code has a unique line spectrum is used in this comparison. In Figure 5-7, the uniqueness of this pattern is shown. The Doppler frequency of satellite 1 is changed from -4 khz to 4 khz and the CW RFI is placed at the band center. The symmetry in the picture clearly shows that the coincidence of interference with a line always results in the same pattern of C/No. This is because the line spectrum is symmetrical. 98

118 Chapter 5 Characterization of the effects of CW and pulse CW interference on the GPS signal quality Actual C/No (dbh) Time (sec) Figure 5-7 C/No calculated using the power ratio technique for satellite 1 with Doppler frequency changing from -4 khz to 4 khz and CW interference at the band center at GHz In Figure 5-8, C/No plots calculated using both the parametric method and power ratio method applied to measured data, are shown together. The fact that for both techniques, the relationship between the values of any two consecutive peaks remains the same is illustrated. In fact in Figure 5-8, we have Figure 5-2 and Figure 5-4 which is flipped and the horizontal axis converted to Doppler frequency (as Doppler frequency changes linearly with respect to time). The reason why it should be flipped over is that in an increasing Doppler frequency scenario, as interference crosses the spectral lines, the last line which is crossed is in fact the lowest frequency spectral line. 99

119 Chapter 5 Characterization of the effects of CW and pulse CW interference on the GPS signal quality Actual C/No Theroretical C/No 45 C/No (dbh) Doppler Frequency (Hz) Figure 5-8 C/No plots calculated using both the parametric method and power ratio method In Figure 5-9, the difference between the estimated and theoretically calculated C/No is shown. This difference has been calculated in the first 1000 khz from the band center. This is in fact the difference between the 8 trough values of Figure 5-4 and the values of 8 troughs of an 8 trough searching window over the whole bandwidth of the C/A code spectrum. The minimum difference in this experiment is at 14 khz which is the frequency at which the RFI is added to the GPS signal. For clarity, only the first 80 khz is shown. In (5-14), other than the frequency of interference, there are three other parameters: signal power, background noise power and interference power. The first two are known parameters as the GPS signal has been generated in a known condition and with a desired signal power. Interference power is also estimated from the AGC level in the RF front-end (Bastide, 2003). Using this value we can have the theoretical C/No fit the actual C/No as much as possible. This will allow us to have less difference between the two calculations at the frequency where interference exists (14 khz in this experiment). 100

120 Chapter 5 Characterization of the effects of CW and pulse CW interference on the GPS signal quality Difference between values of 8 consecuitive troughs (db) Frequency (khz) Figure 5-9 The difference between the estimated C/No and the theoretically calculated one 5.4 The Effect of Integration Time on C/No in the Presence of CW RFI In Figure 5-10(a, b), using a software receiver (Borre, 2007), C/No is examined for PRN1 signal with a Doppler change of 2.4 khz. The dependence of C/No on the loop updating rate (integrator).and the Doppler frequency is shown in these figures. This result is also consistent with the previous results from (5-14). 101

121 Chapter 5 Characterization of the effects of CW and pulse CW interference on the GPS signal quality C/No (dbhz) Time (1.6 sec) (a) C/No (dbhz) Time (1.6 sec) Figure 5-10 (a,b) C/No of PRN1 when interference is at 4.5 khz for a) T d = 16ms and b) T d = 8ms (b) In these two cases, the level of noise has been intentionally kept low so that the effect of interference is better observed. 5.5 The Effect of Pulse CW RFI on the GPS Signal Quality In this section, the effect of pulse CW interference on the GPS signal quality (in terms of C/No) is studied. The expression of this interfering signal can be written as I( t) = p( t) 2J cos[2π f t + ϑ ( t)] i i 102

122 Chapter 5 Characterization of the effects of CW and pulse CW interference on the GPS signal quality where p(t) is a periodic pulse with period T p. The Fourier transform of the pulse p(t) shifted by ±f i (the carrier frequency of the interference). As discussed in Chapter 2, this interference is spread by the code which is generated by the receiver in the carrier tracking loop. In the next stage it is passed through an integrator block which integrates the signal over a specific period T d. It is shown later in this section that this process is equivalent to convolving the signal in the frequency domain with a Sinc function of the type S I (f) = T d Sinc(fT d ) and evaluating the result for f = 0. Figure 5-11 shows the spectrum of I(t), where D C is the duty cycle of pulses, and T p the pulse repetition period. The Sinc function S I (f) is also shown. The multiplication by the periodic code implies the spectrum repetition indicated in Figure 5-11, where T c is the C/A code period, which is 1 ms. 1/Tp 1/DC*Tp 1/Td 1/Tc Figure 5-11Sinc functions associated with the pulse interference (black) and the integrator block of the tracking loop (red) To evaluate the effect of the pulse CW interference on the correlator output power and eventually on the received GPS signal quality, the same strategy as in section 5.2 is followed. The GPS signal and the interference are passed through the correlator of Figure 5-1 and the correlator output power generated by each of them is calculated. The power generated by the signal is similar to the one calculated in Equation 9. In this section the correlator output power generated by the pulse CW interference is investigated. Now if we call the output signal from the first code despreading stage of Figure 5-12 Z i ( ) then we have 1, t Z1, i ( t) = I(t)PN ( t-τ) e j(2πf ct) = 103

123 Chapter 5 Z j(2πδfit ) 1, i ( t) = p(t)pn ( t-τ) e Characterization of the effects of CW and pulse CW interference on the GPS signal quality (5-20) where Δ fi is the frequency difference between the interference carrier and the n th line of the C/A code such that Δ fi < 1 k Hz. The two diagrams in Figure 5-12 and Figure 5-13 show the same process of code despreading and integration in the time and frequency domains respectively. Signal x(t) represents either the GPS signal or the interference and y(t-τ ) represents the code which is generated by the receiver to strip the code off the received GPS signal. The output in both diagrams is called (I+Q). This quantity is a function of T d and τ among other parameters such as the frequency difference between the signal or interference carrier and the receiver estimation of the carrier ( Δ f c or Δ f i ). x(t) X y( t τ ) + Td 2 Td - 2.dt ( I + Q)( τ, Td ) Figure 5-12 Time domain representation of code despreading and integrating blocks 104

124 Chapter 5 Characterization of the effects of CW and pulse CW interference on the GPS signal quality x(t) f =0 ( d I + Q)( τ, T ) T d Sinc ( ft d ) y( t τ ) Figure 5-13 Frequency domain representation of code despreading and integrating blocks Here, to make the calculations easier, the processing is done in the frequency domain. In passing through the first convolution of the block diagram shown in Figure 5-13 we will have Z 1,i (f) = j2πkτ T k c P(f) *( J ncke δ ( f )) * δ ( f Δfi + T k c n T c ) Z 1,i (f) = j2πkτ T k n c P(f) *( J ncke δ ( f + Δfi )) T T k c c (5-21) where J n, C k, τ and Δ f i are the same previously defined parameters and m P(f) = p δ(f T m m ) p This signal is then passed through the second convolution, (in the frequency domain) shown in Figure (f ) = Z i 105

125 Chapter 5 Characterization of the effects of CW and pulse CW interference on the GPS signal quality m k T J p C Sinc(( f d n m k n k T c + m T p Δf ) T )e i d j kπτ 2 T c and the output of the block diagram of Figure 5-13 will be Z 2i (0) = m k n k Td J n. pm. CkSinc(( + T c m T p Δf ). T ). e i d j2k T c πτ (5-22) To simplify this expression for analysis purposes, a few approximations and assumptions are made. The classification is done based on the different situations that the two Sincs in the Figure 5-11 can have with respect to each other. It is obvious that in general non-extreme situations, the behavior of C/No can be inferred from considering these marginal behaviors. By definition the residual term Δ is lower than1 / 2T because it is the offset fi c of the interference spectral line with respect to the lines of the type the code periodicity. i / Tc created by T c >> T p 1/Tp 1/Td 1/DC.Tp Figure 5-14 Sinc functions associated with the pulse interference (black) and the integrator block of the tracking loop (red) 106

126 Chapter 5 Characterization of the effects of CW and pulse CW interference on the GPS signal quality Figure 5-14 shows the Sinc function associated with the integrator (with frequency interval between zeros equal to1 / T ) and the pulse Sinc functions (with frequency interval between zeros equal to1 / D C T and frequency intervals between the lines equal to d p 1 /Tp where D C is the pulse duty cycle). It is clear from this figure that only one spectral line of the Sinc associated to the pulse can pass through the integrator. In other word we will have i d n n i d j 2n T c πτ Z2 ( 0 ) = T J. p0. C Sinc( Δf T ). e (5-23) This expression is similar to the term in the (5-12). The only difference is that this expression is multiplied by p 0 which is related to the D C value of the periodic pulse. Therefore, a pulse with a low duty cycle will result in a lower effect of the pulse interference having a smaller effect at the output of the correlator when compared to a pulse with equal average power level and a higher duty cycle. The fact that the C/No degradation decreases as the duty cycle decreases may seem intuitive. However the trend of this degradation can be worked out using (5-23). The added value of (5-22) is that it gives a clear relationship between the receiver and pulse parameters, which allows us to establish, for example, a threshold for interference detection and mitigation algorithms which take into account the worst case for the Doppler or it can be adaptive with respect to the Doppler. In Figure 5-15, as an example, the carrier to noise ratio for satellite 1 is drawn using this analyses. In this example the interference is placed 3 khz away from the L1 frequency and the Doppler frequency is assumed to change as much as 1000 Hz. In this figure it is shown that in this specific case where T c >>T p, the effect of the pulse CW is similar to that for CW. Also the effect of the duty cycle of the pulse CW is investigated in this figure. For all of the graphs in this figure, equal power levels of interference are used. 107

127 Chapter 5 Characterization of the effects of CW and pulse CW interference on the GPS signal quality 44 Tp << Tc C/No (dbhz) CW DC = 0.5 DC = 0.1 DC = Doppler frequency (Hz) Figure 5-15 C/N 0 calculated using the mathematical expression for satellite 1 with Doppler frequency changing from 0 khz to 1 khz and pulsed CW interference at 3 khz away from the band center at GHz with different duty cycles. In Figure 5-16, the actual measured C/No using a NordNav software GPS receiver is illustrated. In this experiment, using the single channel utility of the signal generator, the GPS satellite 1 signal was generated with the Doppler frequency changing from 2 to 4 KHz. The idea is to build up the same scenario as in Figure 5-15 which is coincidence of the interference carrier frequency where the interference frequency coincides with the third C/A code spectral line. In these two cases (theory and experiment), the C/No is drawn for different duty cycles in the same interference power levels and T p =0.1T c. It is easy to observe that signal degradation increases with increase in the duty cycle. This dependence comes from the fact that only one line from the Sinc spectral function of the interference periodic pulse is affecting the correlator output power as it is shown in the (5-23) and the power of this line varies with the pulse duty cycle. Another important point is that the C/No level at the Doppler frequencies in which the interference does not coincide with the C/A code spectral line decreases with decreasing the duty cycle (Figure 5-16). This can also be explained by the fact that decrease in the duty cycle means an increase in the width of the spectral Sinc function of the pulse and increasing the possibility of coincidence of more Sinc tails with the Sinc function resulting from the integrator in the tracking loop. This behavior however cannot be seen in Figure 5-15 because as was mentioned earlier, in the 108

128 Chapter 5 Characterization of the effects of CW and pulse CW interference on the GPS signal quality theoretical analysis the assumption is that accurate for interferences close to the spectral line of the code. Δ f i 1 <<. In other word, (5-14) is more T c 46 Tp<<Tc C/No (dbh) CW dc=0.5 dc=0.1 dc= Elapsed time (s) Figure 5-16 Actual C/No measured by the receiver for different duty cycles T c << T p In the previous case where the pulse period was much longer than the integration period, in any integration we were likely to see the same thing as in the CW case. However, this case has many pulses in the integration period and hence we face the opposite situation if explained in the time domain. 1/Tp 1/Td Figure 5-17 Sinc functions associated with the pulse interference (black) and the integrator block of the tracking loop (red) Figure 5-17 shows the relative situation of the integrator and the pulse Sinc functions in the frequency domain. It can be seen that the pulse spectral Sinc function is completely covered by the integrator Sinc function. This is equivalent to saying that 109

129 Chapter 5 in (5-22) all the terms in which this case we will have Characterization of the effects of CW and pulse CW interference on the GPS signal quality k n are almost zero. So it is assumed that k = n. In Z M 2i = Td J n pmcnsinc(( Δfi + ) Td ) m= M Tp j 2nπτ Tc m (0) e (5-24) where M depends on how big is T p compared to T d. For example if T p is 10 times larger than T d, then all terms up to the tenth term of the expression in the (5-24) will have significant value. Similar to Figure 5-15, the theoretical value for C/No resulting from (5-24) for this case are shown in Figure 5-18 and Figure 5-20 with duty cycle fixed and different values of T p and with T p fixed and different values of the duty cycle respectively. What Figure 5-18 shows and Figure 5-19 (the actual value of C/No from a similar scenario) confirms is that the signal degrades more with longer PRP. This is because of the fact that M in (5-24) increases with increasing T p. In Figure 5-18, it can be seen that the troughs get wider from T p = 5 to T p = 20 and then get narrower from 20 to 100. This indicates that in the case where we have many pulses during the integration period, the behavioure of the C/No in the presence of pulse CW interference is not quite as predictable as in the CW case. 44 Tp >> Tc C/No (dbhz) Tp = 5*Tc Tp = 20*Tc Tp = 100*Tc Doppler frequency (Hz) Figure 5-18 C/No calculated using the mathematical expression for satellite 1 with Doppler frequency changing from 0 khz to 1 khz and pulsed CW interference with DC= 20% at 3 khz away from the band center at GHz with varying pulse lengths. 110

