The 010 Mltary Communcatons Conference - Unclassfed Program - Waveforms and Sgnal Processng Track A RF Source Localzaton and Trackng System Wll Tdd, Raymond J. Weber, Ykun Huang Department of Electrcal and Computer Engneerng Montana State Unversty Bozeman, MT 59717 Yufe Zhao Department of Electrcal and Computer Engneerng Unversty of Calforna Davs, CA 95616 Abstract Ths paper descrbes an RF source localzaton and trackng system usng an 8-element unform crcular array. The system s capable of precse RF source localzaton, fast target recognton and dscrmnaton, fast spectral sensng, real-tme processng for locaton trackng, and the suppresson of jammng and mult-path sgnals. Exstng systems and technology for precson locaton and trackng of non-cooperatve RF emtters are costly and tme consumng to develop and deploy, and some rely on satellte communcaton whch s not always avalable and s susceptble to jammng. Ths novel RF source localzaton and trackng system offers strategc, fast, and low-cost advantages n the next generaton of RF mappng, sensor networks, and moble wreless systems snce t s capable of automatc target localzaton and trackng, dynamc spectrum access, and drectonal power recepton/radaton wthout relyng on satellte communcaton. Keywords-RF mappng; DOA estmaton; frequency sensng; trackng; Kalman Flter. I. INTRODUCTION The Navy seeks novel solutons for detectng, locatng, and trackng moble rado frequency emtters and for long dstance relable communcaton. We have developed a novel and nnovatve RF source precson localzaton and trackng system that can localze and track moble rado frequency (RF emtters. Exstng systems and technology for precson locaton and trackng of non-cooperatve RF emtters are costly and tme consumng to develop and deploy, and some rely on satellte communcaton whch s not always avalable and s also susceptble to jammng. We have explored the feasblty of mplementng hgh resoluton smart antennas n conjuncton wth frequency swtchng and selecton technologes to create antenna array systems capable of reconfgurng ther spatal and spectral recevng and transmttng characterstcs n response to dynamc envronmental demands. Ths novel RF source localzaton and trackng system offers strategc advantages n the next generaton of RF mappng, sensor networks and moble wreless systems snce t s capable of automatc target localzaton and trackng, dynamc spectrum access, and drectonal power radaton/recepton. The frequency sensng, Drecton of Arrval (DOA estmaton and beamformng ablty help to reject frequency hoppng and nterference whle mantanng good communcaton lnks. suppressng jammng. The system s compact, lghtweght, and energy effcent. The hardware and software of our system s desgned to ncorporate relay hop and mesh capablty nto the Navy s moble communcatons technologes. Passve emtter locaton s an mportant component of the Sgnal Intellgence (SIGINT system. The characterstcs of an emtter sgnal, whch n our case s DOAestmaton, can be determned by multple antenna arrays. By usng the DOA estmates and known locatons of these arrays estmates from multple locatons can then be transmtted to an RF locator whch s responsble for emtter locaton and trackng. Specfcally, the data from multple arrays s collated at a sngle locaton and analyzed usng well-known technques such as multlateraton employng the use of TDOA (Tme Dfference of Arrval to arrve at a sngle DOA estmate. The results of our studes on TOA and TDOA localzaton wth ths array system have been publshed [1, ]. Theoretcal study, algorthm smulaton, and hardware desgn and mplementaton have all been carred out. A prototype system has been bult for lab tests. The system ncludes a unform crcular array that enables 360 degree azmuthal coverage, an RF recever board, a hgh speed A/D board, an FPGA board for Angle of Arrval (AOA estmaton and a PC to provde a user frendly Graphcal User Interface (GUI. The system operates at the 5.8 GHz band wth a bandwdth up to 1.5 MHz. The array gan s about 14dB. The Mean Square Error (MSE of DOA estmaton ranges from 0.01 to 1, dependng on the Sgnal to Nose Rato (SNR, sgnal frequency, samplng number and the algorthm. The current beamformng array and DOA estmaton system use a combnaton of analog and dgtal sgnal processng technques. Ths technology s used as the test-bed for mult-purpose tests and wll be extended nto the dgtal doman to add functonalty, to mprove performance, and to lower cost. The system can be scaled to operate at other frequences. Ths paper wll focus on our smulaton and lab test results on DOA estmaton usng the MUSIC algorthm and DOA trackng usng the Kalman state estmator. Ths paper s organzed as follows. In Secton, an overvew of our RF emtter localzaton and trackng technque s presented. Next, n Secton 3 a smulaton study on frequency search, spectral hole-fndng, DOA estmaton, and beamformng procedures s explaned. Secton 4 provdes an overvew of the hardware desgn. Then, n Secton 5 the lab test results are dscussed. Fnally, the conclusons are presented n the last Secton, 6. Ths work s supported by the Montana Board of Research and Commercalzaton Technology program under grant #09-09. 978-1-444-8179-8/10/$6.00 010 IEEE 67
II. A 3D RF EMITTER LOCATION AND TRACKING TECHNIQUE We consdered an m - element adaptve array wth arbtrary shape. The array receves sgnals from M spatally separated users. The proposed 3D RF emtter localzaton estmaton and trackng technque s summarzed as follows: A. Frequency Sensng There are numerous ways to sense the RF envronments frequency spectrum. A standard way s to compute an FFT of a block of data and then locate the peaks. Another approach for accurate frequency estmaton s a temporal doman ESPRIT frequency estmaton algorthm [1, ]. Ths uses two sgnal sets, wth one orgnal data set and one delayed data set where the tme between the two data sets s less than or equal to the Nyqust samplng nterval. The sgnal frequences of multple sgnals wthn the baseband are determned from the generalzed egenvalues of a matrx composed of crosscorrelatons between the two sgnal paths. The appled metrc s a matrx constructed from the auto-correlaton sequence for ether path. B. DOA Estmaton There are many DOA estmaton algorthms that can be used for ths purpose. For arbtrarly shaped arrays we chose the hgh resoluton Multple Sgnal Classfcaton (MUSIC DOA estmaton algorthm [3]. MUSIC performance approaches the Cramér Rao Bound (CRB, the theoretcal lmt to the performance achevable by an unbased estmaton algorthm. Ths s acheved through the use of matrx subspace technques whch force the nose and sgnals nto orthogonal subspaces from whch the orthogonal matrx can be computed. The DOA estmaton results from multple arrays wll be used for locaton estmaton. C. Jont Frequency and DOA Estmaton We have developed a novel method for jont frequency and D angle estmaton of ncomng sgnals mpngng upon a crcular array over a very wde frequency band (-18 GHz.[4] A temporal doman ESPRIT frequency estmaton algorthm was extended for azmuth and elevaton angle estmaton. These drecton estmaton results are automatcally pared wth frequency. The algorthm provdes non-ambguty, hgh resoluton, and accuracy n both frequency and drecton of arrval estmaton over the frequency range. The DOA estmaton accuracy and resoluton were compared wth the orgnal MUSIC DOA estmaton results. D. Locaton Estmaton The Hough Transform was used for locaton estmaton [5]. The dstance from a reference ste at ( x, z to the locaton (x, y, z s found to be, R ( xr x ( y y ( z z r r The range dfference between a locaton on the grd, array, and the reference ste 1 s then: r y r r c,1 R R ( x ( x r x 1 ( y ( r y r x y y z z 1 r1 r1 ( z ( r z Assume that the measurement errors n DOA estmaton are Gaussan. The votng functon s defned n terms of the condtonal probablty densty functon (pdf. The DOA estmaton pdf s: exp p( x, y Where s the measured angle at array, s the calculated angle to pont xy (z on the grd and s the standard devaton n the angle measurement. The resultng