Adaptive Array Processing for GPS Interference Rejection

Transcription

1 Adaptive Array Processig for GPS Iterferece Rejectio David S. De Lorezo, Staford Uiversity Jeifer Gautier, Staford Uiversity Jaso Rife, Staford Uiversity Per Ege, Staford Uiversity Deis Akos, Uiversity of Colorado at Boulder BIOGRAPHY David De Lorezo is a member of the Staford Uiversity GPS Laboratory, where he is pursuig a Ph.D. degree i Aeroautics ad Astroautics. He received a Master of Sciece degree i Mechaical Egieerig from the Uiversity of Califoria, Davis, i David has worked previously for Lockheed Marti ad for the Itel Corporatio. Dr. Jeifer Gautier is curretly the Istitute of Navigatio Executive Fellow for Previously, Dr. Gautier was a Research Associate i the GPS Laboratory at Staford Uiversity, where she led the Lab s research program for the Joit Precisio ad Approach Ladig System (JPALS). She received the Bachelor s degree i Aerospace Egieerig from Georgia Tech ad completed the Master s ad Ph.D. degrees i Aeroautics ad Astroautics at Staford Uiversity. Dr. Gautier has worked for Lockheed, Hoeywell Labs, ad Trimble Navigatio, Ltd. Dr. Jaso Rife is a Research Associate studyig JPALS ad the Local Area Augmetatio System (LAAS) at Staford Uiversity. After receivig his B.S. i mechaical ad aerospace egieerig from Corell Uiversity (1996), he spet oe year workig i the turbie aerodyamics group of the commercial egie divisio of Pratt & Whitey. He resumed his studies at Staford ad eared M.S. (1999) ad Ph.D. (2004) degrees i mechaical egieerig. His thesis work focused o positioig ad avigatio techologies for uderwater robots. Dr. Per Ege is a Professor of Aeroautics ad Astroautics at Staford Uiversity, where he is the Kleier-Perkis, Mayfield, Sequoia Capital Professor i the School of Egieerig. He directs the GPS Research Laboratory, which develops satellite avigatio systems based o the Global Positioig System. Dr. Ege has received the Kepler, Thurlow, ad Burka Awards from the Istitute of Navigatio for his work. He is a Fellow of the Istitute of Navigatio ad the Istitute of Electrical ad Electroics Egieers. Dr. Deis M. Akos is a Assistat Professor with the Aerospace Egieerig Sciece Departmet at the Uiversity of Colorado at Boulder. He also has served as a faculty member with the Luleå Techical Uiversity, Swede, ad as a Research Associate i the GPS Laboratory at Staford Uiversity. Dr. Akos completed the Ph.D. degree i Electrical Egieerig at Ohio Uiversity withi the Avioics Egieerig Ceter. ABSTRACT Oe area of active iterest i the GNSS commuity is determiig the vulerability of multi-atea sigal processig algorithms give calibratio errors i ateas ad aalog frot-ed hardware. Errors itroduced by the ateas or by the aalog hardware that are ot completely mitigated by software or calibratio may lead to degradatio i steerig-vector accuracy. Of iterest are the characterizatio ad mitigatio of ay cosequet reductio i C/N 0 ad biasig of code ad carrier phase estimates. Whe a referece sigal is available (as is the case for GPS) ad whe there may be degradatio i steerig-vector accuracy (due to hardware errors), the a adaptive mea-square-error based approach may have sigificat C/N 0 advatages. Further, exercisig both spatial ad temporal degrees of freedom is desirable i maximizig C/N 0 for adaptive atea arrays. Staford Uiversity s multi-chael hardware testbed demostrates the feasibility of implemetig adaptive array processig for GPS sigals i real time. INTRODUCTION The Joit Precisio ad Approach Ladig System (JPALS) is a Uited States Navy ad Air Force program to provide local-area augmetatio to the o-board GPS avigatio solutio for pilots o approach to aircraft