130 Chapter 5 Characterization of the effects of CW and pulse CW interference on the GPS signal quality 46 Tp >> Tc C/No (dbhz) Tp = 20*Tc Tp = 100Tc Tp = 5*Tc Elapsed time /s Figure 5-19 Actual C/No measured by the receiver for different pulse periods The match between the theory and actual measurement of the C/No is not as precise as that of the CW. This is probably because of the approximation used in (5-24). Also from Figure 5-20 it is easy to observe that the signal degradation in the presence of pulse CW interference is not overly sensitive to the duty cycle of the pulses for equal power levels and T p = 100 ms. This can be explained noting the fact that for all the three duty cycles, all the lines of the Sinc spectral function of the pulse pass through the low pass filter of the carrier tracking loop and regardless of the duty cycle, the interference has its strongest effect on the carrier to noise ratio. Similar to Figure 5-18, the behavior of C/No for different duty cycles is not perfectly predictable because of the approximation. 111

131 Chapter 5 Characterization of the effects of CW and pulse CW interference on the GPS signal quality 44 Tp >> Tc C/No (dbhz) CW DC = 0.5 DC = Doppler frequency (Hz) Figure 5-20 C/No calculated using the mathematical expression for satellite 1 with Doppler frequency changing from 0 khz to 1 khz and pulsed CW interference with Tp = 0.1s at 3 khz away from the band center at GHz. 46 Tp >> Tc C/No (dbhz) CW dc = 0.5 dc = Elapsed time /s Figure 5-21 Actual C/No measured by the receiver for different duty cycles T c T p 112

132 Chapter 5 Characterization of the effects of CW and pulse CW interference on the GPS signal quality Pulse sincs 1/Tc 1/Td 1/Tc Figure 5-22 Sinc functions associated with the pulse interference (wide) and the integrator block of the tracking loop (narrow) Figure 5-22 shows the integrator and the pulse Sinc functions. It can be seen from the figure that the spectral lines of the pulse spectrum are separated by 1/T p which in this case is almost the same as 1/T c. This means that all these spectral lines have their effect on the correlator output power. In this case we will have Z 2i (0) = + M m= M T d J n p m C n m m Sinc(( + T c m T p Δf ) T i d ) e j 2( n m)πτ Tc (5-25) Similar to the previous case, in this case 2M+1 terms will mainly contribute to the power that the pulse CW interference will transform to the correlator output. M 1 1 depends on two parameters the first of which is the value. The smaller this T p T c value the larger the number of terms that contribute the correlator output power. In other words, pulse CW becomes more dangerous to the GPS receiver if the pulsing period becomes very close to the pseudo-random noise code period. M also depends on the duty cycle of the pulse CW interference and this is because p m tends to zero more quickly for higher duty cycles as can be observed in Figure Just as in the first two cases, the effect of the pulse interference with PRP equal to the C/A code period is investigated theoretically and by experiment. The spectral Sinc of the pulse in this case is not narrow enough to be covered by the spectral Sinc of the integrator in the carrier tracking loop. Neither is it wide enough so that we can assume that only one line coincides with the filter at a time. Therefore in this case, regardless of the duty cycle, since the frequency distance between the spectral lines of the pulsing is close to those of the C/A code lines (PRP is close to the C/A code period) all the lines have their effect on the correlator output power. With 113

133 Chapter 5 Characterization of the effects of CW and pulse CW interference on the GPS signal quality this analysis it is easy to explain the insensitivity of C/No in the presence of pulse CW interference with the same power levels to the duty cycles (T p = T c ). Figure 5-23 (theory) and Figure 5-24 (experiment) are both consistent this analysis. 44 Tp ~ Tc C/No (dbhz) DC = 0.1 DC = 0.5 CW Doppler frequency (Hz) Figure 5-23 C/No calculated using the mathematical expression for satellite 1 with Doppler frequency changing from 0 khz to 1 khz and pulsed CW interference with Tp = s at 3 khz away from the band center at GHz 46 Tp ~ Tc C/No (dbhz) CW dc = 0.5 dc = Elapsed time /s Figure 5-24 Actual C/No measured by the receiver for different duty cycles 5.6 Summary In this chapter, the effects of two types of interference are studied using the Carrier to Noise Ratio (C/No) of the received GPS signal as an indicator of the quality of that signal; Continuous Wave (CW) and pulse CW. For CW interference, it 114

134 Chapter 5 Characterization of the effects of CW and pulse CW interference on the GPS signal quality has been shown analytically that the C/No of the signal can be calculated using a closed formula after the correlator in the receiver. This result was shown to be accurate by calculating the C/No using the I and Q data from a software GPS receiver. At this stage a method for assessing the carrier frequency was introduced. For the pulse CW, a similar analysis is performed to characterize the effect of parameters such as pulse repetition period (PRP) and duty cycle on the quality of the received signal. It is specifically shown that for equal interference power level, for cases where the PRP is far less than the pseudorandom noise code period, the signal degradation increases with increasing duty cycle whereas it doesn t change if the two periods are equal or the PRP is far greater than the code period. These results indicate that the vulnerability to interference is highly variable. It is important to be aware of this in RFI monitoring and interference mitigation. It is not enough to limit the analysis to a discrete set of possible RFI signals. Another important result is that the mitigation techniques should take into account this high variability of the C/No degradation, dependence on Doppler, etc. Hence adaptive mitigation techniques are necessary, even if we know a priori the possible RFI signals in some specific applications. Betz s jamming resistance quality and Spectral Separation Coefficients have the problem of ignoring Doppler frequency which is a major issue when dealing with very narrowband RFI. These two factors are used in the calculation of C/No which conveys the problem to this parameter too. In the expression introduced in this chapter, the effects of Doppler frequency and also integration duration time are both considered. In these calculations the effects of AGC and ADC are not considered and should be considered in future work. It is clear that the value of SNIR will decrease not only because of the existence of RFI but also because of reducing in signal power due to AGC. This effect is less when more ADC bits are used. 115

135 Chapter 6 A PREVENTATIVE APPROACH TO MITIGATING CW INTERFERENCE IN GPS RECEIVERS 6.1 Introduction In this chapter a new CW interference mitigation technique is introduced for GPS receivers. This technique is based on the characterization of the effects of CW interference on the received signal quality which is investigated in Chapter 5. For this purpose, first a new concept of exclusion zone for each satellite is introduced. This exclusion zone, where that satellite should not be used due to interference degradation, is shown to be predictable for each satellite as a function of time. Using this prediction, the CW interference effect on the positioning quality of the receiver can be mitigated by ignoring the affected satellites within exclusion zones when performing position evaluation. The threshold beyond which a satellite should be excluded is then derived by studying the mutual effects of the geometry and the signal quality of that satellite on the positioning quality. Receiver Autonomous Integrity Monitoring (RAIM) uses redundancy in measurements to perform an internal consistency check to see if all of the measurements are satisfactory. In this chapter this technique is also used to mitigate the effect of CW interference on the positioning accuracy. Finally it is shown that the prediction of the exclusion zone for each satellite outperforms the RAIM algorithm in mitigation the effect of the interference when 5 satellites are visible. In section 6.2 the effect of CW interference on the carrier to noise ratio C/No is used as a background to define the new concept of the satellite exclusion zone. In section 6.3, the elements of satellite positioning quality are studied to obtain an appropriate threshold to determine the exclusion zone. The proactive mitigation algorithm is also presented in this section. Section 6.4 is dedicated to the experiments to mitigate interference from real GPS signals collected by a software GPS receiver using both a RAIM algorithm and the proposed preventative technique. Section 6.5 summarises the results. 116

136 Chapter 6 A Preventative Approach to Mitigating CW Interference in GPS Receivers This work has been published in Tabatabaei (2006a, 2006c and 2007d). Beatrice Motella helped doing the experiments of section 6.4. The author developed the theories and ideas and performed rest of the experiments. 6.2 The Effect of CW RFI on the GPS Signal Quality It was shown in the previous chapter how C/No behaves in the presence of CW RFI. In Figure 6-1, as an example, using (5-14) and assuming a specific environmental noise power, the C/No is shown for satellite 1 with Doppler frequency changing from 0 khz to 10 khz and CW interference with a specific power at 14 khz away from L1 frequency (i.e. at GHz). The deep troughs in this graph correspond to the coincidence of CW RFI with the code spectral lines. It is clear that this happens at 1 khz spacing in the Doppler frequency, as expected because the code repeats each 1ms. As expected from (5-14), there are different values for different lines. This difference comes from the difference between the coefficients of different lines in the code spectrum. This particular line spectrum would thus be different for different satellite codes. The other point which is noticeable in this figure is the inverted Sinc functions occurring around each trough. The width of each inverted Sinc function is related to the integration period, as can be seen in (5-14). The longer the integration period is the narrower will be the Sinc functions and the more immune the receiver will be to CW interference away from the code spectral lines (and the more vulnerable at the spectral lines). 117

137 Chapter 6 A Preventative Approach to Mitigating CW Interference in GPS Receivers C / No (db-hz) Doppler frequency (Hz) Figure 6-1 C/No calculated using the mathematical expression for satellite 1 with Doppler frequency changing from 0 khz to 10 khz and CW interference at 14 khz away from L1 frequency Figure 6-2 shows the variation of the Doppler frequency for different satellites for 24 hours for a specific almanac file in the presence of narrowband CW interference. Gaps in the plots indicate where an interferer at the L1 frequency (or any integer multiply of 1 khz away from that) may cause these signals to be lost. Depending on satellite number, signal power, strength of the interference, the integration period of the tracking loops and the background noise power, the width of this gap changes. Instead of losing lock, we can set a threshold for the C/No which is a good indication of the signal quality. For any value of C/No less than this threshold, that specific signal will be taken out of the operating channels. We call the zones that are identified by this algorithm exclusion zones. This is the frequency region in which the interference knocks out that satellite and the pseudorange for that satellite should be excised from the solution. 118

138 Chapter 6 A Preventative Approach to Mitigating CW Interference in GPS Receivers Doppler frequency (Hz) Elapsed Time (h) Figure 6-2 Variation of Doppler frequencies for the visible satellites over 24 hours. Exclusion zones are indicated at multiples of 1 khz. There are a number of different techniques for detecting the RFI and calculating its frequency and power. Statistical inference applied to the Fourier transform is a common technique which is used for this purpose in Chapter 4. Once the exact frequency of the RFI is calculated, the second step is to find the exclusion zone for each of the lines. In Figure 6-3 this quantity is shown in terms of the corresponding trough depth in the C/No calculated theoretically (5-14). In this experiment, a conservative value of 40 dbhz is chosen for the C/No threshold. In Figure 6-2, each line represents the Doppler frequency of each satellite. The gaps in the lines of this graph are the Doppler frequencies in which the C/No of that satellite is bellow the threshold. Obviously the more power in the spectral line, the greater the effect of the interference, the deeper will be the trough in the C/No and the wider will be the gap. Figure 6-3 clearly shows that the deeper the trough, the wider will be the exclusion zone for that satellite around that specific line. As an example, the first point in this graph shows 27.7 db-hz and 136 Hz. This means that if a trough with the depth of 27.7 db-hz is generated in the trend of C/No (like one of the troughs of Figure 6-1), then with the receiver threshold of 40 db-hz for C/No, the value of C/No is less than the threshold for 136 Hz of Doppler frequency change (note that the horizontal axis of Figure 6-1 is Doppler frequency change of the satellite). So by using Figure 6-1 which provides an idea of how to calculate the exclusion zone in terms of C/No and (5-14) with which we can calculate the C/No, the exclusion zone of any satellite signal at any Doppler frequency can be calculated. 119

139 Chapter 6 A Preventative Approach to Mitigating CW Interference in GPS Receivers Exclusion zone (Hz) Minimum C/No (db-hz) Figure 6-3 Relation between trough depths in C/No calculated theoretically and the corresponding satellite exclusion zone In Figure 6-3, it can be seen that there is a relatively linear relationship between the width of the exclusion zones of different C/A code spectral lines and the C/No trough depth which is calculated theoretically as the result of those lines. In the following experiment the NordNav software receiver is used to capture the IF data to be analyzed and post processed and a signal generator (HP8648B) is used to generate the CW interference which is combined with the GPS signal generated by a SPIRENT GSS6560. The aim of this experiment is to characterize the effect of interference power on the level of C/No. The exclusion zone is characterized for just the two lines circled in Figure 6-4. Line one in Figure 6-4 corresponds to the 4 th spectral line of the code away from L1. So the 6 th and 7 th lines are at 9 and 10 khz away from the band centre in this figure. Both the theoretical and the actual calculated C/No (using the I and Q samples from the NordNav software receiver) are shown to closely correspond to each other in this figure. 120