votng equaton for DOA only estmaton s: A M 1 ( x, y p( x, y DOA m M m1 E. Trackng The mplemented Kalman state estmator was heavly based upon an artcle by Dan Smon; mplementaton and code documented n [6]. The code focused on vehcle navgaton applcatons and was modfed prmarly n the nputs such that t tracked angular poston nstead of Cartesan coordnate postons. The Kalman flter s a smple and well known algorthm for trackng. Ths algorthm s a specal case of a recursve least squares flter that predcts the velocty and acceleraton of a source n the presence of nose. Whle many algorthms have been developed for use n systems such as ths, the Kalman flter s used n the followng smulaton study n ths paper. III. SIMULATION STUDY All the smulatons were performed for an 8-element crcular array. The array dameter s.0. The operaton frequency s 5.8GHz. A. DOA locaton estmaton As an example, consder three statonary arrays located at (0km, 0km, (0km, 0.5km and (0.75km, 0.01km respectvely and a sngle source located at (.5km, 1km. The SNR s 0 db and the mesh spacng s 10 m. At each array, DOA estmaton was made and the standard devaton of each estmaton was assumed as 0.01 durng our Monte Carlo smulaton. The correspondng Hough Transform space for these DOA measurements s shown n fgure 1. The red dots represent the arrays and the peak n the HT space (represented as a black dot s the RF source. In ths smulaton we used three arrays for comparson purposes. 68
on each prevous sgnal frequency and angle. Ths approach s vald assumng that a target wll not move more than a predefned range n one update perod. The thrd approach lmted our trackng abltes so smulaton studes were only consdered and carred out usng the frst approach. Ths Kalman flter technque was appled to the results of the jont frequency and DOA estmaton algorthm. The smulaton consdered a sgnal movng at a constant angular velocty from 10 to 60 and then mmedately ceasng movement. The results of ths smulaton are shown n Fgures 3 and 4 where True s the Kalman estmate, Measured s the DOA estmates, and Estmated s the current Kalman state estmate. These fgures show that the trackng works as expected and quckly converges upon the necessary velocty's to track the target. Fgure 1. HT space for D DOA measurement for one sgnal locates at (km,1km. Next, we ran a 3D locaton estmaton wth two arrays. Two statonary arrays located at (km, 1km, 1km and (km, 3km,.5km respectvely and a sngle source located at (9km, 4km, 1km. The SNR s 0 db and the mesh spacng s 5 m. The correspondng Hough Transform space for these DOA measurements s shown n fgure. The green dots represent the arrays and the peak n the HT space (represented as a black dot s the RF source. (a Kalman Trackng Poston Fgure. HT space for 3D DOA measurement for one sgnal locates at (9km, 1km,150m. F. Trackng Three approaches to trackng were consdered. One avenue was to use a closed loop adaptve algorthm, such as the Kalman flter, to track angular speed and locaton of the detected sgnals. Smulaton studes were carred out for trackng a moble target. It starts at 10, movng 5 per run. The sgnal s a 5.8 GHz contnuous wave (CW. The samplng rate s GHz. The sgnal to nose rato s 0 db. The samplng tme s 1000 per run. The second approach conssted of runnng the ESPRIT based algorthm perodcally to make sure that the sgnal frequency and angle estmatons were current. Ths approach was to computatonally ntensve and was not mplemented. The thrd approach was to use the prevous sgnal frequency and angle as a startng pont and only run frequency and angle estmaton n narrowed wndows centered (b Kalman Resultng Poston Error Fgure 3. DOA estmaton Kalman trackng smulaton results. The Kalman flter was also smulated for trackng two separate targets by keepng separate tracks of each target and assumng that the pont closest to the estmated next track of a target was a contnuaton of the same target. These tracks then kept ther data separate whle runnng the Kalman estmaton algorthm ndvdually on each track. For the smulaton n Fgure 5, two sgnals were tracked wth the MUSIC DOA estmates beng fed nto the Kalman estmaton algorthm as they move lnearly around the array and then stopped. 69