2 carrier, fixed base, ad tactical airfields. Sea-based JPALS provides carrier-phase differetial avigatio ad as such, there is the requiremet to maitai highitegrity GPS measuremets while rejectig iterferece ad multipath. There are a broad rage of iterferece mitigatio techiques described i the literature, icludig atea desig chages to suppress low-elevatio sigals, filterig ad sigal processig to reduce iterferece power, ad chages to the trackig loops icludig augmetatio by iertial measuremets [1]. Oe of the techologies beig studied for JPALS is a cotrolled receptio patter atea (CRPA) array. The baselie JPALS architecture ivolves a multi-elemet atea array to icrease available sigal-to-iterfereceplus-oise ratio (SINR); a additioal compoet likely would be adaptive beamformig ad ullsteerig to further improve iterferece rejectio. The sub-meter accuracy required for carrier ladigs ecessitates limitatios o avigatio errors due to sigal distortio, chael-to-chael timig differeces, ad low carrier-tooise ratio (C/N 0 ). No-adaptive spatial-oly beamformig, combied with kowledge of platform attitude ad satellite costellatio ephemeris, is suitable for icreasig the C/N 0 of desired satellite sigals. The drawbacks to this approach are high sidelobes, the requiremet at all times for accurate kowledge of platform orietatio ad satellite ephemeris, ad the susceptibility to atea ad aalog frot-ed timig ad distortio errors. However, with suitable iterferece detectio ad localizatio processig, a ullsteerig array ca sigificatly reduce sidelobe gai i the directio of iterferece sources. Adaptive processig for oise-rejectio or powermiimizatio allows the automatic suppressio of arrowbad iterferece ad jammig, ad itroduces o further frequecy-depedet phase distortio o the spread-spectrum GPS ragig sigal [2-3]. Greater iterferece rejectio, particularly of widebad sources, may be realized by icorporatig temporal filterig i the array processig e.g., with a tapped delay lie atea array. However, addig time taps to allow space-time adaptive processig (STAP) yields a fiite impulse respose filter that may distort the spread-spectrum GPS ragig sigal. The topic of multi-elemet steered beam ad adaptive atea arrays has received attetio withi the GPS commuity, icludig its impacts upo the carrier ad code phase observables [e.g. 4-12]. Errors i the calibratio of ateas ad frot-ed compoets ca itroduce additioal, ucompesated phase ad group delay o the icomig satellite sigals. The resultig impact to steerig-vector accuracy, effective sigal-tooise ratio, ad code ad carrier phase estimatio errors has ot bee adequately examied withi the cotext of adaptive atea arrays. This paper quatifies the effects to sigal-to-oise ratio due to calibratio errors i the ateas ad frot-ed compoets, ad explores the sesitivity to spatial ad temporal degrees of freedom withi the array. I this paper, it is show that whe a referece sigal is available (as is the case for GPS) ad whe there may be degradatio i steerig-vector accuracy (due to hardware errors), the a algorithm that miimizes the measquare-error betwee the actual array output ad the ideal array output may have sigificat C/N 0 advatages. Further, it is show that exercisig both spatial ad temporal degrees of freedom is desirable i maximizig C/N 0. To verify that the methods that are developed ad tested i simulatio are applicable to real-world scearios, Staford Uiversity has pursued costructio of a datacollectio hardware testbed. This testbed is based o iexpesive, commercially available frot-ed compoets ad high-speed/high-resolutio aalog-todigital coverter cards. This testbed allows collectio of high-quality sigal-plus-jammig data records for subsequet playback through a variety of software tools, trackig loops, ad specialized software avigatio receivers. ADAPTIVE ALGORITHMS FOR GPS The followig sectio presets a overview of adaptive algorithms, particularly as they apply to GPS sigal processig. This brief review motivates the remaider of the discussio o atea ad aalog frot-ed calibratio errors, ad spatial ad temporal degrees of freedom. For a determiistic atea array with maximum gai achieved for a particular icomig sigal vector, weights are calculated give kowledge of atea baselies ad icomig sigal azimuth/elevatio (ad atea gai/phase respose if available). The resultig array gai patter yields maximum costructive iterferece i the desired directio, but the sidelobes/ulls of the array are ucotrolled ad cosequetly may ot adequately suppress radio frequecy iterferece (RFI) or multipath. The array output sigal x(t) is formed as the product of the weight vector W times the sigal vector X(t): x ( t) W T X(t = ) (Eq. 1) I cotrast to the determiistic array, the cetral cocept of a adaptive atea array is the use of feedback to optimize some performace idex (Figure 1). The optimizatio criteria ca be broadly classified as either