140 Chapter 6 A Preventative Approach to Mitigating CW Interference in GPS Receivers C / No (db-hz) Doppler frequency (Hz) Figure 6-4 Actual C/No calculated using the I and Q samples (red) and Theoretical C/No calculated using 1 (blue) for PRN 1 and interference at 4 khz away from L1 Instead of 10 khz, the interference is moved only 2 khz in 4 minutes (which guarantees crossing two 1 khz-spaced lines separated by 2 minutes). In this experiment, the wideband noise power and the signal power are kept constant. The experiment is performed for four different RFI powers (Table 6-1). Table 6-1 Exclusion zone widths for four different RFI powers for two consecutive C/A code spectral lines RFI Power/ exclusion zone - 82dBm - 85dBm - 88dBm - 91dBm Line 1 94 Hz 87 Hz 22 Hz 0 Hz Line Hz 95 Hz 14 Hz 7 Hz 121

141 Chapter 6 A Preventative Approach to Mitigating CW Interference in GPS Receivers C / No (db-hz) Elapsed time (s) Figure 6-5 C/No for four different values of RFI power (magenta, black, blue and red for -82, -85, -88, -91 dbm respectively) 40 db-hz chosen as threshold In Figure 6-5, the effect of these four power levels of RFI is shown on the C/No. It is obvious that where the RFI lines up with the C/A code spectral line, the higher power has the more serious effect. The other thing that can be seen from Figure 6-5 is that where the RFI does not line up with the C/A code line, the power of RFI does not have a significant effect on the C/No. In other words CW RFI affects C/No only when it lines up with the C/A code line. In Table 6-1, it can be seen that the exclusion zone for the two lines increases with increasing power of RFI. This is expected, as the depth of the trough is greater for greater RFI power, and the width is also greater. 6.3 Positioning Quality Elements Satellite geometry and satellite signal quality, mathematical approach The accuracy of positioning, using the measured pseudoranges from the receiver to each of the satellites, depends on several different factors. The position evaluation in the GPS receiver estimates four quantities (x, y, z and time) using four or more pseudoranges. Briefly there is the following relationship between the pseudorange error ( Δ ρ ), position error ( Δ x ) and the geometry matrix (G) (Parkinson, 1996a). T 1 T Δx = ( G G) G Δρ (6-1) 122

142 Chapter 6 A Preventative Approach to Mitigating CW Interference in GPS Receivers It is easy to see in this equation that the position accuracy is decided by two factors, the measurement quality and the user-to-satellite geometry. These factors separately are extensively discussed in Parkinson (1996a). In this section we will make a quantitative comparison of the effect of each one of these two factors on the positioning accuracy. The aim is to establish if pseudorange error is large due to the poor signal quality (low C/No), under which circumstances we can achieve better position accuracy by eliminating that satellite, noting the fact that eliminating the satellite will affect the geometry. To achieve this goal, we simplify the scenario. The assumption is that there are 5 satellites available to the receiver and of these only one has affected signal quality. This means that only this satellite is affected by interference. This situation is likely to occur when there is moderate blockage of sky and a single CW interferer is present. The position error covariance is studied in this investigation: T cov( Δ x) = E( ΔxΔx ) where E(.) operates as an expected value operator. From (6-1) we have: cov( Δx) = E(( G T G) G T T Δρ Δρ G( G G) ) (6-2) 1 T 1 At this stage two different cases are considered: 4 satellites all having the same pseudorange error (ε ) and 5 satellites one of which has a degraded signal with larger pseudorange error (η ). Without loss of generality we can assume here that the pseudorange error in the first case for each satellite is ε = 1 m and that of the second case to beξ = η / ε = η. Then for the two cases we will have: T 1 T T cov 4 ( Δ x) = ( G4 G4 ) ( G4 W4 ( ξ ) G4 )( G4 G4 ) (6-3) where W 4 ( ξ ) = I 4 and T 1 T T cov5 ( Δ x) = ( G5 G5 ) ( G5 W5 ( ξ ) G5 )( G5 G5 ) (6-4) I 4 0 where W 5 ( ξ ) = 2. 0 ξ G 4 and G 5 represent the G matrix respectively for the cases of 4 and 5 satellites. The difference between the above two quantities comes from the difference between G 4 and G 5 on the one hand and W 4 and W 5 on the other. In the scenario 123

143 Chapter 6 A Preventative Approach to Mitigating CW Interference in GPS Receivers explained in the following section, the effects on the position error covariance of W and G are studied. The data used for this scenario are a set of real data collected with a GPS software receiver NordNav-R30 at the University of the New South Wales on the 6th November Six satellites (1, 11, 20, 23, 25 and 31) are acquired by the receiver. To compare the covariance matrices (6-3 and 6-4), one way is to compare their determinants. Two satellite sets of (1, 23, 25, 31) and (1, 11, 23, 25, 31) are chosen. These two sets are chosen because during the initial epochs of the data, satellite 11 plays a fundamental role in delivering good geometry of the constellation. In this experiment, as explained earlier, only the pseudorange error of the satellite 11 is changed. By using (6-3) and (6-4), the amount of pseudorange error of satellite 11, which makes the position error for the two configurations equal, is found to beξ 0 = 22 m. Figure 6-6 shows the position error for the two cases with respect to the pseudorange error of satellite 11 with (blue) and without (red) the use of this satellite in the positioning calculations. It is obvious that the position error should not change by changing the pseudorange error of satellite 11 (red) whereas it should change in the case where satellite 11 is considered in the positioning calculations (blue). It is clearly seen that the position error for the two configurations (4 satellites and 5 satellites) become equal to each other at ξ 0 = 23.1 m which is very close to what is predicted theoretically satellite HDOP = 21 5 satellite HDOP = 1.95 Position error (m) Satellite 11 pseudorange error (m) Figure 6-6 Position error vs. pseudorange error for the 4 and 5 satellite configuration at a particular time for satellite 11 which is important to the geometry of the 5-satellite constellation 124

144 Chapter 6 A Preventative Approach to Mitigating CW Interference in GPS Receivers In another scenario, another two satellite sets are chosen which have very similar and good geometries. These two sets are (1, 11, 20 and 23) and (1, 11, 20, 23 and 25). The satellite of which the pseudorange error has been increased is satellite 25. Again using (6-3) and (6-4), the pseudorange error at which the two position errors become equal is calculated. This error is found to beξ 0 =1.45 m. Figure 6-7 shows that this value in the experiment was found to be 1.65 m. This means that there will be times when eliminating the satellite because of its poor CNo will have advantages and others when the degradation has to be significant before elimination helps satellite HDOP = satellite HDOP = Position error (m) Satellite 25 pseudorange error (m) Figure 6-7 Position error vs pseudorange error for the 4 and 5 satellite configuration for satellite 25 which contributes little to constellation geometry Mitigation Algorithm Here the algorithm for the proposed technique to mitigate the effects of CW interference is described. 1. As was explained in section 6.2, using the information regarding the frequency and power of the interference and the constellation information, the carrier to noise ratio of each channel is predicted using (5-14). 2. Using (6-3) and (6-4), for each of the satellites in view, the minimum pseudorange error at which the positioning error becomes higher when including that particular satellite is found. This value is calledξ 0. Here it is assumed that the relationship between the pseudorange error and the carrier to noise ratio for the GPS receiver in which this algorithm is used, is known (Ndili, 1998). Usually the 125

145 Chapter 6 A Preventative Approach to Mitigating CW Interference in GPS Receivers pseudorange is calculated using the code phase so what is really important is to see the effect of the interference on the code tracking error. This has been done in Betz (2000). Different techniques are discussed in that paper regarding the code phase error among which is the effect of C/No on this parameter. Also in Kaplan (2005), a lower bound is given for code tracking error in terms of C/No which is even independent of code tracking system design. 3. If the predicted carrier to noise ratio of a satellite is less than the value which corresponds to ξ 0 m pseudorange error, then that satellite will be excluded from the positioning evaluation. This process is schematically shown in Figure 6-8. Figure 6-8 Algorithm description flow chart It was shown in section 6.2 that degradation in satellite signal quality (C/No) due to interference can prevent one or more satellites being available. In Fante (2000) the probability of availability of N satellites in the presence of interference in terms of C/No was investigated. Due to the nature of this impact of CW RFI on the GPS satellite signals, one can reasonably think of using a RAIM algorithm to mitigate this effect. In Kim (2006) the position domain errors are assessed using traditional leastsquares estimation in the presence of interference, mitigated by a RAIM scheme. Also in Yun (2006), an RFI mitigation technique is introduced based on integrity monitoring for a Differential GPS (DGPS) system. In this section the dependency of our proposed mitigation algorithm on the satellite geometry was studied. The dependency of RAIM on geometry was also discussed in Brown (1990). So, in the fact that the two algorithms are dependent on the geometry, they are similar but there are differences between the two mitigation 126

146 Chapter 6 A Preventative Approach to Mitigating CW Interference in GPS Receivers techniques which are analysed and discussed in the next section using some experiments. So in the fact that the two algorithms are dependent on the geometry they are similar but there are differences between the two mitigation techniques which in the next section using some experiments are analyzed and discussed. 6.4 Experiment Discussion and Results The Impact of the HDOP The goal of this section is to demonstrate the theoretical analysis discussed in sections 6.2 and 6.3 with some experiments. The data used for this purpose are the same of those have been used for the simulation results. The question to be addressed is in which cases the exclusion zones algorithm can be applied. In other words: what is the trade-off between the loss of positioning accuracy due to degraded geometry and a loss of position accuracy due to the use of the satellite affected by CW interference? This is then used to decide if the affected satellite should be eliminated from position estimation. By analyzing the following two examples this question is approached. In the first example we analyze the possibility of applying the exclusion zone algorithm in the case where the HDOP is comparable before and after excluding one satellite. The constellation considered is composed of satellites 1, 11, 20, 23 and 25. The CW interference affects the signal of satellite 1. It can be seen from Table 6-2 that the value of the HDOP does not change significantly when satellite 1 is removed. In fact the HDOP for both cases stays almost constant for the 38 minutes of data. Table 6-2 HDOP for the 5 and 4 satellite configuration HDOP - 5 sat HDOP - 4 sat The position error is described versus time (Figure 6-9 a) on the scatter diagram (Figure 6-9 b). Blue lines refer to the 5 satellite configuration, while red lines refer to the position error evaluated omitting satellite

147 Chapter 6 A Preventative Approach to Mitigating CW Interference in GPS Receivers sat 4 sat 25 position error (m) time (minute) (a) (b) Figure 6-9 Position error vs. time (a) and scatter diagram (b). The comparison is between the 5 satellites configuration (blue line) and the 4 satellite one (red line) when the interference affects the satellite 1 (kept out in the 4 sat configuration) It is observed that the performance of the 4 satellite configuration is comparable with the 5 satellite case, except where the interference matches one of the lines of the PRN 1 code (between min 16 and min 22). On the contrary, the second example analyses the case where the HDOP varies significantly after removing one satellite. The constellation is composed of satellites 1, 11, 20, 25 and 31. The CW interference affects the C/No of satellite 11 (Figure 6-10 a). Moreover in the 4 satellite configuration the value of the HDOP varies from 22.5 to 4 during the 38 minutes (Figure 6-10 b). 128

148 Chapter 6 A Preventative Approach to Mitigating CW Interference in GPS Receivers (a) 25 5 sat 4 sat HDOP time (minute) (b) Figure 6-10 C/No for the PRN 11 (a) HDOP for the 5 and 4 satellites configuration (b) During the first 18 minutes the difference between the HDOP of the two constellations is very high. This means that the HDOP of the 4 satellite configuration is extremely high and this is reflected in a very high position error (see Figure 6-11). The maximum error in this case is 35 m and it exceeds 15 m several times. With this level of position accuracy there is no motivation to apply the exclusion zone algorithm, because the HDOP does not allow acceptable position estimation. On the contrary, when the difference in HDOPs is not significant (less than 7 after min 18), it is easy to observe that the performance of the 4 satellite configuration, after the application of the exclusion zone algorithm, is much better than the 5 satellite configuration when the interference matches one of the lines of the 129

149 Chapter 6 A Preventative Approach to Mitigating CW Interference in GPS Receivers PRN 11 code (Figure 6-11 a). Note that the interference is present throughout the experiment. In order to see how much the interference can affect the position accuracy (Figure 6-11 b) represents the difference with and without interference. Also in these figures the solutions have piecewise smooth sections that all seem to be the same size and there are small jump after each smooth section. This seems to be related to the positioning algorithm of NordNav software receiver. Fortunately these jumps do not affec the overall trend of positioning and the validity of the point that we are making here sat 4 sat 35 positioning error (m) time (minute) (a) sat interefrence 5 sat - no interefrence positioning error (m) time (minute) (b) Figure 6-11 Comparison between the positioning error using 5 and 4 satellites in presence of interference (a) Comparison between the positioning error using the 5 satellites constellation with and without interference (b) Exclusion Zone RAIM Comparison 130