(a Kalman Trackng Velocty Fgure 6. An 8-monopole crcular array wth a ground skrt. The system we mplemented n the lab, as shown n fgure 7, determnes a target's DOA and passes the relevant data nto a Kalman trackng algorthm. Fgure 7. Array system dagram. (b Kalman Trackng of Velocty Error Fgure 4. Angular velocty trackng smulaton results. The system uses a recever board (RX to amplfy and down-convert the rado frequences to baseband. Then a Data AcQuston (DAQ card creates a dgtal representaton of the tme sgnal and sends the data to MATLAB for FFT processng. The recever board prototype that has been bult s depcted n fgure 8. The RF sgnal from the antenna array s fed nto the top 8 SMA nputs shown n the photo below. The lower 8 SMA connectors are the down-converted baseband output, and the local oscllator (LO feed s the center SMA. The automatc gan control settngs are at baseband frequences and are located on the back of the board. Fgure 5. Two targets Kalman DOA estmates tracks. IV. HARDWARE OVERVIEW We desgned an 8-element unform crcular array (UCA as shown n Fgure 6. The array electrcal sze s 3.0. The nterelement spacng s 0.375. The array gan s 14 db. The desgned operatng frequency s 5.8GHz. Because the UCA offers 360 beam steerng wthout sgnfcant effect on the beam shape and no nherent angular estmaton ambguty, a crcular geometry was chosen nstead of a lnear one. Fgure 8. RF recever board snapshot. A LabVIEW PC nterface has been developed to control all the nstrumentaton and hardware. It s used to control the processng of the ncomng sgnals after acquston and dgtalto-analog converson. Frst, t s used to determne the frequences of nterest, DOA estmates, and avalable spectral holes. Second, t s used to beamform toward the users, null the nterferers, and track the moble users. 70
V. LAB TEST RESULTS AND DISCUSSIONS The experments were conducted n an anechoc chamber at the Electrcal and Computer Engneerng (ECE Department of Montana State Unversty (MSU. These experments were desgned to test the accuracy and robustness of the MUSIC DOA and Kalman trackng system n a test-bed envronment, and to provde emprcal proof for the prevously conducted smulatons. We frst verfed the functonalty of the MUSIC DOA algorthm n our system wth an SNR of 15 db. In fgure 9, we show the MUSIC algorthm returns a sngle peak at the estmated locaton of the source wth respect to the array, and the peak s sharp as a result of a calbraton technque compensatng for the mutual couplng of the array head. Wthout movng the turntable and takng repeated tests, the standard devaton of the angle estmates vared between 0.03 for 100 runs at approxmately 35 db SNR and ncreasng to about 1 wth lower SNRs (5 10 db whch closely resembled theoretcal predctons. All experments and smulatons were conducted wth a sample rate of.5 MSPS and 1000 samples. SNR of 15 db was placed n the chamber whle the 8-element array was located opposte the source on a turntable. The turntable was steered on a random path ncludng changes n drecton and speed to smulate the target movng at dfferent veloctes around the array. Fgure 11 shows the true angle measured wth the MUSIC DOA algorthm versus the Kalman estmated DOA as well as the error between the two. The teratons are unt-less values of tme on the order of seconds. Fgure 11. Kalman trackng poston and error for 1 target. Fgure 1, plots the true velocty of the source, found by takng the dervatve of the true angle above wth respect to tme, versus the estmated Kalman velocty. The error between the two s plotted as the velocty estmated error on the rght sde axs of the plot. Fgure 1 s assocated wth the same source path as fgure 11. Fgure 9. Sngle source DOA pseudo power-spectra. In fgure 10, we repeated the experment wth two sources at slghtly dfferent frequences wth a