3 Gai ad Phase Adjustmet Feedback Cotrol Σ Performace Idex Array Output optimized to balace rejectio of RFI with suppressio of ucorrelated oise received o each atea chael. I other words, if the RFI power goes up, the ull depth icreases, while if white Gaussia oise (WGN) power goes up, the sidelobe levels go dow. I short, RFI that is icidet o the array is correlated across atea elemets, ledig structure to the covariace matrix this structure is suppressed by the adaptive algorithm. As update equatios, the algorithms are: Figure 1. Geeric adaptive atea array. maximizig the sigal-to-iterferece-plus-oise ratio (SINR) at the array output [e.g ] or miimizig the mea-square-error (MSE) betwee the actual array output ad the ideal array output [e.g. 15; this is termed a leastmea-square (LMS) approach]. I either case, the array adapts to maximize the desired sigal ad to reject iterferece. The optimal steady-state weight vector satisfies the Wieer solutio: 1 W = μ Φ T 1 W = Φ S Applebaum Widrow LMS (Eq. 2) This otatio, ad that used heceforth i this paper, comes from Compto, Here T * is the array steerig vector ad S is the referece correlatio vector. Also, ote that solutio of the equatios i this form requires estimatio ad the iversio of the sigal covariace matrix Φ, which is defied as the expected value of X * X T. Estimatio requires sigificat time-averagig (reducig the ability of the array to adapt quickly to chagig sigal eviromets) ad sigal bufferig capacity, while iversio requires computatioal complexity. However, this is a approach that has bee employed successfully i adaptive beamformig/ullsteerig GPS architectures [e.g. 5, 8]. I recursive form, these algorithms look like: + 1 [ I γφ ] W + γμt Applebaum [ I γφ ] W + γs Widrow LMS W + 1 = (Eq. 3) W = With this formulatio, estimatio of Φ may be doe at each sample epoch. This form of the weight solutio requires o bufferig or matrix iversio, ad with suitable pre-coditioig, it adapts quickly to a chagig sigal eviromet. Now, i additio to the composite sigal output calculatio from Eq. 1, there is the simple calculatio of Φ at each sample epoch (ad the calculatio of S for LMS). This also brigs up a iterestig characteristic of the steady-state gai patter that results from applicatio of a adaptive algorithm: the depths of the patter ulls ad the levels of sidelobes are W + 1 = W + ΔW ΔW = γ [ μt ΦW ] Applebaum (Eq. 4) ΔW = γ [ S Φ W ] Widrow LMS This last form is actually the most istructive the algorithms reach steady-state whe the bracketed terms i the ΔW equatios are zero, i.e., whe the weight vector suppresses from the covariace matrix everythig but the steerig or referece vector. Of course, sice oly a estimate of the covariace matrix Φ is available, this is a approximate solutio. It is the misadjustmet parameter γ (equivalet to 2μ i the treatmet of Widrow ad Stears, 1985) that cotrols covergece speed ad, as the ame implies, steady-state misadjustmet. For a covetioal determiistic beamsteerig atea array without sigal feedback or adaptatio, the required iputs to the weight cotrol algorithm are sigal arrival directio, platform orietatio, ad baselie geometry or array maifold (Figure 2). The Applebaum array requires a additioal measure or estimate of sigal covariace allowig it to cotrol sidelobes, steer ulls, ad achieve rejectio of RFI. The LMS array requires sigal covariace as well as a ideal array output sigal; the LMS algorithm steers the array respose to alig the output sigal with a locally-geerated referece sigal. For the case of adaptive array processig for GPS, this ideal array output sigal is derived from the code ad carrier trackig loops. Note that there does exist a ambiguity if avigatio data are ecoded oto the icomig sigal; however, this ca be addressed either by bufferig of the avigatio message, by suitable logic i the weight cotrol algorithm (moitorig the polarity of the weight at the referece atea tap), or by ruig i parallel two weight computatios, oe each for +1 ad -1 data bit ecodig. ADAPTIVE ARRAY SENSITIVITY STUDY From the previous discussio, it becomes apparet that the Applebaum array is well-suited to situatios of kow sigal arrival directio but ukow sigal waveform, whereas the LMS array is appropriate for ukow sigal arrival directio but kow (or correlated) referece waveform. But what happes whe there are errors itroduced by the hardware that are ot completely