150 Chapter 6 A Preventative Approach to Mitigating CW Interference in GPS Receivers In order to show the advantages of the exclusion zone algorithm in comparison to a RAIM technique, an example is used. The maximum solution separation RAIM method (Parkinson, 1996b) has been chosen for the comparison. To apply the exclusion zone algorithm, we consider the example in the previous subsection, (Figure 6-9). In (Figure 6-12 b), which is the trend of the C/No of satellite 1, it is observed that the decision to exclude the satellite cannot be based only on measuring the actual C/No level. This level is in fact always quite high (> 34 db-hz) even when the position quality is severely affected by the interference. This is explained in Chapter 5 where the behavior of C/No in the presence of CW interference is characterized. It is shown that this type of interference has an improved effect on the C/No, when its frequency is very close to the frequency of one of the lines of the C/A code. The reason is that the carrier tracking loop tracks the interference and the stronger the interference, the higher will be the C/No. On this basis, the C/No by itself can be a misleading indicator in this specific situation. Instead of monitoring the C/No, by predicting the C/No this problem is resolved. In this experiment the frequency of the CW RFI is chosen so that the signal of GPS satellite 1 is affected significantly during the course of the 38 minutes. The Doppler frequency of this satellite is shown in (Figure 6-12 a). The interference is chosen to be at frequency 12.1 khz away from L1. As the Doppler frequency of this signal is 2 khz, so the RFI will coincide with the 10 th spectral line of this code in the middle of the experiment. The RFI power is chosen to be -82 dbm because this guarantees C/No degradation see Table 6-1. In section 6.3, we proposed a method to calculate, at any given time, the C/No threshold ξ 0 at which a satellite should be excluded. Following that analysis, ξ 0 for this specific case is 2.1 m. In the next step the carrier to noise ratio, corresponding to this level of pseudorange error should be found. In Ndili (1998) the relationship between pseudorange error and the correlator output power or the C/No is characterized. This relationship varies based on the receiver. The fact that ξ 0 is very close to 1 m means that satellite 1 does not have a significant effect on the geometry and therefore should be excluded from the positioning calculation as soon as the positioning error is affected by the poor signal quality. For this specific PRN code line and RFI power level and the threshold, the exclusion zone from (5-14) is found to be 101 Hz (Figure 6-12). Again the RFI is 131

151 Chapter 6 A Preventative Approach to Mitigating CW Interference in GPS Receivers present throughout the experiment. However its detrimental effect is worst only during the exclusion zone. As is clear from Figure 6-9, the position error has been improved significantly after applying this technique. In Table 6-3, this improvement is quantified. Table 6-3 Maximum position error before and after applying the mitigation technique Maximum position error before Maximum position error after applying applying the preventative RFI the preventative RFI mitigation mitigation 30.5 meter 7 meter Hz Doppler frequency (Hz) time (minute) (a) Exclusion zone C/No (db-hz) time (minute) (b) Figure 6-12 (a) Doppler frequency for PRN 1 (b) Exclusion zone for PRN 1 132

152 Chapter 6 A Preventative Approach to Mitigating CW Interference in GPS Receivers The advantages of this algorithm are most appreciable when we have only 5 satellites in view. This is because in this situation the RFI affected satellite can have a stronger effect on the positioning quality whereas for higher numbers of satellites, it is possible that even though the affected satellite does not have significant effect on the geometry, elimination of that satellite does not improve the positioning quality either. The comparison with RAIM is also less stark as RAIM algorithms can remove a faulty satellite from a constellation of 6. This is shown schematically in another scenario from the same data. We have 6 satellites in view, (PRN 1, 11, 20, 23, 25 and 31). Using the exclusion zone algorithm, we know that the signal from satellite 1 can be seriously damaged by the presence of the interference at a particular time we also know the other satellites are not affected in this way. Figure 6-13 show the position error before and after the exclusion zones in the scatter plot and versus time respectively. 133

153 Chapter 6 A Preventative Approach to Mitigating CW Interference in GPS Receivers 10 Position scatter plot (m) constellation: constellation: (m) (a) constellation: constellation: Positioning error (m) Time (min) (b) Figure 6-13 Position error for 6 satellites, one of which is affected by interference scatter plot (a) Positioning error before and after the exclusion zone (b) The advantage of removing satellite 1 from the position estimation does not bring big advantages in terms of position accuracy. As we saw in the first example, the advantages of the algorithm became more appreciable when we have only 5 satellites in view. The RAIM algorithm needs at least 5 satellites to detect the presence of an error in one of the pseudorange and 6 satellites to identify which is deficient. In order to compare the exclusion zones algorithm with RAIM, the situation with 5 satellites in view is analyzed. Figure 6-14 shows the maximum distance between the five position solutions evaluated removing one satellite at a time (constellation: PRN 1, 11, 23, 25, 31). 134

154 Chapter 6 A Preventative Approach to Mitigating CW Interference in GPS Receivers In this case it is difficult to detect the presence of the error using the maximum distance of solutions algorithm. The different positions are evaluated using only 4 satellites. This means that the HDOP can significantly affect the accuracy in the positioning evaluation. In other words, the effect of poor HDOP is more than the effect of the existing interference and between minute 16 and minute 22, the maximum separation distance is not higher than the other time where RFI has no effect. Therefore in this experiment, even though 5 satellites are available, because of poor geometry, the presence of error caused by interference cannot be detected Maximum Separation Distance (m) Time (min) Figure 6-14 Maximum distance between positioning from different satellite configuration for the 5 satellite set, one affected by interference Here of course the investigation is carried out on a particular real data example. In this example it was shown how the exclusion zone algorithm results in significant improvement of the positioning quality for low numbers of satellites. RAIM algorithms are not competitive in those situations because of lower redundancy which is essential to RAIM algorithms. 6.5 Summary In this chapter a new technique to mitigate the effects of CW interference on GPS C/A code signal quality is introduced. No analogue or digital filter is used in this algorithm and this helps keep the GPS signal phase and amplitude from being distorted. Unlike other mitigation techniques which are responsive, this technique works preventatively but it does require knowledge of the interference frequency and power, which may be estimated by known techniques plus the relationship between the C/No and the pseudorange error for the receiver. This means that it predicts and 135

155 Chapter 6 A Preventative Approach to Mitigating CW Interference in GPS Receivers prevents the error before it happens. The specific signal structure in GPS allows us to predict the effect of CW interference on each of the satellites signal (C/No) at any given time. It usually affects one signal at a time and in the technique proposed in this chapter that affected satellite is removed from the positioning calculations provided that its C/No is less than a threshold. This threshold depends on the effect of that satellite on the user-to-satellite geometry. Receiver autonomous integrity monitoring techniques work on a similar basis. The difference is that in the RAIM approach the affected signal is detected after the error appears in the positioning calculations. The other difference between the RAIM technique and the proposed preventative algorithm is that RAIM needs a higher number of received satellites whereas the preventative approach is effective in the presence of few satellites. It is shown in a case study that in the five satellite case, the positioning error was improved from 30 m to 7 m. For that example the maximum solution separation RAIM algorithm could not identify the error caused by the interference. 136

156 Chapter 7 GNSS SATELLITE AVAILABILITY IN THE PRESENCE OF CW RFI 7.1 Introduction Considering the designer s intention of maintaining interoperability between different satellite navigation systems, it is reasonable to seek a quantified comparison between the different systems in terms of vulnerability to CW interference. In this chapter, considering the signal structures, the characterization of the effect of CW interference on the C/No for GPS and Galileo is investigated and compared. It is shown that for the available Galileo signal (GIOVE-A BOC(1, 1) in the E1/L1 band), the worst spectral line happens far from the L1 frequency whereas for GPS, it usually happens close to the L1 frequency. If the frequency of the CW RFI coincides with the worst spectral line the effect is more serious to the quality of the signal. Considering these facts, the probability of availability of one GPS satellite is compared with that of a Galileo satellite in terms of interference power and frequency. It is shown that these two systems can be considered as alternatives to each other in the presence of different RFI frequencies as their availability in the presence of CW RFI is different in terms of RFI frequency and power (L5, L2C, E5 and E6 are not considered). In section 7.2 of this chapter the commonality between the GPS C/A code and Galileo signal is studied as a basis to build up a common term for satellite availability. In section 7.3 the probability of satellite availability in the presence of CW interference is defined. In section 7.4 the two currently available satellite navigation systems (GPS L1 signal and Galileo signal (GIOVE-A BOC(1, 1) in the E1/L1 band)) are compared in terms of probability of availability in the presence of CW RFI. In this section the effect on the satellite availability of important interference parameters like frequency and power are also investigated on the satellite availability. Section 7.5 summaries the chapter. The ideas and theory covered in this chapter were developed by the author. A fellow UNSW PhD student Jinghui Wu helped conducting the experiments. This work has been published in Tabatabaei (2007f). 7.2 Commonality of Signal Structure of different GNSS Systems Based on the signal generation schemes of the Galileo ICD (GIOVE-A) (ICD1, 2007) and (ICD2, 2004), each sampled Binary Phase-Shift Key (BPSK) signal modulates a carrier. The signal generation for GPS and Galileo are shown in Figure 7-1 and Figure 137

157 Chapter 7 GNSS Satellite Availability in the Presence of CW RFI 7-2 respectively. As the secondary code (S-code) of GIOVE-A pilot channel is invisible when 8ms integration times are used, it can be regarded in the same way as the navigation bits, similar in a way to the Galileo data channel and the GPS L1 signal. Therefore the effects of the S-code in the GIOVE-A pilot channel and the navigation data for GPS are not included in the simulation data. The linear Doppler frequency shift applies only to the carrier frequency while the minor effect on code phase is neglected. Moreover as the changing Doppler frequency shift will determine how fast a CW RFI sweeps across the PRN spectral lines in the frequency domain, the Doppler frequency shift rate needs to be slower when the spectral lines are close to each other (Galileo case), so that the effect of the RFI while coinciding with the spectral line is better observed. The rate of the Doppler frequency change is too low compared to the reciprocal of the integration time (for both cases see section 7.2.2) of the carrier tracking loops. This means that the behavior of C/No is still similar in the two cases. Therefore, in order to better investigate the interference effect on C/No, a simulated Galileo satellite signal should have a slower Doppler frequency changing rate (0.5Hz/Second and 1Hz/Second) for 10mins compared to two simulated GPS satellite signals (2Hz/Second and 3Hz/Second for 20minuts data). The longer time for GPS is because of the greater distance between two adjacent spectral lines (1 khz) compared to Galileo (125 Hz). Finally, additive white noise and interference are added on to each of the signals. Figure 7-1 GPS signal generation for L1 band C/A code 138

158 Chapter 7 GNSS Satellite Availability in the Presence of CW RFI (a) (b) Figure 7-2 Galileo BOC(1,1) modulation scheme a) Simulating pure satellite signal b) signal plus noise and interference Worst Spectral Line for GPS PRN1 and Galileo GIOVE-A According to the mathematical expression for C/No (Chapter 5), at the PRN spectrum line which has the highest power in the frequency domain the GNSS receiver will be most vulnerable to CW RFI. This particular spectral line is called the worst line (Kaplan, 2005). Different PRNs have different worst lines. In order to compare the worst case in the environment of CW RFI for both GPS and Galileo receivers, the unique worst lines for both GPS PRN1 and Galileo GIOVE-A are found in this section. GPS PRN1 has a full length code period of 1ms, so the spectrum lines (neglecting the navigation data) are distributed across the frequency band evenly 1 khz away from each other (Kaplan, 2005). The GIOVE-A C-code has 8ms full length, so the Dirac lines of the pilot channel PRN spectrum are thus distributing evenly but closer (125 Hz) to each other. By selecting an appropriate resolution in the frequency domain, each of the separate spectral lines for both PRN1 and the GIOVE- A pilot channel PRN can be clearly seen in Figure 7-3. The worst line for GPS PRN1 is at 42 khz and the worst line for the GIOVE-A pilot channel C-code is at 771 khz. 139

159 Chapter 7 GNSS Satellite Availability in the Presence of CW RFI Amplitude Amplitude Frequency (Hz) x 10 7 (a) Frequency (Hz) x 10 4 (b) 2.5 x 104 x Amplitude 1 Amplitude (c) x 10 7 Frequency (Hz) Figure 7-3 PRN spectrum and their worst lines. GPS PRN1 spectrum (a) and its 42 khz frequency component (b); Galileo C-PRN (Square wave modulated with C-code) spectrum (c) and its 771 khz frequency component (d) Frequency (Hz) (d) x Tracking Loop design The experiments in this section used the Kai Borre software receiver (Borre, 2007). This software receiver is a development receiver in MATLAB. For fair comparison between the effects of CW RFI on the two signals, the tracking loops for both GPS and Galileo are designed based on the same receiver. This is done using a conventional Costas tracking loop. However, the conventional delay lock loop (DLL) for BOC (1, 1) is more sensitive to noise (Julien, 2004) because of the ambiguity property of the DLL for binary offset code. Therefore in order to avoid the possibility of false lock and lost lock, the noise power for this experiment is set to be as low as only 4 times (6dB) higher than the satellite signals power. Also, the threat of acquisition ambiguity is eliminated during the acquisition process by always acquiring a signal whose C/No is higher than 50dBHz (Julien, 2004). In this way, the effect of interference on both GPS and Galileo signals dominates the effects of noise. Although this is not a realistic situation, both the GPS and Galileo tracking loops will maintain lock and only the effect of interference on both signals can be observed. Besides noise level and interference level, the C/No computed from the Costas loop outputs of I-Prompt and Q-Prompt depends on the tracking loop parameters (e.g. 140

