Analysis and Practical Comparison of Wireless LAN and Ultra-Wideband Technologies for Advanced Localization

Transcription

1 Analysis and ractical Comparison of Wirlss LAN and Ultra-Widband Tchnologis for Advancd Localization Stfan Gallr, Jns Schrodr, Golalh Rahmatollahi, Kyandoghr Kyamakya, and Klaus Jobmann Abstract In this papr w compar th capabilitis of Wirlss LAN and Ultra-Widband to nabl advancd localization systms. Th two tchnologis ar compard rgarding thir suitability towards tim-dlay-stimation by thortical bounds as wll as by mans of practical tim-of-flight rang stimation in LOS and NLOS conditions in an industrial nvironmnt. Indx Trms Ultra-Widband, UWB, Wirlss LAN, WLAN, localization, positioning, TDOA I. INTRODUCTION Indoor localization has bcom an important fild of rsarch with its potntial for applications in privat and industrial nvironmnts. Espcially for production, logistics, factory and building automation or scurity and safty applications, combind indoor communication and localization systms with sub-mtr accuracy ar rgardd as a ky nabling tchnology to a whol varity of nw srvics. Two tchnologis sm to b appropriat candidats to fulfill th dsird functionality: Wirlss LAN (IEEE.) bcaus of its high markt pntration for communications and also incrasingly for localization basd on rcivd signal strngth mthods, offring accuracis of som mtrs, and Ultra Widband (UWB), which is just starting for high-rat, short-rang multimdia transfr, but rsarch and standardization activitis targt lowr-rat and longr-rang concpts, bcaus of its promising prcis localization. To achiv robust and rliabl sub-mtr accuracy localization in challnging aras as industrial nvironmnts, advancd localization tchniqus hav to b applid. Tim-offlight masurmnts instad of rcivd signal strngth mthods, in combination with advancd filtring and tracking, Manuscript rcivd May,. This work was supportd in part by th DFG (Grman Rsarch Foundation) undr Grant KIO, and th Minna- Jams-Hinmann Foundation. S. Gallr, J. Schrodr, G. Rahmatollahi and K. Jobmann ar with th Institut of Communications Enginring (IKT, Univrsity of Hannovr, Hannovr, Grmany (phon: +9 7-; fax: ; - mail: (gallr, jns.schrodr, gola, jo)@ant.uni-hannovr.d). K. Kyamakya is with th Institut of Informatics Systms, Univrsity of Klagnfurt, Klagnfurt, Austria (phon: , fax: , -mail: kyamakya@isys.uni-klu.ac.at). rlying on prcis tim dlay stimats as wll as th capability to rsolv multipath componnts and th ability to distinguish btwn lin-of-sight (LOS) and non-lin-of-sight (NLOS) conditions ar rgardd as ssntial for such systms. Th goal of this papr is hnc to practically compar th capabilitis of th two tchnologis by mans of tim-of-flight rang stimation in LOS and NLOS conditions in an industrial nvironmnt. Th papr is organizd as follows. Sction II compars rang stimation minimum variancs capabilitis using th Cramr-Rao Lowr Bound. In Sction III, th masurmnt systm stup and th signal procssing applid ar dscribd. Sction IV prsnts th masurmnt campaign and rspctiv rsults. Th papr finalizs with a conclusion in Sction V. II. RANGE ESTIMATION BOUNDARIES Th accuracy of tim-of-flight basd localization dpnds on th gomtrical constllation of sourcs and rcivrs and on th accuracy of th rang stimats. In this papr w concntrat on th lattr. Th Cramr-Rao lowr bound is widly usd as a lowr bound for an unbiasd stimator. In th cas of an all-whit-gaussian-nois channl, th varianc σ ˆτ of th tim dlay stimat is boundd by [], () πβfsnr σ τ ˆ whr SNR rprsnts th signal to nois ratio availabl at th rcivr and β f is th ffctiv bandwidth of th rcivd signal dfind by, () β f f Sf () df Sf () df and S( f ) is th Fourir transform of th signal. As it can b obsrvd, th impact of th ffctiv bandwidth β f is quadratic, compard to th availabl SNR, which is linar. Th ffct on tim dlay stimation using th corrlation mthod can b sn in Fig.. Th normalizd autocorrlation pak of an Ultra-Widband signal having.7 GHz of bandwidth is dpictd in th top figur, th normalizd autocorrlation pak