4 covariace estimate IMU attitude Weight Cotrol Algorithm replicated sigal Σ Trackig Loops ephemeris PVT code replica carrier replica Figure 2. Iputs to weight cotrol algorithm. mitigated by software or calibratio? These errors degrade the C/N 0 performace of steerig-vector-based array schemes; however, LMS-based adaptatio is robust to these effects. To this ed, a simulatio study was performed to ivestigate the sesitivity of adaptive algorithms to varyig evirometal coditios. Specifically, C/N 0 was assessed for differet cases, icludig errors ad imperfectios i the atea ad frot-ed hardware, errors i estimatig the directio of arrival of the icomig sigal, ad the presece of RFI sources (Figure 3). This study was doe for a sigle-elemet fixed receptio patter atea (FRPA), as well as for a 7- atea array i which sigals were combied either by a determiistic cotrolled receptio patter atea (CRPA) beamformig solutio, by a steerig-vector approach (Applebaum array), or by a algorithm that miimizes mea-square-error betwee the array output ad a referece (LMS array). Te satellites were placed radomly i the sky, represetig C/A-code PRN sequece umbers from 1 to 10, each trasmittig with a C/N 0 compared to backgroud WGN of 50 db-hz; for this simulatio, badpass samplig was employed rather tha mixig ad dowcoversio to a itermediate frequecy. The bars show i the accompayig illustratio show the effective C/N 0 as reported by the GPS receiver carrier trackig loop for the previously metioed sigal combiatio methods (Figure 4). The text labels uder each group of C/N 0 bars idicate which simulatio sceario is represeted: base case; atea phaseceter effects; phase-ceter, frot-ed, ad satellite arrival errors; RFI sources preset; ad the worst-case combiatio with all degradatio preset. The superimposed vertical lie o each bar graph represets ± oe stadard deviatio from the mea. Not oly is a higher C/N 0 desirable, but also a smaller stadard deviatio; the smaller the stadard deviatio, the lower the likelihood that oe or more satellite sigals will fall below the trackig threshold. For the case of perfect kowledge of sigal arrival directio ad array orietatio, o iterferece sources, ad isotropic ateas ad matched aalog frot-eds, it ca be see that the determiistic CRPA ad the Applebaum adaptive array achieve a similar improvemet i C/N 0 versus the sigle-elemet FRPA; this is to be expected, sice the Applebaum array i this case uses as a steerig vector the weights calculated for the determiistic CRPA, ad the white, ucorrelated ature of the icidet oise requires o additioal adaptatio to Figure 3. Satellite ad RFI geometry for sesitivity study.

5 60 Effective CN0 vs. Sigal & Adaptatio Parameters 55 C/N0 Effective (db-hz) CN FRPA Det. CPRA Applebaum LMS base dtpc dtpc,dtfe jammig worst Sceario base Δ pc Δ pc Δ fe RFI worst-case Sceario Figure 4. Adaptive array sesitivity study results. achieve maximal SNR. There is, however, a slight pealty for the LMS array, due to gradiet oise causig the coverged solutio to vary slightly from the ideal Wieer solutio (which is equivalet to the CRPA/Applebaum weight vector solutio). Next, there is the iclusio of motio of the atea phase-ceters as a fuctio of icomig sigal arrival directio, as determied by aechoic chamber testig ad computer simulatio [16]. The rage of advace/delay values applied to the icomig sigals was ±150 of a L1 carrier cycle, ad idetical across all ateas. This phase-ceter motio causes a reductio i the C/N 0 for the CPRA ad Applebaum arrays, sice the steerig/weight vector o loger matches the actual set of weights that would deliver maximal SNR; i other words, the effective maibeam of the array does ot coicide exactly with the actual desired icomig sigal vector. The LMS array suffers o such C/N 0 degradatio, as log as trackig is maitaied to provide a referece waveform, sice the algorithm uses i its adaptatio oly the desired output from the ceter, referece atea elemet. Note that this discussio oly applies to C/N 0 ad sigal trackig performace; biases itroduced by atea phase-ceter or