160 Chapter 7 GNSS Satellite Availability in the Presence of CW RFI integration time, loop gain, and DLL discriminator early and late chip distance). So in order to have a tracking loop performance comparison by considering the different inputs of Doppler frequency acceleration, integration time for GPS and Galileo signal tracking simulation and their different auto-correlation function shapes, the different tracking loop parameters are chosen as shown in Table 7-1. Table 7-1 Tracking Loop setting for different data GPS Galileo Integration time T 8ms 64ms 8ms Early-Late distance (d) 0.35chip 0.1chip sampling frequency Hz (F s ) Intermediate frequency Hz (f IF ) The settings in Table 7-1 are selected to maximize the possibility of keeping lock for both GPS and Galileo signals in the presence of varying frequency and power of CW RFI. 7.3 Probability of Satellite Availability in the Presence of CW RFI The concept of exclusion zone is introduced in Chapter 6. If there is a certain value of C/No that has been set as a threshold and regarded as the minimum signal quality that a GNSS receiver tracking loop can handle, the interference affected satellite signals whose C/No are below the threshold, are regarded as unavailable. Different satellite systems might have different C/No characteristics in a hybrid GPS/Galileo L1 receiver when the signal is exposed to the same CW RFI frequency and power level. In order to analyze the system interoperability in terms of satellite availability due to the effect of CW RFI on two satellite signals, an experiment was designed. Here Figure 7-4 is used to explain this concept.. The C/No in this experiment is calculated using the narrowband wideband power estimation technique (5-19). First of all a mean value of vectors of C/No (Pink) are calculated when there is no interference present in the generated noisy satellite data (S/N=-6dB). The mean values of C/No (light Blue) of GPS and Galileo signals are computed respectively. Hence each of the C/No thresholds (red) are set at a value with the same amount of margin (5 141

161 Chapter 7 GNSS Satellite Availability in the Presence of CW RFI db). In other word, the threshold is a fixed amount (in db) lower than the average C/No calculated for the system when interference does not exist. The reason of choosing this method to calculate the satellite probability of availability in the presence of CW RFI is to make the comparison between the two system as fair as possible. The Galileo data used in this experiment has Doppler frequency changing at 0.5Hz/sec for 10min while the GPS data has a change of 3Hz/Sec for 20 min. As discussed in Chapter 5, the width of the main lope of C/No trough is inverse proportional to integration time T and proportional to the value of the power of CW RFI. So in order to discard the effect of different integration times on the two tracking loops, the integration time 8 times their PRN code periods is selected (T= 8ms for GPS and T=64ms for Galileo). The width of each symmetric trough main lobe will occupy 2/8 times of each interference affected zone area (which equals the distance between two troughs in the time domain) caused by the coincidence between CW RFI and each of the spectral lines. In other words, the width of the trough main lobe in this experiment will depend on the PRN code period and the effect of CW RFI on each crossing spectral line. In Figure 7-4, d1 and d2 are the exclusion zone periods. The satellite availability probability (P a ) is then obtained accordingly (7-1) P a = =... d N i i 1,2 1 (7-1) D where N is the number of troughs crossed by the threshold. 142

162 Chapter 7 GNSS Satellite Availability in the Presence of CW RFI CNo (dbhz) d1 d2 46 D CNo1: An/As=2; Ai/An=0 Mean of CNo1 (AV) Threshold =AV-5dbHz CNo2: An/As=2; Ai/An=2;Nf=12 Threshold crossing d1 Threshold crosing d Time (0.064sec) Figure 7-4 Definition of probability of satellite availability The longer the integration time, the narrower is the loop bandwidth. Therefore, a further experiment with T=8ms for Galileo signal was also carried out for a more direct comparison with GPS, to ensure same CW RFI power passes through the two loop filters. 7.4 Comparing the Effect of CW RFI on the GPS and Galileo Available Signals RFI Frequency Based on the previously described experiment setup for satellite availability analysis, different frequency CW RFI signals with the same power were added to each of the satellite signals before tracking. The frequency of the CW RFI changes from 0Hz to 771 khz. As expected, the lowest P a for both signals occurs at a CW RFI frequency close to their worst spectral lines. However, Galileo appears to be more seriously affected (its P a is very low) by a CW RFI whose frequency is close to its worst line in both integration times (Figure 7-5). When the CW RFI comes close to 771 khz for the Galileo case, the satellite probability of availability is about which seems to be lower than the GPS case when CW RFI frequency is close to the worst line of GPS PRN1. However, Galileo (64ms, blue) also exhibits a less vulnerable character over a wider range of CW RFI frequencies than GPS does. For Galileo (8ms, green) that is not the case. As can be seen for the P a for GPS PRN1, Figure 7-5(red) is almost always lower 143

163 Chapter 7 GNSS Satellite Availability in the Presence of CW RFI than that of Galileo GIOVE-A (blue) over the chosen frequency range except for the frequencies very close to the Galileo worst line. This can be explained noting that the troughs in C/No in the case when integration time is 64ms is narrower than those of the case when integration time is 8ms (this is shown in section 5.4 of Chapter 5) and thus according to (7-1), the Pa in 8ms integration time is usually lower than 64ms Probability of Satellite Availability GIOVE-A 64ms GPS PRN-1 8ms GIOVE-A 8ms CW RFI Frequency (khz) Figure 7-5 GPS and GIOVE-A probability of availability versus frequency of CW RFI RFI RFI Power In another scenario, the CW RFI is chosen to sit right in the middle of the two worst lines of GPS and Galileo signal spectral worst lines. This experiment is designed to investigate what role interference power plays for the two cases. Therefore the RFI frequency is chosen to be fixed. In Figure 7-6, the green line which is the Galileo (8 ms) C/No, is always higher than that of GPS (the red line) until a specific point where 10log10(Pi/Ps)=18. From this point on, GPS has higher satellite availability in the presence of CW RFI. This result is completely consistent with what was expected theoretically. The power in the Galileo signal is distributed among more lines than GPS. This means for the same signal power, each of the spectral lines for the Galileo signal has less height compare to GPS. In other words, in the Galileo signal, we have a high number of smaller spectral lines distributed closer to each other. Because of this, when interference power is lower than a threshold, the Galileo signal is most probably available and after that threshold, it is most probably not available. 144

164 Chapter 7 GNSS Satellite Availability in the Presence of CW RFI Probability of Satellite Availability GIOVE-A 64ms GPS PRN-1 8ms GIOVE-A 8ms I/S (db) Figure 7-6 GPS and GIOVE-A probability of availability versus power of CW RFI 7.5 Summary In this paper, after looking at the structure of the GPS and Galileo signals, the effect of CW interference on these two signals is analysed. The difference of effects is first explained based on the difference between the two signal structures. Then the difference is investigated for different interference powers and frequencies. It is shown that the integration time which is a receiver parameter can affect the shape of C/No and is considered in the analyses. Worst spectral lines are evaluated for the both signals and the effect of CW RFI on a real Galileo signal collected by the NordNav RF front-end is examined. It is shown that for similar conditions, the drop in C/No in the presence of RFI located in the middle of the bandwidth is always larger than that of Galileo signal. Finally giving a definition of probability of availability of a satellite within its visible period, it is shown that a Galileo signal is more probable than GPS signal to be available in the presence of CW RFI close to L1 frequency and a GPS signal is more probable to be available in the presence of interference which is far from L1 frequency (for the two particular codes selected). This means that these two systems can be considered as alternatives to each other in the presence of different RFI frequencies as their availability in the presence of CW RFI is different in terms of RFI frequency. 145

165 Chapter 8 SUMMARY AND FUTURE WORK 8.1 Introduction This research was funded by Cooperative Research Center for Spatial Information (CRC-SI), Australia. The aims were to target problems that presently exist in CORS network site installation and the quality of the raw data derived from CORS networks concentrating on signal interference. The aim have been successfully achieved by introducing novel ideas and theories in the areas of RFI detection, characterization and mitigation. The section 8.2 of this chapter summarizes the content of this thesis. Section 8.3 lists some suggestions for the possible directions that the research in this thesis can be pursued. 8.2 Summary and Conclusions In this thesis, the problem of the effects of interference on GNSS systems is addressed. The focus is primarily on CW interference as it is shown to be the most dangerous type of interference to the performance of GNSS systems. Firstly, a locally optimal technique is proposed to detect the interference. A second approach uses the detected interference parameters to characterize the effect of the interference on the tracking part of GNSS receivers. Finally, based on this investigation an interference mitigation technique is proposed and shown to be effective in improving the position accuracy of the receiver. The theory of signal detection was used in the RFI detection algorithm in Chapter 4. In this algorithm, statistical inference is used to produce an optimal detection scheme to detect real interference. A signal free of interference is assumed to be the null hypothesis and an interfered signal is assumed to be the alternative hypothesis. The window of data assumed to be interference-free is called the assessment window and the window of data to be evaluated for the presence of interference is called the evaluation window. Using properties of the fast Fourier transform (FFT) and the effects of the number of FFT points on the test statistic mean and variance, it is shown that an appropriate data block size and window size can be selected to achieve the 146

166 Chapter 8 Summary and future work optimum solution for minimum detectable interference power. Based on how the data block is broken into sub-blocks, the detection threshold changes with a fixed probability of false alarm. There is an optimal way to break up this data window to achieve the lowest detection threshold for a fixed value of probability of false alarm in which the computational load is also minimal. The optimal detector algorithm is proposed. At this optimal point, for a fixed Detection Threshold, probability of false alarm becomes minimal and for a fixed probability of false alarm, we can achieve the minimum value for the detection threshold. Experiments show that at this point we have the minimum computational load. This theoretical result is supported by real experiments. Finally this algorithm is employed and tested to detect a real GPS interference signal generated by a TV transmitter in Sydney. Having detected the interference and estimated its frequency, it is possible to characterize its effect on the received signal quality. In Chapter 5, by looking at the GPS C/A code L1 signal structure and analyzing the effect of noise, signal and the interference on the carrier tracking loop separately, a closed-form formula is derived for C/No which is an indicator of the received signal quality. It is shown that the value of the C/No after the carrier tracking loop depends on different parameters like the powers of interference and the GPS signal, the height of the closest spectral line of the signal to the RFI frequency, the frequency difference between them and finally the integration duration time in the tracking loop. For pulsed CW, a similar analysis is performed to characterize the effect of parameters such as pulse repetition period (PRP) and duty cycle on the received signal quality. It is shown that for equal interference power levels, in the cases where the PRP is far less than the pseudorandom noise code period, then the signal degradation increases with increasing duty cycle, whereas it doesn t change if the two periods are equal or the PRP is far greater than the code period. These results show vulnerability to interference is highly variable. It is important to be aware of this in RFI monitoring and interference mitigation. It is not enough to limit the analysis to a discrete set of possible RFI signals. Another important result is that the mitigation techniques should take into account this high variability in the C/No degradation, which depends on Doppler frequency, so adaptive mitigation techniques are necessary, even if we know a priori the possible RFI signals that may occur in some specific applications. It is also shown in Chapter 5 that because different satellites have different timevarying Doppler frequencies, then the effect of CW interference on each satellite 147

167 Chapter 8 Summary and future work varies with time and is different for different satellites. This means that it is possible that in each constellation we have only one satellite affected by interference. This is the basis of the mitigation algorithm presented in Chapter 6. The Doppler frequency for each signal is predictable once the receiver position is known. In Chapter 6 the effect on different satellites is studied analytically. The concept of an exclusion zone is defined and analyzed for each satellite. This exclusion zone, where that satellite should not be used due to interference degradation, is shown to be predictable for each satellite as a function of time. Using this prediction, the CW interference effect on the positioning quality of the receiver can be mitigated by ignoring the affected satellites within exclusion zones when performing position evaluation. The threshold beyond which a satellite should be excluded is then derived by studying the mutual effects of the geometry and the signal quality of that satellite on the positioning quality. Receiver Autonomous Integrity Monitoring (RAIM) uses redundancy in measurements to perform an internal consistency check to see if all of the measurements are satisfactory. In Chapter 6 this technique is also used to mitigate the effect of CW interference on the positioning accuracy. Finally it is shown that the prediction of the exclusion zone for each satellite outperforms the RAIM algorithm in mitigation the effect of the interference when 5 satellites are visible. Unlike other mitigation techniques which are responsive, this technique works preventatively but it does require knowledge of the interference frequency and power, which may be estimated by known techniques plus the relationship between the C/No and the pseudorange error for the receiver. This means that it predicts and prevents the error due to interference before it happens. Finally giving a definition of probability of availability of a satellite within its visible period in the presence of CW RFI, it is shown that a Galileo signal is more probable than GPS signal to be available in the presence of CW RFI close to L1 frequency and a GPS signal is more probable to be available in the presence of interference which is far from L1 frequency (for the two particular codes selected). This means that these two systems can be considered as alternatives to each other in the presence of different RFI frequencies as their availability in the presence of CW RFI is different in terms of RFI frequency. 8.3 Future Work 148

168 Chapter 8 Summary and future work In this thesis and other previous works, whenever the effect of RFI has been investigated on each tracking loop (code or carrier), the other has been assumed to be perfectly locked without any error. In reality of course this is not the case. In the GNSS receiver the code and the carrier tracking loops are coupled with each other and it is possible to look at them as a multi-input multi-output (MIMO) control system. One direction from this thesis is to produce a general model, which looks at the interference as a disturbance to this process and based on the disturbance rejection performance of this system characterize the effect of the RFI on the overall performance of the receiver tracking section. In the calculation of the C/No in the presence of CW interference, the noise power has been evaluated as the summation of the thermal noise power and the noise equivalent power of the interference. In other words, the amount of interference that can pass through the matched filter is added to the existing noise and constitutes the total post-correlation noise in the receiver. This approach is performed in the presence of AGC in the RF front-end. However, the AGC does affect the signal and the RFI power that passes through the front-end bandwidth. This effect has not been considered in the calculation of the C/No. Characterization of the effect of AGC on the C/No is a task that can be done in the future. Having a more accurate estimate of C/No can also help in the mitigation technique presented in Chapter 6. In that chapter a threshold is defined for the signal quality below which the satellite should be eliminated from the positioning evaluation. That algorithm assumes a specific relation between the C/No and the pseudorange error for each receiver {satellite?}. If there is a closed-form relationship between the C/No and the pseudorange error then at each particular epoch and for each particular satellite the threshold can be found for its pseudorange error. The fact that at each particular time only one satellite might be affected by interference can also be used in other mitigation algorithms. For example, if using a notch filter to eliminate CW interference, the closed-form formula for C/No presented in Chapter 5, makes it possible to apply the notch filter only for the satellite which is affected by RFI. In this case it is possible to eliminate the unwanted side effect of this mitigation technique on the other satellites. In Chapter 7, using the characterization of the effect of CW RFI on the C/No, the satellite availability of two available signals (GPS and Galileo GEOVE-A) is investigated. In the future this investigation can be taken further to other signals such as L5, L2C, E5 and E6. 149