combned SNR of approxmately 0 db to verfy that we can detect the locaton of targets wth approxmately 90 of separaton. Fgure 1. Kalman trackng velocty and error for 1 target.. Second, two CW sgnal sources/targets wth a combned SNR of 0 db were placed at a dstance satsfyng the far feld crteron from the 8-element array. Intally both were movng around the array at a constant angular velocty. The second target was then moved at a dfferent velocty n the path of the frst target wth the ntenton of testng the robustness of the trackng algorthm and then moved back to ts statonary poston wth respect to the frst source. Fgure 13, shows the true DOA and estmated Kalman DOA for both sources. Fgure 10. Two source DOA pseudo power-spectra. Next, we tested and verfed the functonalty of the Kalman flter algorthm. Frst, one CW sgnal source wth an 71
Fgure 13. Kalman trackng poston for targets. Fgure 14, shows the Kalman trackng error as the dfference between the true MUSIC DOA and the estmated Kalman DOA for both sources, respectvely. Fgure 14 s assocated wth the same target paths as fgure 13. poston based on the prevous velocty and poston, slght overshoot can be notced n the Kalman estmaton DOA when the source suddenly stops or changes drecton. The overshoot s exacerbated when the source s movng at faster veloctes. Some small DOA error s nduced when usng the Kalman trackng system, but the reducton n computatonal complexty far outweghs the nduced angle error exemplfyng the feasblty of the Kalman state estmator. The expermental results renforced the smulaton results and proved the abltes of the current hardware platform to quckly ascertan hghly accurate DOAs and track multple sources even when ther paths nterfere. The system's DOA and trackng estmates do experence degradaton wth lower SNR values, ncreased target veloctes, or unorthodox target trajectory paths. The DOA estmaton accuraces are drectly proportonal to SNR levels translatng nto ncreased DOA and trackng error wth lower SNR. Ths relatonshp has yet to be quantfed, but t can be assumed for low SNR (< 0 db or very fast movng objects (> 90 /sec the DOA and trackng estmaton methods become hghly senstve and may breakdown. These algorthms are desgned to work n an envronment that does not exceed these lmts. The overall system desgn s a small, lghtweght, energy effcent devce, whch can acheve hghly accurate DOA estmatons and concurrently track multple sources usng a Kalman based trackng algorthm. Fgure 14. Kalman trackng error for targets. VI. CONCLUSIONS The experments tested and verfed the MUSIC DOA and Kalman trackng algorthms wth one and two targets at varyng speeds and drectons. The results of the DOA estmaton experments show the system's ablty to perform DOA estmaton wth the MUSIC algorthm for sngle or multple targets. The calbraton mplemented allows for these estmates to approach the theoretcal predctons of accuracy as measured through the standard devaton of multple runs. The experments tested the DOA and Kalman trackng algorthms wth one and two targets at varyng speeds and drectons. Because the Kalman estmator predcts current ACKNOWLEDGMENTS The authors thank Andy Olson, Ahmed Khallaayoun, and Aaron Traxnger for ther help n hardware desgn, calbraton and testng. REFERENCES [1] Raymond J. Weber, Ykun Huang. A local postonng system for wreless networks. Natonal Rado Scence Meetng, Unversty of Colorado at Boulder, January 010. [] Raymond J. Weber, Ykun Huang. 3d RF emtter locaton estmaton. 010 IEEE Aerospace Conference, Bg Sky, MT, March 010. [3] Ralph O. Schmdt. Multple emtter locaton and sgnal parameter estmaton. IEEE Trans. on Antennas and Propaga., AP-34(3:76-80, March, 1986. [4] Raymond J. Weber and Ykun Huang. A wdeband crcular array for frequency and D drecton estmaton and trackng, IEEE Aerospace Conference, March, 010. [5] P. V. C. Hough. Method and means for recognzng complex patterns. US Patent 3069654, 196. [6] Dan Smon. Kalman Flterng. Embedded Systems Programmng, pp. 7-79, June 001. 7