group-delay characteristics (or aalog frot-ed effects) also ca cause code/carrier biases that would itroduce errors i the avigatio solutio ad impact the ability to successfully determie itegers for carrier-phase differetial RTK. Whe a additioal frot-ed differetial timig delay is itroduced for each atea sigal (equivalet to 0.25 wavelegths at L1, RMS with respect to the referece atea elemet) as well as a error i the kowledge of sigal arrival directio (2.5 RMS error i azimuth ad elevatio, equivalet to a equal error i platform attitude kowledge), there is further reductio i C/N 0 values. At this poit, the CRPA ad Applebaum arrays uderperform the sigle-atea FRPA system, while agai the LMS array suffers little C/N 0 degradatio. The ext sceario itroduces six swept-cw RFI sigals o or ear the horizo, ad each at a power level 10 db higher tha the received satellite sigals, but with o other errors preset i the simulatio. As a ote, it should be metioed that the CW sigals were swept i frequecy to avoid possible lock-o from the receiver carrier trackig loop; the frequecy ramp was set radomly to ±25 khz/msec. Compared to the C/N 0 degradatio see for the FRPA, the CRPA, Applebaum, ad LMS arrays all perform equally well. Fially, the last case icludes ot oly the six CW RFI sources, but also the atea phase-ceter, frot-ed

6 Correlatio Fuctio vs. Code Phase - PRN #22 Iphase Quadrature Total Code Offset (chips) Correlatio Fuctio vs. Code Phase - PRN #22 Iphase Quadrature Total Code Offset (chips) Correlatio Fuctio vs. Code Phase - PRN #22 Iphase Quadrature Total Code Offset (chips) delay, ad sigal arrival errors described above. As expected, the CRPA ad Applebaum performaces agai fall well below that of the FRPA. The LMS array suffers o further C/N 0 degradatio tha for the case with RFI preset oly. For this sceario, the average performace of the LMS array exceeds that of the CRPA or Applebaum array by at least 10 db-hz. The coclusios from this simulatio study are twofold. First, i beig sigal coditios, ay algorithm is sufficiet for sigal receptio; i fact, i the absece of RFI or multipath the predomiat reaso to employ a multi-elemet atea is to icrease gai toward the satellite to raise C/N 0 ad improve trackig accuracy. Secod, whe a referece sigal is available (as is the case for GPS) ad whe there may be degradatio i steerig-vector accuracy, the a LMS-based approach may have sigificat C/N 0 advatages. SPATIAL AND TEMPORAL D.O.F. I additio to the spatial degrees of freedom (d.o.f.) associated with icreasig the umber of ateas available for sigal receptio, there are temporal d.o.f. as well. By usig a tapped delay lie, the array sythesis may take advatage of sigal iformatio from more tha oe samplig epoch. Spatial d.o.f. correspod roughly with the geometry of the array gai patter ad the width (directivity/selectivity) of beams ad ulls. Temporal d.o.f. are related to the spectral respose of the array ad the ability to ull widebad RFI sigals. The total d.o.f. of the array (the product of the umber of ateas times the umber of time taps) determies the umber of beams, the depths of ulls, ad may allow ullig of widebad iterferece. There are sigificat SNR beefits from a multi-atea space-time adaptive processor (STAP) feedig ito the GPS code ad carrier trackig loops. I geeral, it is beeficial to use as may temporal d.o.f. as possible withi ±½-chip of the referece tap, ad withi the limit established by the auto-correlatio properties of the icomig sampled sigal. Spatial d.o.f. ca the be traded agaist temporal d.o.f. to reach a optimum give implemetatio costraits. To examie the ifluece of spatial ad temporal d.o.f. o GPS sigal trackig performace, data were collected ad the processed either from a sigle atea chael or for two ateas with sigals combied with a LMS adaptive algorithm. For this experimet, data were collected with two NordNav Techologies receiver froted uits operatig i clock sychroy, storig 4-bit data sampled at MHz with a MHz itermediate frequecy. The sigals were the acquired ad tracked with a software receiver implemeted i Matlab. There were 8 satellite sigals preset i the data records with C/N 0 values ragig from db-hz. Correlatio Power (ormalized) C/N0 = 48 db-hz 1 atea Correlatio Power (ormalized) C/N0 = 51 db-hz 2 ateas 1 time tap per chael LMS Algorithm Figure 5. Trackig of C/A-code PRN22. Correlatio Power (ormalized) C/N0 = 59 db-hz 2 ateas 5 time taps per chael LMS Algorithm For oe of the satellite sigals preset i the data record (C/A-code for PRN22), code ad carrier trackig usig a sigle atea chael yielded a C/N 0 estimate of 48 db- Hz (Figure 5). Whe trackig the combied total of both sigals, usig weights determied by a LMS adaptive algorithm, the resultig C/N 0 was approximately 51 db- Hz, a 3 db-hz icrease, as would be expected give the icrease i sigal power from two ateas. It should be emphasized that these results are achieved usig the desired sigal as the oly iput to the weight cotrol circuit the algorithm operated blid i that there was o survey of atea locatios, o calculatio of satellite ephemeris, o frot-ed or atea calibratio, ad o measuremet of atea cables or frot-ed delays. I cotrast to the sigle time tap example, with a 5 time tap STAP the effective C/N 0 icreased further to 59 db- Hz, with a correspodig reductio i carrier phase ad code phase errors (taps separated by 1/f s = μs). This ca be thought of as beig equivalet to addig equal amouts of balaced early/late multipath to the desired sigal available at the ceter, referece time tap. The reaso for the additioal icrease i C/N 0 is attributable to the fact that the filtered oise o the iput sigal (due to the receiver s aalog frot-ed) is o loger white. As ca be see from the auto-correlatio plot of a filtered WGN iput sequece, there is correlatio betwee samples; the arrower the filter passbad the loger the time-extet of correlatio. Sice a adaptive processor uses the sigal covariace matrix to drive the weight cotrol feedback, it meas that there is further oisereductio possible with the additio of temporal d.o.f. (give a sufficietly high samplig frequecy), sice the idepedece of the iput sequece has bee partially elimiated. I fact, for this choice of samplig frequecy there is i fact a icrease i C/N 0 for a sigle atea elemet with multiple time taps. This effect may be more clearly see by comparig the C/N 0 estimates for each satellite sigal preset i this data record (Figure 6). This plot shows estimated C/N 0 for each sigal as a thi lie, ad tracked either with a sigle atea, with both ateas ad a sigle time tap, or with icreasig umbers of time taps (3, 5, 7, 9, ad 11 tap algorithms); the esemble average is show as the thicker, red lie. These results emphasize the improvemets

7 Effective CN0 (db-hz) CN0 Improvemet from Base Case (db-hz) FRPA 1-tap 3-tap 5-tap 7-tap 9-tap 11-tap Processig Method: FRPA or LMS-Based Adaptive w/ Number of Time Taps icreasig the other d.o.f. Additioally, for GPS it may be fuctioally easier to add time taps tha ateas eve with oly 2 ateas, STAP will dramatically improve weak sigal trackig performace. So, midful of practical desig costraits, it is beeficial i terms of output C/N 0 to employ multiple time taps, recogizig the limits imposed by the GPS sigal structure, samplig frequecy, ad auto-correlatio properties of the sampled time series. Spatial d.o.f. ca also be exercised as required to meet RFI rejectio ad C/N0 trackig thresholds, ad give practical implemetatio costraits. The complemetary ature of spatial ad temporal d.o.f. is to be emphasized, ad for maximum array performace both aspects of a desig should be optimized. Effective CN0 (db-hz) Figure 6. Estimated C/N0 with LMS-STAP. possible by exploitig spatial ad temporal d.o.f., ad also show the dimiishig returs as the temporal extet approaches both the ±½-chip limit as well as the time extet of the sigal auto-correlatio fuctio. A additioal simulatio study further examies spatial ad temporal d.o.f. effects. For this simulatio, the fial test cofiguratio described earlier (6 swept-cw RFI sources, atea phase-ceter ad aalog frot-ed delay effects, ad errors i S/V ephemeris kowledge) was used, ad the umber of ateas varied from 1 to 4 to 7, while the umber of time taps wet from 1 to 3 to 5 (Figure 7). As ca be see here, icreasig either the umber of ateas or the umber of time taps yields icreasig C/N 0. Further, eve whe oe d.o.f. is costraied, there are still beefits to be realized by Timetap 3 Timetaps 5 Timetaps Effective CN0 vs. Spatial & Temporal D.O.F. STANFORD 4-CHANNEL HARDWARE TESTBED To verify that the methods that are developed ad tested i simulatio are applicable to real-world scearios, Staford Uiversity has pursued costructio of a datacollectio hardware testbed. This testbed is based o iexpesive, commercially available frot-ed compoets ad high-speed/high-resolutio aalog-todigital coverter cards. This testbed allows collectio of high-quality sigal-plus-jammig data records for subsequet playback through a variety of software tools, trackig loops, ad specialized software avigatio receivers [Figures 8 & 9; ad 17-19]. Sigals come i through 4 ateas placed o the roof of oe of the laboratory buildigs o the Staford Uiversity campus, ad are combied with a iterferece sigal, cetered at GPS L1, sythesized i the lab. These composite sigals are routed to the NovAtel SuperStar II GPS cards, where they eter the aalog frot-ed circuitry of the Zarlik GP2015 chip. The four SuperStar II GPS cards are clock sychroized by disablig their respective o-board oscillators ad usig istead a exteral 10 MHz OCXO as a timig iput. After sigal mixig, dowcoversio, ad filterig o the Zarlik frot-ed chip, a aalog IF sigal is tapped from pi #1 for routig to a buffer/isolatio circuit ad the o to the ICS-650 A/D samplig card. The itermediate frequecy from the Zarlik chip has bee desiged to be MHz, so samplig at 20 MHz is more tha adequate to capture the GPS C/A-coded sigal badwidth; alteratively, badpass samplig at MHz allows aliasig of the IF dow to MHz. Samplig is doe at 12-bit A/D resolutio to maximize dyamic rage; this A/D resolutio ca be artificially degraded for study durig post-processig Atea 4 Atea 7 Atea Number of Ateas i LMS Adaptive Array Figure 7. STAP & D.O.F. sesitivity study. Samples are stored i the o-board buffer of the A/D card with 500,000 samples per chael prior to buffer switchig (which results i a data-record gap), ad the writte to disk. Due to the extremely large throughput of

8 Iterferece Geerator GPS Display Host PC NovAtel SuperStar NovAtel SuperStar Serial cable Aalog lie Aalog lie PCI Bus ICS-650 master ICS bit A/D 5-65 MHz 2-chael NovAtel SuperStar Aalog lie ICS-650 slave Sychroizatio Cables 10 MHz Clock NovAtel SuperStar 4 Ateas 4 SuperStars 2 ICS-650 cards Commo Clock Aalog lie Samplig Clock SuperStar II Zarlik GP MHz IF IDE HD power/groud wires ad buffer circuits left off for clarity Figure 8. Staford Uiversity 4-chael test hardware schematic. data comig i, it has ot bee possible to stream the data to disk; rather, a millisecod sap-shot is take (depedig o samplig frequecy), stored o-card, ad the writte to disk. At this poit, the saved data record is available for post-processig. As a ote, the IF sigal from the GP2015 is still routed to the remaider of the SuperStar circuitry where it is processed by the o-board GPS receiver. A serial coector o the hardware chassis is utilized to coect ay of the four GPS receiver cards to the data acquisitio computer, allowig easy trouble-shootig i real-time with NovAtel s StarView iterface software. A 85 ms data record was collected usig the 4-chael hardware testbed samplig at MHz. Satellite sigals from each atea were idividually acquired ad tracked, ad also a composite sigal was formed by a LMS-based adaptive algorithm ad the tracked as well (usig acquisitio data from atea #1). The C/N 0 for each atea for five of the satellite sigals preset i the data record are show, as well as the C/N 0 for the combied sigal (Figure 10). Compared to the C/N 0 for atea #1, which was a high-quality piwheel atea ad had uiformly the highest C/N 0 of ay atea, the combied sigal achieved a average C/N 0 improvemet of 2.2 db-hz. CONCLUSIONS Figure 9. Staford Uiversity 4-chael test hardware. Multi-elemet ateas are a importat compoet of the JPALS architecture, ad adaptive algorithms will improve array performace i degraded coditios (iterferece, multipath, etc.). I this paper, it was show that whe a referece sigal is available (as is the case for GPS) ad whe there may be degradatio i steerigvector accuracy (due to hardware errors), the a LMSbased approach may have sigificat C/N 0 advatages. The LMS adaptive algorithm offers a efficiet hardware