169 Chapter 8 Summary and future work The last but not the least thing to consider is the improvement of the hardware implementation of the RFI detection unit. In the prototype presented here, using the second generation of Namuru FPGA board, it is possible to have a GPS receiver in parallel to the detection scheme. In such an architecture, using the acquired data from the receiver and the information from the detection unit, the C/No for each channel can be predicted in advance. 150

170 REFERENCES ABA (2002) DRAFT LICENCE AREA PLAN Sydney Analog Television Australian Broadcasting Authority Canberra October 2002 Akos, D., (1997). A Software Radio Approach to Global Navigation Satellite System Receiver Design. Ohio University, Athens, OH. Altera (2006) FFT Megacore Function User Guide 6.1 December Altera (2006) Avalon Streaming Interface Specification 1.0 November Amoroso, F. (1983) Adaptive A/D Converter to Suppress CW Interference in DSPN Speard-Spectrum Communications. IEEE Transaction on Communications Bahl, P., Padmanabhan, V.N, (2000) Radar: An In-Building RF-Based User Location and Tracking System, Proc. IEEE INFOCOM 2000, Tel Aviv, Israel, vol. 2, Mar., pp Barnes, J., Rizos, C., Kanli, M., Pahwa, A., Small, D., Voigt, G., Gambale, N., & Lamance, J., (2005). High accuracy positioning using Locata's next generation technology. 18th Int. Tech. Meeting of the Satellite Division of the U.S. Institute of Navigation, Long Beach, California, September, Bastide, E., Macabiau, C., (2001) GPS Interference Detection And Identification Using Multicorrelator Receivers. ION GPS Bastide, F., Macabiau, C., Akos, D.M. (2003) Automatic Gain Control (AGC) as an Interference Assessment Tool. ION GPS/GNSS Bayes, T. An essay toward solving a problem in the doctrine of chances. Phil. Trans. Roy. Soc., vol. 53, pp , (Reprinted in Biometrika, vol. 45, pp , 1958.) Benson, D. (2005) The design and implementation of a GPS receiver channel An FPGA plus DSP creates a powerful combination GPS Magazine, October Betz, J.W., (2001) Effect of partial-band interference on receiver estimation of C/No: Theory, in Proc. of ION National Technical Meeting, Long Beach, CA, Jan. 2001, pp Betz, J.W. (2000) Effect of narrowband interference on gps code tracking accuracy. in in Proc. of ION National Technical Meeting, Anaheim, CA, Jan. 2000, pp. 16 Betz, J.W., Goldstein, D.B. (2002) Candidate designs for an additional civil signal in GPS spectral bands. Technical Papers, MITRE, Bedford, Mass, USA, January 151

171 References Betz, J.W. Bryan, L., Titus, M., (2004) Intersystem and Intrasystem Interference with Signal Imperfections IEEE Position Location and Navigation Symposium Bond, K., Brading, J. (2000) Location of GPS Interference Using Adaptive Antenna Technology. Proceedings of the ION GPS-2000, September, Borio, D., Lo Presti, L., Mulassano, P., (2006) Spectral Separation Coefficient for Digital GNSS receivers, European Signal Processing Conference (EUSIPCO), Florence, Italy - September 4-8, Borre, K., Akos, D.M. Bertelsen, N. (2007) A Software-Defined GPS and Galileo Receiver. A Single-Frequency Approach. Brown, A., Reynolds, D., Roberts, D., Serie, S. (1999) Jammer and Interference Location System-Design and Initial Test Results Proceedings of the ION GPS- September Brown, A., Atterberg, S., Gerein, N. (2000) Detection and location of GPS interference sources using digital receiver electronics NAVSYS Corporation Proceedings of ION Annual Meeting, June, San Diego, CA Brown, R.G., Hwang, P.Y. (1986) GPS failure detection by autonomous means within the cockpit. Proceedings of Institute of Navigation s Annual meeting (Seattle, WA), June 24-26, pp.5-12 Brown, A., Sturza, M. (1990) The effect of geometry on integrity monitoring performance. proceedings of the ION annual meeting, June 1990 Bucco, G., Trinkle, M., Gray, D., Cheuk, W. (2004) FPGA Implementation of a Single Channel GPS Interference Mitigation Algorithm Journal of Global Positioning Systems Vol. 3, No. 1-2: Bullock, J.B., King, T.M., Kennedy, H.L., Berry, E.D., Zanfino, G. (1997) Test results and analysis of a low cost core GPS receiver for time transfer applications Frequency Control Symposium, Proceedings of the IEEE International May Page(s): CAT I (2002) Local Area Augmentation System Ground Facility. U.S. Federal Aviation Administration, Washington, D.C., FAA-E-2937 Chang, H. (1982) Presampling filtering, sampling and quantization effects on digital matched filter performance. Proceedings of the international telemetering conference, (San Diego, CA), 1982, pp

172 References Chen, S.W. (1997) Linearity requirements for digital wireless communications Gallium Arsenide Integrated Circuit (GaAs IC) Symposium, Technical Digest, 19th Annual Oct. Page(s):29 32 Chow, Y.H. et al, (2006) A 1V, 0.9dB Noise-Figure High Linearity LNA MMIC for Concurrent GPS Handset application", Proc Asia Pacific Microwave Conference, Clynch J.R., Parker, A.A., Badger, G., Vincent, W.R., McGill, P., Adler (2003) The Hunt for RFI. GPS World January 1, Data Fusion Corporation (2004) Kent Krumvieda, GPS Program Manager Bannock Street Suite 246 Northglenn, CO USA Tel: X303 Fax: Dempster, G.A. (2006) Dilution of precision INS angle-of-arrival positioning systems. Electronics Letters, vol 42, no 5, pp , 2 Mar., Dempster A. G Aperture (2004) Jitter Effects in software radio GNSS receivers. Journal of Global Positioning Systems, 3(1/2):45-48, Dempster, G.A., Mumford, P., Parkinson, K., Qaisar, S., Wu, J., Tabatabaei Balaei, A., Glennon, E., (2007) NAMURU, a platform for GNSS research, Location April Devore, J.L. (1995) Probability and statistics for engineering and the sciences. Fourth edition by Wadsworth, Inc. Engel, F., Heiser, G., Mumford, P., Parkinson, K., Rizos, C. (2004) An Open GNSS Receiver Platform Architecture, The 2004 International Symposium on GNSS/GPS, 6-8 Dec. Sydney, Australia. Evans, A.G., Swift, E.R. Cunningham, J.P., Hill, R.W., Blewitt, G., Yunck, T.P., Lichten, SM., Hatch, RR., Malys, S., Bossler, J. (2002): The Global Positioning System Geodesy Odyssey, Navigation, 49(1), Fante, R.L., Vaccaro, J.J. (2000) Ensuring GPS availability in an interference environment Position Location and avigation Symposium, IEEE March Page(s):37 40 Federal Aviation Administration, Washington, DC, Advisory Circular C: Criteria for Approval of Category III Landing weather Minima, Mar Federal Aviation Administration, Satellite Navigation Program Office, Operational Requirements Document: Local Area Augmentation System, February

173 References Feller, W. (1971) An Introduction to Probability Theory and its Applications. John Wiley and Sons, NY, Figure Fisher, R.A., (1922) On the mathematical foundations of theoretical statistics. Phil. Trans. Roy. Soc., vol. A 222, pp , Gauss, C.F. (1823) Theoria Combinationis Observationum Erroribus Minimis Obnoxiae. G ottingen, Germany: Dieterich. Geyer, M., Winer, B., Frazier, R. (1997) Airborne GPS RFI Localization Algorithms, Proceedings of ION GPS-97, Kansas City, Sept , pp Godfrey O.O. (2003) Setting up of GPS Reference Stations and Investigating the Effects of Antenna Radome. Thesis submitted in partial fulfillment for the requirements of admission into the Ph.D. studies in Navigation, Institute of Navigation University of Stuttgart August, Goiser, A., Sust, M. (1989) Adaptive interference rejection in a digital direct sequence spread spectrum receiver. Military Communications Conference, Gromov, K. (1999) Interference Direction Finding for Aviation Applications of GPS, Proceedings of the ION GPS-99, September Gromov, K., (2000) GIDL: Generalized Interference Detection and Localization System. Proceedings of the ION GPS-2000, September. Gromov, K. (2002) GIDL; Generalized interference detection and localization system. PhD thesis, Stanford University Groves, P.D., (2005) GPS Signal-to-Noise Measurement in Weak Signal and High- Interference Environments. Journal of the Institute of Navigation. Vol. 52, No. 2, Page(s):83 94 Harris, J.M.; Wheeler, M.L.; Daher, J.K. (1994) Commercial GPS receiver susceptibility evaluation, Electromagnetic Compatibility. Symposium Record. Compatibility in the Loop. IEEE International Symposium on Aug Page(s):68 72 Helstrom, C.W. (1995) Elements of signal detection and estimation. Englewood Cliffs, NJ: Prentice-Hall, Heppe, S., Ward, P. (2003), RFI & Jamming and its Effects on GPS Receivers, based on communication theory, Navtech seminars course 452, Dahlgren, VA, May

174 References Hewlett Packard (1996) HP 8647A and 8648A/B/C/D Signal Generators Technical Specifications, Hoffman, J.R., Cotton, M.G., Achatz, R.J., Richard N., Statz, R., Dalke, A. (2001) Measurements to Determine Potential Interference to GPS Receivers from Ultrawideband Transmission Systems U.S. Department of commerce NTIA Report February Holmes, K.J. (2007) Spread Spectrum Systems for GNSS and wireless communications Artech House, INC. ICD1 Galileo Project Office of European Space Agency. GIOVE: Galileo In-orbit Validation Element. Retrieved March 6, 2007, from EESA: ICD2 GPS JOINT PROGRAM OFFICE. INTERFACE SPECIFICATION: IS-GPS- 200D. Retrieved Dec , from U.S. Coast Guard Navigation Canter of Excellence: Jacobson, L., Expert (2007) Advice Why we need eloran? GPS WORLD Mar. Jing, W., Moeness, A. (2007) Direction finding of GPS receiver interference based on the nulling weights Wireless Sensing and Processing II. Edited by Raghuveer, R., Dianat, S.A., Zoltowski, M.D. Proceedings of the SPIE, Volume 6577, pp (SPIE Homepage) 05/2007 Jan, S.S. Enge, P. (2001) Finding Source of Electromagnetic Interference (EMI) to GPS Using a Network Sensors in Proceedings of the ION National Technical Meeting NTM-2000, Institute of Navigation Jan John, A. (2001) Vulnerability assessment of the transportation infrastructure relying on the global positioning system Volpe Natinal Transportation system center August 2001, US DoT Julien, O., Cannon, E., Lachapelle, G., Mongrédien, C. (2004) A New Unambiguous BOC(n,n) signal Tracking Technique Source: Proceeding of The European Navigation Conference GNSS 2004,Rotterdam,17-19 May Kailath, T., Vincent, H. (1998) Poor Detection of Stochastic Processes. Ieee Transactions On Information Theory, Vol. 44, No. 6, October 1998 Karsi, M.F., Lindsey, W.C. (2000) Effects of CW interference on phase-locked loop performance Communications, IEEE Transactions on Volume 48, Issue 5, May Page(s): Kaplan E. (2005) Understanding GPS: Principles and Applications, Artech House. 155