2 of an IEEE.b Wirlss LAN having MHz in th bottom tim in ns tim in ns Fig.. Normalizd autocorrlation pak of Ultra-Widband (.7 GHz) and Wirlss LAN ( MHz) signals. Th much narrowr pak of th Ultra-Widband autocorrlation as a rsult of th largr bandwidth, allows mor prcis tim dlay stimats, spcially in th cas of multipath conditions, wr multipl copis of th autocorrlation function form an additiv ovrlay. III. MEASUREMENT SYSTEM SETU This sctions dscribs th stup of th masurmnt systm, th applid signal procssing, th mtrics usd to compar th masurmnts and th tst of systm functionality. A. Gnral Systm Stup Th gnral stup of th masurmnt systm is th sam for both tchnologis, Ultra-Widband and Wirlss LAN. Cntral componnt of th masurmnt systm is a high-spd digital sampling oscilloscop, which in conjunction with a C running Matlab oprats as a fully digital rcivr at Nyquist rat (s Fig. ). triggrd, th oscilloscop simultanously capturs signals rcivd by th four antnnas connctd to its inputs. For signal procssing, all rcordd data is subsquntly transfrrd to a standard C running Matlab. Commrcially availabl Arc Frdom antnnas ar usd, which fatur accptabl puls as wll as continuous wav radiation and rcption charactristics in th utilizd frquncy rang []. Th Ultra-Widband tst dvic is basd on a simpl monophas pulsr circuit dscribd in [3]. Th dsign has bn xtndd to b capabl of using th avalanch ffct of two transistors, rsulting in a bi-phas puls gnrator. ulsdurations ar in th ordr of sub-nanosconds, corrsponding to a masurd bandwidth of approximatly.7 GHz. Assuming a fully balancd systm, puls pak powr according to puls pak amplitud of 7.3 V can b stimatd to b 3 dbm into a Ω load. Avrag puls powr at puls rptition frquncy 3. MHz can b stimatd to b dbm into a Ω load. As masurmnt signal, w us a 3 chip biphas modulatd RN puls-train. Bcaus of th bandpass charactristics of th antnnas, only th frquncy rang from.7 to.7 GHz was usd, rsulting in an ffctiv bandwidth of GHz. A mor dtaild dscription of th Ultra-Widband tst dvic can b found in []. Th Wirlss LAN tst dvic is a standard off-th-shlv IEEE.b D-Link DI- accss point routr, configurd to oprat on channl (. MHz). Bcaus of its good auto-corrlation proprtis, w us th IEEE.b physical layr convrgnc protocol (LC) prambl as masurmnt signal []. B. Signal rocssing To calculat th tim dlay stimats, all masurmnt data ar transfrrd from th oscilloscop to th C. Fig. 3 shows th dirct cross-corrlation rcivr structur implmntd in Matlab, considring as xampl channl and channl : s ( n) H (ω) s ( n) i i N k= s ( k) s ( n+ k) intrpolation tim dlay stimation Δt s ( n ) H (ω) Fig. 3. Dirct cross-corrlation rcivr implmntation in Matlab. Fig.. Gnral Masurmnt Systm Stup. As sampling oscilloscop, w us an Agilnt infinium A, running at a sampl rat of F S = GS/s with a mmory dpth of. sampls pr channl. Whn Bfor cross-corrlation, ach channl si( n), i is pr-filtrd with H( ω ) to supprss out of band intrfrnc. In th cas of Ultra-Widband, H( ω ) is implmntd as a bandpass filtr with passband frquncis from.7 to.7 GHz. In th cas of Wirlss LAN, H( ω ) is implmntd as a bandpass filtr with MHz bandwidth at cntr frquncy. MHz.