9 Effective C/N0 (db-hz) C/N0 for FRPA vs. 4-elemet LMS Array Atea #1 Atea #2 Atea #3 Atea #4 4-elemet LMS PRN-23 PRN-13 PRN-20 PRN-27 PRN-16 Satellite Number Figure 10. Sigal trackig results. PRNs 23, 13, 20, 27, 16 i dataset. implemetatio, ad is robust to errors i the aalog electroics (Figure 11). Further, it was show that exercisig both spatial ad temporal degrees of freedom is desirable i maximizig C/N 0. This paper also icluded a discussio of Staford Uiversity s multi-chael hardware testbed ad a example of sigal trackig results. Based o iexpesive, commercially available frot-ed compoets ad highspeed/high-resolutio aalog-to-digital coverter cards, this testbed allows collectio of high-quality sigal-plusjammig data records for subsequet playback through a variety of software tools, trackig loops, ad avigatio receivers. The impacts to sigal trackig ad pseudorage ad carrier-phase estimatio are strogly depedet o the errors itroduced by atea phase-ceter ad group delay variatios with icomig sigal directio ad also by differetial propagatio through the aalog frot-ed chaels. The ext phase of research ivolves idetifyig compesatio methods for these various effects, ad determiig the total impact to avigatio accuracy ad itegrity. ACKNOWLEDGMENTS The authors gratefully ackowledge the ivaluable electroics assistace of Gil Gutterma ad Godwi Zhag i the costructio of the data collectio hardware described i this paper. The authors also gratefully ackowledge the support of the JPALS Program Office, ad the Naval Air Warfare Ceter Aircraft Divisio through cotract N C REFERENCES 1. Spilker, J.J. ad Natali, F.D., 1996, Iterferece Effects ad Mitigatio Techiques, i Global Positioig System: Theory ad Applicatios, Volume I, Parkiso, B.W. ad Spilker, J.J., eds., AIAA, Ic., pp Widrow, B. ad Stears, S.D., 1985, Adaptive Sigal Processig, Pearso Educatio, Ic. 3. Compto, R.T., 1988, Adaptive Ateas, Pretice- Hall, Ic. 4. Geca, A. ad Zoltowski, M., 1995, Power Miimizatio Techiques for GPS Null Steerig Atea, Proc. ION GPS 1995, pp Nicholso, B.W., Upto, D.M., Cotterill, S., Marchese, J., Upadhyay, T., ad Vader Velde, W.E., 1998, Computer Simulatios of Digital Beam Adaptive Weight Algorithm LMS-based Adaptive Blid Beamformig LMS referece Σ Trackig Loops code replica PVT LO carrier replica S/V-specific sigal processig Figure 11. Proposed 1 st -geeratio HW implemetatio, multi-elemet GPS ASIC.

10 Formig Adaptive Ateae for GPS Iterferece Mitigatio, Proc. ION NTM 1998, pp Hatke, G.F., 1998, Adaptive Array Processig for Widebad Nullig i GPS Systems, Proc. Thirty- Secod Asilomar Cof. o Sigals, Systems & Computers, vol. 2, pp Fate, R.L. ad Vaccaro, J.J., 2000, Widebad Cacellatio of Iterferece i a GPS Receive Array, IEEE Tras. o Aerospace ad Electroic Systems, vol. 36, o. 2, pp Gupta, I.J. ad Moore, T.D., 2001, Space-Frequecy Adaptive Processig (SFAP) for Iterferece Suppressio i GPS Receivers, Proc. ION NTM 2001, pp Moore, T.D., 2002, Aalytic Study of Space-Time ad Space-Frequecy Adaptive Processig for Radio Frequecy Iterferece Suppressio, PhD. Dissertatio, The Ohio State Uiversity. 10. Hatke, G.F. ad Phuog, T.T., 2004, Desig ad Test of a GPS Adaptive Atea Array Processor: The Multipath Adaptive Multibeam Array (MAMBA) Processor, Licol Laboratory Project Report GPS-16, Massachusetts Istitute of Techology. 11. Fate, R.L., Fitzgibbos, M.P., ad McDoald, K.F., 2004, Effect of Adaptive Array Processig o GPS Sigal Crosscorrelatio, Proc. ION GNSS 2004, pp McGraw, G.A., Rya Youg, S.Y., ad Reicheauer, K., 2004, Evaluatio of GPS Ati-Jam System Effects o Pseudorage ad Carrier Phase Measuremets for Precisio Approach ad Ladig, Proc. ION GNSS 2004, pp Frost, O.L., 1972, A Algorithm for Liearly Costraied Adaptive Array Processig, Proc. IEEE, vol. 60, o. 8, pp Applebaum, S.P., 1976, Adaptive Arrays, IEEE Tras. o Ateas ad Propagatio, vol. 24, o. 5, pp Widrow, B., Matey, P.E., Griffiths, L.J., ad Goode, B.B., 1967, Adaptive Atea Systems, Proc. IEEE, vol. 55, o. 12, pp Kim, U.S., 2005, Aalysis of Carrier Phase ad Group Delay Biases Itroduces by CRPA Hardware, Proc. ION GNSS 2005, i press. 17. NovAtel Ic., 2004, SuperStar II User Maual. 18. Zarlik Semicoductor Ic., 2002, Datasheet for GP2015 GPS Receiver RF Frot Ed. 19. Iteractive Circuits ad Systems Ltd., 2001, ICS- 650 Operatig Maual.