175 References Kay, S.M. (1998) Fundamentals of Statistical Signal Processing, Volume 2: Detection Theory Prentice Hall Signal Processing 1st edition (January 27, 1998) Khan, F.A. (2007a) Behavior of the GPS Timing Receivers in the presence of Interference ION GNSS Khan, F.A., Dempster, A.G., (2007b) Interference Effects on GPS Timing Receivers and Their Impacts on Communications Networks Globecom Kim, N. (2006) Interference Effects on GPS Receivers in Weak Signal Environments. University of Calgary, Department of Geomatics Engineering Masters Thesis. Krizman, K.J., Biedka, T.E., Rappaport, T.S. (1997) Wireless Position Location: Fundamentals, Implementation Strategies, and Sources of Errors, In Proc. IEEE Vehicular Technology Conf. vol 2, pp , May Krim, H., Viberg M., (1996) Two decades of Array Signal Processing Research, IEEE Signal Processing Magazine July Landry, R.J., Renard, A. (1997) Analysis of potential interference sources and assessment of present solutions for GPS/GNSS receivers, 4th Saint-Petersburg on INS, May. Landry, R.J., Calmettes, V., Bousquet, M. Impact of interference on a generic PS receiver and assessment of mitigation techniques SUPAERO 10 AV. Edouard Belin Toulouse Cedex, France bousquet@supaero.fr Landry, R.J., Renard, A. (1997) Analysis of potential interference sources and assessment of present solutions for GPS/GNSS receivers, 4th Saint-Petersburg on INS, May Lawson, J., Uhlenbeck, G.E. Threshold Signals. New York: McGraw-Hill, Lee, Y.C. (1986) Analysis of range and position comparison methods as a means to provide GPS integrity in the user receiver. Proceedings of Institute of Navigation s Annual meeting (Seattle, WA), June 24-26, pp.1-4. Leung, S.C., DiEsposti, R., Giron, C., Weiss, I. (1997) Analysis of Algorithms for GPS Interferer Direction Finding. Proceedings of ION GPS-97, Kansas City, Sept , pp Lu, D., Feng, Q., Wu, R. (2006) Survey on Interference Mitigation via Adaptive Array Processing in GPS Progress In Electromagnetics Research Symposium, Cambridge, USA, March Luo, M., Xie, G., Akos, D., Pullen, S. and Enge, P. (2003). Radio Frequency Interference Validation Testing for LAAS using the Stanford Integrity Monitor 156

176 References Testbed. Proceedings of the Institute Of Navigation National Technical Meeting, Anaheim, CA. Jan Middleton, D., (1977) Statistical physical models of electromagnetic interference. IEEE Trans. Electromagn. Compat., vol. EMC-19, pp , Aug Marti, L., Van Grass, F., (2004) Interference Detection by Means of the Software Defined Radio ION GNSS 17 th International technical meeting of the satellite division, Maxim-IC APPLICATION NOTE 3447 Complete Stand-Alone GPS Receiver Solution with MAX2742 A complete GPS solution is achieved with minimal external components, using the MAX2742 integrated CMOS RF front-end GPS receiverhttp://pdfserv.maxim-ic.com/en/an/an3447.pdf. Marti, L., Van Grass, F., (2004) Interference Detection by Means of the Software Defined Radio ION GNSS 17 th International technical meeting of the satellite division. Marti, L., Van Grass, F., (2003) The effect of the radio frequency front-end onto signal estimation. Proceedings of the 24 th International technical meeting of the satellite division of the Institute of Navigation, ION GPS Marti, L., Van Grass, F., (2004) Interference Detection by Means of the Software Defined Radio. PhD Thesis, Ohio University. The BPSK spectrum is split into two symmetrical components with no power at the carrier frequency (Martin, 2003) Michels, J.H., Himed, B., Rangaswamy, M. (2004) Robust STAP detection in a dense signal airborne radar environment, Signal Processing, Volume 84, Issue 9 Pages: September Moelker, D., van Der Pol, T., Bar-Ness, Y. (1996) Adaptive antenna arrays for interference cancellation in GPS and GLONASS receivers, Proceeding of the IEEE 1996 Position Location and Navigation Symposium, , Motella, B., Tabatabaei Balaei, A., Lo Presti, L., Leonardi, M., Dempster, G.A. (2006) Characterization of Radar Interference Sources on Galileo E6 Band Submitted to the IEEE Transaction on Aerospace and Electronic systems, November Mumford, P., Tabatabaei Balaei, A., Dempster, G.A., (2007) UNSW GNSS Interference Detection Unit Proceedings of IGNSS, Australia, December Mumford P., Parkinson, K., Dempster, G.A. (2006) Open GNSS receiver platform. IEEE/ION PLANS, San Diego, California, April,

177 References Myrick, W., Zoltowski, M.D., Goldstein, J.S., (1999) Smoothing of space-time power minimization based preprocessor for GPS. Second Joint USA/Australia Workshop in Defense Application of Signal Processing (DASP99), Chicago, IL, Starved Rock State Park, , 22 26, August Myrick, W., Goldstein, J., Zoltowski, M.D. (2001) Low complexity anti-jam spacetime processing for GPS, Proceedings of the 2001 IEEE International Conference on Acoustics, Speech, and Signal Processing, Vol. 4, , Navman Ndili, A., Enge, P. (1998) GPS Receiver Autonomous Interference Detection. Proceedings of the Institute of Electrical and Electronics Engineers Position Location and Navigation Symposium, Palm Springs, CA. Ndili, A., (1998) Robust GPS autonomous signal quality monitoring. PhD thesis, Stanford University. Neyman, J., Pearson, E., (1933) On the problem of the most efficient tests of statistical hypotheses. Phil. Trans. Roy. Soc., vol. A 231, no. 9, pp , Nguyen, T.X., Ely, J.J., (2002) Determination of receiver susceptibility to radio frequency interference from portable electronic devices. NASA Langley Research Center Document ID: , NordNav Technologies AB, Stadsgården 10, S Stockholm, Sweden, Phone: , Fax Nyunook Kim Interference Effects on GPS Receivers in Weak Signal Environments University of Calgary, Department of Geomatics Engineering Masters Thesis. January 2006 Oshman, Y., Koifman, M. (2006) Robust navigation using the global positioning system in the presence of spoofing. Journal of guidance, control, and dynamics Vol. 29, no. 1, January Febrary Parkinson, B.W., Spilker J.J. (1996a) Global positioning system: Theory and Applications Volume I, American Institute of Aeronautics and Astronautics, Inc. Parkinson, B.W., Axelrad, P., Enge, P. (1996b) Global positioning system: Theory and Applications. Volume II Patwari, N., Ash, J.N., Kyperountas, S., Hero, A.O., Moses, R.L., Correal, N.S. (2005) Locating the Nodes [Cooperative Localization in Wireless Sensor Networks] IEEE signal Processing Magazine July 158

178 References Poor, H.V. (1994) An introduction to signal detection and estimation, Second Edition. New York: Springer-Verlag, Proakis, G.J., Salehi, M. Communication System Engineering, Prentice Hall, 1994 Rizos, C., Yan, T.S., Development of SydNET permanent real-time GPS network Journal of Global Positioning Systems (2004) Vol. 3, No. 1-2: Robinson, E., Quazi, A. (1985) Effect of sound-speed profile on differential timedelay estimation. Journal of Acoustical Society of America, vol. 77, no. 3, pp , Ross, J.T. et al., (2001) Effect of Partial-Band Interference on Receiver Estimation of C/N0: Measurements, Proceedings of ION 2001 National Technical Meeting, Institute of Navigation, January. RTCA/DO-208 (1991) Min. Operational Perf. Standards for Airborne Supplemental Navigation Equipment using Global Positioning System (GPS), July 12, Change 1 issues September 21, Sayed, A.H., Tarighat, A., Khajehnouri, N. (2005) Network-Based Wireless Location. IEEE signal Processing Magazine July Schneuwly, D., Sa O. (2002) Robust GPS-based synchronization of cdma mobile networks. 33rd annual precise time and time interval (PnZ) meeting BrCvards 16, CH-2002 NeuchQtel, Switzerland Sharawi, M., Akos D.M., Aloi D.N, (2007) GPS C/No Estimation in the Presence of Interference and Limited Quantization levels IEEE Transactions on Aerospace and Electronic Systems VOL. 43, NO. 1 JANUARY 2007 Shi, K., Zhou, Y., Kelleci, B., Wayne, T., Erchin, F., Ilker, A. (2007) Impacts of Narrowband Interference on OFDM-UWB Receivers: Analysis and Mitigation IEEE Transactions On Signal Processing, Vol. 55, NO. 3, MARCH 2007 Shnidman, D.A. (1995) Radar detection probabilities and their calculation, IEEE Trans. Aerosp. Electron. Syst., vol. 31, no. 3, pp , July. Shnidman, D.A. (2005) Comparison of low angle radar clutter models, IEEE Trans. Aerosp. Electron. Syst., vol. 41, no. 2, pp , Apr. Shnidman, D.A. (2005) Radar detection in clutter. IEEE Trans. Aerosp. Electron. Syst., vol. 41, no. 3, pp , July Soderstrand, M.A.; Johnson, T.G.; Strandberg, R.H.; Loomis, H.H., Jr.; Rangarao, K.V. (1997) Suppression of multiple narrow-band interference using real-time 159

179 References adaptive notch filters.; Circuits and Systems II: Analog and Digital Signal Processing, IEEE Transactions on [see also Circuits and Systems II: Express Briefs, IEEE Transactions on] Volume 44, Issue 3, March 1997 Page(s): Soderstrand, M.A.; Johnson, L.G.; Phillips, S.R. (2006) New Technique for Attenuation of Narrow-Band Interference. Applications in Control and Communications Systems; Signals, Systems and Computers, ACSSC '06. Fortieth Asilomar Conference on Oct.-Nov. Page(s): Spiegel, S. et al, (2003) Improving the Isolation of GPS Receivers for Integration with Wireless Communication Systems, Proc IEEE RFIC Symposium, pp Spilker J. and Natali F., (1996) Interference Effects and Mitigation Techniques, Chapter 20 of Global Positioning System: Theory and Applications, AIAA. Stewart, M. (2004) Enhancing Australia s core geodetic infrastructure. Cooperative Research Centre for Spatial Information (CRC-SI), Project Agreement 1.1, Stensby (2004) it is shown that the interference will always cause a periodic phase error, the period of which is related to the frequency difference between the signal and the interference. Tabatabaei Balaei, A., Wu, J., Dempster G.A., (2007f) Comparison Between GPS and Galileo Satellite Availability in the Presence of CW Interference Proceedings of IGNSS, Australia, December Tabatabaei Balaei A., Motella, B., Dempster, G.A., (2007e) GPS Interference Detected in Sydney- Australia Proceedings of IGNSS, Australia, December Tabatabaei Balaei, A., Motella, B., Dempster, G.A., (2007d) A Preventative Approach to Mitigating CW Interference in GPS Receivers accepted for publication in GPS Solution October Tabatabaei Balaei, A., Dempster, G.A., Lo Presti, L., (2007a) Characterization of the effects of CW and pulse CW interference on the GPS signal quality submitted to IEEE Transaction on Aerospace and Electronic systems March Tabatabaei Balaei, A., Motella, B. (2007b) Satellite exclusion zone in the presence of CW interference Experimental results ENC-GNSS May-June Tabatabaei Balaei, A., Motella, B., Dempster, G.A. (2007c) Mutual effects of satellite signal quality and satellite geometry on positioning quality ION/GNSS September, Fort Worth, USA 160

180 References Tabatabaei Balaei, A, Dempster, G.A., (2006e) A Statistical Inference Technique for GPS Interference Detection Submitted to the IEEE Transaction on Aerospace and Electronic systems, November Tabatabaei Balaei, A., Motella, B., Dempster, G.A. (2006d) Exclusion zones for GNSS signals when reconfiguring receiver hardware in the presence of narrowband RFI. IAIN / GNSS October, Korea Tabatabaei Balaei, A. (2006c) Statistical Inference Technique in Pre-Correlation Interference Detection in GPS Receivers. ION/GNSS September, Fort Worth, USA. Tabatabaei Balaei, A., Dempster, G.A., Barnes, J. (2006a) A novel approach in detection and characterization of CW interference of GPS signal using receiver estimation of CNo PLANS( ION IEEE)- April Tabatabaei Balaei, A., Dempster, G.A., Barnes J. (2006b) Application of Post- Correlation Interference Detection and Characterization of the GPS Receivers in the Receiver Reconfigurability IGNSS July Tabatabaei Balaei, A., Barnes, J., Dempster, A.G., (2005) Characterization of interference effects on gps signal carrier phase error SSC Melbourne Tomtom Tsui, J. (2000) Fundamentals of Global Positioning Receivers, A software receiver approach, John Wiley & Sons, Inc, 605 Third Avenue, New York, 2000, ISBN: Uren, J., Price, F.W. (2006) Surveying for engineers 4 th edition, chapter 7, PALGRAVE MACMILLAN Van Trees, H.L. (1968) Detection, Estimation and Modulation Theory Part I. New York: Wiley,. Van Dierendonck, A. J. GPS Receivers. In Parkinson, B.W., Spilker, J.J., Axelrad, P., Enge, P., (1996) Global Positioning System: Theory and Applications, vol. 1, American Institute for Aeronautics and Astronautics Vitzilaios, G., Papananos, Y., Theodoratos, G., Vryssas, K.S. (2006) Magnetic- Feedback-Based Pre-distortion Method for Low-Noise Amplifier Linearization IEEE Transactions On Circuits And Systems Vol. 53, No. 12, December 2006 Viterbi, A.J. (1966) Principles of Coherent Communication, McGraw-Hill, Inc. Ward, P. (2007) What s going on? RFI situational awareness inside GNSS

181 References Wegman, E.J., Schwartz, S.C., Thomas, J.B. (1989) Topics in Non-Gaussian Signal Processing. New York: Springer, Weiss, J.P., Anderson, S., Fenwick, C., Song, L., Axelrad, P., Belay, P., Brinkly, R., (2005) Development and validation of an aircraft multipath model for land-base JPALS, Proceedings of Institute of Navigation s Annual meeting Willett, P., Chen, B. (1999) Robust Detection of Small Stochastic Signals. IEEE Transactions on Aerospace and Electronic Systems Vol. 35. NO. 1 JANUARY Wohlfiel, J.E., Tanju, B., (1999) Location of GPS Interference, Proceedings of the ION GPS-99, September Wu, J., Dempster, G.A. (2007) Galileo GIOVE-A acquisition and trackinganalysis with a new unambiguous discriminator. Proceedings of IGNSS, Australia, December Yun, Y., Kee, C., Rife, J., Luo, M., Pullen, S., Enge, P. (2006) Detecting RFI Through Integrity Monitoring at a DGPS Reference Station THE JOURNAL OF NAVIGATION 59, p The Royal Institute of Navigation Printed in the United Kingdom 162

182 Appendix A HARDWARE IMPLEMENTATION A.1 Hardware The hardware platform is the Namuru field programmable gate array (FPGA) GNSS receiver board. The Namuru board includes an L1 band RF front-end, Altera Cyclone FPGA chip, memory, various I/O options including serial ports and Ethernet socket, and other support devices. Details about the board can be found in (Mumford, 2006) (see Figure A-1) Figure A-1 Namuru circuit board The RF front-end amplifies, filters, down-converts and band-pass samples the incoming signal. It passes the sampled intermediate frequency (IF) to the FPGA for digital processing as two-bit, sign and magnitude values. This process is shown in the frequency domain in Figure A-2. In this application, the GNSS receiver may be disabled and/or not uploaded into the FPGA. The intention at this stage is just to get hold of the L1 band using the board s front end and process that signal as per Chapter

183 Appendix A Hardware Implementation Figure A-2. RF down-conversion and band-pass sampling diagram Coordinating all this activity is a number of finite state machines these in turn are controlled by the state of registers that can be set by software and appear as a memory-mapped peripheral to the Nios processor. The DMA is responsible for making the data available to be processed in software, and provides a set of memorymapped registers for status and control; see (Altera, 2006). Figure A-3 provides a detailed block diagram of the system. The FFT block receives data in a 2048 sample (real only) sequence, then takes some cycles to process, and finally outputs 2048 bin (real and imaginary) values in sequence, along with a scale factor. The Altera Avalon Streaming Interface protocol defines how to control data flow in and out of the FFT block. More information on this protocol can be found at (Altera, 2006). The FFT scale factor presents a problem, as while it varies only a little between FFT processes in the normal mode, it varies a lot in the change to zoom mode and needs to be taken into account. A trade-off between dealing with scaling in hardware, and adjusting scaling from software has been found to keep the bit width of the resulting data to a suitable size. In the present design, the values coming out of the final math block (the square-root) and going into the DMA is 8 bits. This is a very convenient size, and makes for an easy and efficient implementation of the DMA data transfer. 164

184 Appendix A Hardware Implementation dump counter carrier mixer accumulator sample counter sign& mag buff c arrier 2 s comp table NCO switch FIFO buffer FFT sink side Contro l FSM s NIOS core memory FFT source side real imag exp scaling real mult im ag mult buff buff ` add buff sqrt DMA Figure A-3. Detailed hardware block diagram An example of output from a 512 bin Altera FFT block is provided in Figure A-4. Here a MHz sine-wave signal at -80dBm is injected into the RF front-end, sampled and fed into the FFT block as a 2 s compliment real value. The output is a positive and negative frequency sweep from DC to half the sample rate. In the prototype hardware implementation, a 2048 point FFT block is used, but only the first half of the output data is used, providing magnitude values over 1024 frequency bins. The second half of the output is close to a mirror image and for our purpose is redundant. All RF down conversion steps, sampling and FPGA digital processes are driven by one TCXO crystal oscillator running at 10MHz. Ultimately, all frequency determinations are based on this reference clock, and the accuracy, stability and possible calibration of this clock determine the absolute accuracy of the measurements made by the system. Having said this, it must be noted that the 165

185 Appendix A Hardware Implementation oscillator used on the Namuru board is a quality Rakon TX0215BR TCXO with an overall accuracy estimated at 2ppm or better. So why do so much processing in hardware? The first reason is that the Altera function blocks are tested and easily available, and can fit within the available FPGA chip space. The second reason is to allow the Nios processor to operate on the higher levels of the system, without being burdened with time-consuming DSP functions such as the FFT. This also makes software development much quicker and easier. Finally, the Nios processor potentially has enough spare capacity for communication to users, through serial ports, a display such as an LCD screen, or via the internet through a TCP/IP software stack and the on-board Ethernet connector. 5 x p42-80dBm.txt average power bin Figure A-4. Output of 512 bin FFT engine with L1 center frequency input A.2 Software Software for the prototype system was written in C using the Altera NiosII IDE. The functions the software performs are; initialise hardware, control DMA transfers, collect and analyse data, control zoom processing and report on results. A breif overview of these activities is provided in this section. The hardware is attached to the Nios core as a memory-mapped peripheral, status is observed by reading registers and hardware controlled by writing to registers. The DMA, control state machines and serial ports all require initialisation. After this, the 166

186 Appendix A Hardware Implementation software goes into a looping sequence of data collection, statistical testing and deciding if a switch to zoom processing is required. Zoom processing follows a similar sequence, and can fall back into normal processing if statistical tests fail over a number of trials. Various data from each processing sequence can optionally be logged to an external computer via an RS232 serial link for further off-line processing or viewing. Because most of the hard work is performed in hardware, the software is relatively straight-forward and small. The most critical part was found to be setting up and controlling DMA functions. The software is still in the development stage and still requires refinement and expansion before evolving towards release status. Items on the list for inclusion are; on-board display and user interface, watchdog timer and reset system to reboot if crashes occur, ethernet TCP/IP stack for on-line status reporting and frequency calibration capability. A.3 Experiments The system was tested to determine performance in three areas. The first was the detection of a continuous sine-wave in the L1 band to confirm correct overall system functioning and an initial indication of the performance of the statistical method used. The metrics of this test include an estimate of the sine-wave frequency and the t-value used in the hypothesis testing. The second test was to confirm the resolution of the zoom frequency determination system. The third test was to assess the operation of the device in the field, in an area suspected of having occasional interference. The first two tests had the following setup; a Hewlett Packard 8648B RF signal generator was used to provide a sine-wave at a known frequency and level. The specifications for this generator for the L1 band are; frequency accuracy +/- 4.7kHz, frequency resolution to 0.001Hz, power level accuracy +/- 1dB, within one year of calibration, see (Hewlett Packard, 1996). The internal oscillator is rated as +/- 2ppm/year. Unfortunately, the calibration status of this unit is unknown, and it cannot be assumed to be within specification. This signal is injected into the antenna connection of the Namuru s RF front-end. The signal is down-converted and sampled into the FPGA chip where processing is performed. Software running on a Nios core on the FPGA communicates via a JTAG serial interface to a console window 167

187 Appendix A Hardware Implementation providing status information, and also data can be logged to file via an RS232 serial link for further analysis. Some results from testing for the detection of a sine-wave in the L1 band and frequency determination using the zoom technique at zoom 10 is provided in Table A-1 and Table A-2. The threshold for detection was set at 95% confidence level, providing a t-value threshold of In Table A-1, the RF input level from the HP signal generator was set at -100dBm, and Table A-2 the input was set at -110 dbm. It can be seen that -110dBm is close to the limit of detection at 95% confidence for this system. It can also be seen from the tables that there is a near constant frequency offset between the frequency set on the HP signal generator and the reported frequency determination from the zoom process. This offset is most likely due to the differences in reference clock frequencies in the two devices (more likely the receiver provided that the signal generator is calibrated), but also may include other factors. Further investigation is needed in this regard. Table A-1 Detection of 100dBm sine zoom 10 Input frequency -100dBm (MHz) Detection t-value Zoom 10 frequency determination Delta (Hz) (MHz) Table A-1 reveals a good correlation between the frequency set on the HP RF generator and the determined frequency, with an average delta of 3794Hz with standard deviation 98Hz. The t-values are all well above the 95% confidence threshold. 168

188 Appendix A Hardware Implementation Table A-2 also reveals good correlation, with average delta of 3754Hz and standard deviation 50Hz. Most of the t-values clear the 95% confidence threshold convincingly, except for the L1 center frequency that clears the threshold by just It is clear from this that detection would begin to fail at signal power levels lower than this at the 95% confidence level. Sensitivity could be improved by using a larger FFT block with more bins, FPGA space permitting. Table A-2. Detection of 110dBm sine zoom 10 Input frequency -110dBm (MHz) Detection t-value Zoom 10 frequency determination Delta (Hz) (MHz)

189 Appendix A Hardware Implementation Table A-3. Detection of -110dBm zoom 100 Input frequency -110dBm (MHz) Detection t-value Zoom 100 frequency Delta (Hz) determination (MHz) Table A-3 shows improved frequency correlation with an average delta of 3299Hz and standard deviation of 9.9Hz. Higher resolution for frequency estimation is valuable if the mitigation algorithm presented in Chapter 6 is to be applied after the detection of CW RFI. In Chapter 5 and Chapter 6 it was shown that the width of troughs in the C/No depends on the integration duration time. This width was the basis of the mitigation scheme of exclusion zone. To apply this algorithm after the detection block in this hardware, a high resolution in the estimation of the frequency is needed. A resolution of 10Hz can be acceptable for the case in which the width of the trough is 100Hz. This corresponds to an integration duration time of 10ms. It is always possible to have better resolution with higher zoom with the cost od computational load. The t-values in this case are similar in Table A-2 and Table A-3 showing a marked dip in the center. 170

190 Appendix A Hardware Implementation Table A-4. Frequency resolution at zoom 100 Input frequency -110dBm (Hz) Zoom 100 frequency (f1) determination (Hz) Diff Adjusted frequency (Hz) (f1 3299) During normal operation, the FFT covers the digitized IF range from DC to 2.8MHz. Each of the 1024 bins have a frequency range of about 2790Hz. To zoom in on an interfering signal found in a particular bin, the zoom level should cover the range of the bin to avoid ambiguities. At a zoom level of 10, the FFT covers a range of about 285kHz, with each bin being 279Hz wide, providing a potential frequency resolution to 279Hz. The overall requirement for proactive GPS interference mitigation as proposed in section 2 is for frequency resolution down to around 10Hz, a zoom level of 300 is the minimum to achieve this with a 1024 bin FFT, giving a range of 9523Hz and 9.3Hz per bin. Unfortunately, with 300 accumulations a problem can occur where the input to the FFT overflows producing errors in the system. The alternate method of decimation solves this problem and in addition reduces the scaling range significantly, removing the need to monitor and adjust the scaling range settings. A zoom of around 100 appears to be the upper limit to avoid overflow with this method and an 8-bit FFT. At a zoom level of 100 the FFT range is 28.5kHz with bin width of 27.9Hz. Results from the second test are given in Table A-4. This table reveals the frequency resolution with zoom level 100, showing frequency jumps of 27 and 28Hz for actual frequency changes of 30Hz in 10Hz steps. Table A-4 also shows the frequency adjusted to account for the offset calculated from Table A-3. It is evident that the system can work quite well and could provide accurate frequency 171

191 Appendix A Hardware Implementation determination once the internal frequency reference (either through the hardware by adjusting the receiver crystal or through the software) is calibrated. Spectrum Analyzer TV1 3RD Ref Level : -10 M1: MHz dbm db / Div : db M Frequency ( MHz) CF: MHz SPAN: MHz Attenuation: 30 db RBW: 30 khz VBW: 1 khz Detection: Pos. Peak Std: Min Sweep Time: 1.00 Milli Sec Date: 06/07/2007 Time: 04:28:06 Model: MS2711D Serial #: dbm Figure A-5. Frequency sweep around MHz near the Artarmon transmitter tower The third test was a trial to evaluate the prototype system in the field. The area around the television and radio broadcast tower at Artarmon in the north of Sydney was chosen. Previous work by (Tabatabaei, 2006) identified this site as a potential source of interference due to the third harmonic of a broadcast signal falling near the L1 band. There is a signal broadcast from the tower centred around MHz. Figure A-5 shows a spectrum analyser sweep around this frequency. The third harmonic of this broadcast frequency would lie around MHz, just outside the GPS L1 signal bandwidth of 2MHz centred at MHz. The prototype system was setup and run at a number of sites near the tower. A console window (see Figure A-6) provided status information from the system, updated every few seconds. In Figure A-6, the results of statistical processing including the maximum t-value and the frequency of the associated FFT bin can be seen over three cycles. Bin 533 keeps coming up as having the maximum t-value of around 0.57, with a frequency range of to MHz. The t-value threshold is set according to the level of confidence required in declaring the existence of interference. For a confidence level of 95% the t-value threshold is 172

192 Appendix A Hardware Implementation 1.64, clearly the obtained t-values are well under this level. The t test is used to trigger the switching to zoom processing to determine the frequency of the suspected interference signal to a higher resolution. The screen shot in Figure A-7 reveals processing when the confidence level is relaxed to investigate bin 533 in more detail. Figure A-7 shows the transition to zoom processing, with the maximum power level found in bin 55 or 56 repeatedly over a trial of several minutes. This relates to a frequency of MHz, and removing the constant offset calculated previously of 3299Hz gives MHz. Figure A-6. Example of statistical result output to a console window 173

193 Appendix A Hardware Implementation Figure A-7. Example of zoom processing to console window While this test is inconclusive, and certainly does not declare the existence of interference at a high level of confidence it does provide an example of the processing flow employed by the prototype system. The test was also performed in an area removed from known transmitters and maximum t-values (and associated bin number) over many trials where observed. There appeared to be no trend. Over a 50 trial sample the average t-value was found to be 0.39 with a standard deviation of