3 Th tim dlay diffrncs Δt, Δt3, Δt, Δt3, Δ t and Δ t 3 ar stimatd by a maximum sarch of th crosscorrlation function T / Δ tij = ti tj = arg max sj( t) si( t + τ) dt, (3) τ T / whr t i and t j dnot th absolut tims of arrival at antnna i and j, rspctivly, and i < j N. () N rprsnts th total numbr of antnnas. Th TDOA stimats Δ tij ar convrtd to rang diffrncs Δ dij through multiplication by th spd of light c : Δ d = cδ t = c( t t ) = d d. () ij ij i j i j As in our cas cross-corrlation is prformd in th digital discrt-tim domain, th tim dlay stimation rsolution is limitd to th duration of th sampl priod. In ordr to furthr incras th rsolution, intrpolation algorithms can b applid, as th xact tim dlay stimat is dtrmind by th maximum of th corrsponding continuous-tim corrlation function, instad of th discrt-tim squnc. Espcially for quasi sinusoidal continuous wav signals, as th Wirlss LAN signal, intrpolation prior to maximum dtction is ssntial, as bcaus of sampling, th discrttim maximum dos not ncssarily corrspond to th quivalnt continuous-tim maximum [].. C. Error Mtrics Th rror mtrics usd throughout this papr to compar masurmnt rsults ar: th simpl rror <> = xˆ x, () l l l dfind as th Euclidan distanc btwn th ral position x l and th stimatd position x ˆl, th arithmtic man μ <> l l = =, (7) of masurmnts, th root man squar (RMS) rror <> l < > l= RMS = th standard dviation σ = <> ( l μ) l=, (), (9) and an stimat of th corrsponding cumulativ distribution function cdf ( <> ). D. Tst of Systm Functionality Systm functionality has bn vrifid in two stps. First, th systm was tstd fully wird with qual lngth cabls, in ordr to avoid all antnna and radio channl propagation ffcts. Fig. xmplarily shows th propagation dlay diffrncs stimatd for channl and channl using th Wirlss LAN tst dvic. Normalizd corrlation Δd [Sampl] Tim Fig.. Corrlation pak of a quasi sinusoidal signal sampld slightly abov Nyquist rat (Th continuous lin is showing th continuous-tim signal and th crosss ar showing th discrt-tim sampld signal). As it can b sn in Fig., th maximum sarch in th discrt-tim domain can rsult in tim dlay stimation rrors of multipl priods of th quasi-sinusoidal signal, vn whn sampld abov Nyquist rat. In ordr to avoid such rrors in th masurmnt systm, th corrsponding continuous-tim corrlation function is furthr approximatd in th digital domain by intrpolation through zro padding and subsqunt lowpass filtring [7] Indx of masurmnt Fig.. Wird propagation dlay diffrnc stimats using Wirlss LAN Th arithmtic man μ of approximatly.7 Sampls (. cm) from xpctd valu zro rsults from physically not xactly qually long cabls. Th standard dviation σ of. Sampls (. mm) originats from nois in th tst signal, timing jittr and analog-to-digital convrsion nois of th oscilloscop and can b rgardd as th ovrall masurmnt systm s accuracy.

4 In a scond stp, th systm was tstd in an anchoic chambr, including antnna ffcts but minimizing radio channl ffcts. Fig. xmplarily shows th propagation dlay diffrncs stimatd for channl and channl using th Wirlss LAN tst dvic.. ral stimat Δd [m] Indx of masurmnt Fig.. Anchoic chambr propagation dlay diffrnc stimats using Wirlss LAN. Th arithmtic man μ of ths masurmnts is 3. cm, with a standard dviation σ of 3. cm. This accuracy is still accptabl, considring antnna ffcts as wll as rrors of ral position dtrmination, and th systm is hnc accptd as functional. IV. MEASUREMENT RESULTS Ral world masurmnts wr conductd in an industrial storag nvironmnt, quippd with concrt cilings, floor, and columns as wll as multipl mtallic objcts (s Fig. 7). Fig.. Masurmnt stup showing th straight lin masurd along and th positions of th four rciving antnnas. Th four rciving antnnas wr positiond at locations x y z x y z x3 y3 z3 = x y z.99.. m. () All ral positions of th mobil dvic and th infrastructur wr dtrmind using a Ziss Elta C3 total station. Two masurmnt sris wr conductd. Th first sris consists of tim-diffrnc-of-arrival masurmnts with an always xisting LOS componnt. Th scond sris was takn with th LOS path componnt xplicitly blockd, rprsnting NLOS conditions. A. Lin-of-Sight Masurmnts Bcaus of th indoor nvironmnt dscribd abov, th channls wr subjct to svr multipath. Howvr, in this masurmnt sris, LOS paths wr xistnt at all positions of th mobil dvic. Fig. 9 shows th stimatd rang diffrncs using th Ultra-Widband puls gnrator as mobil dvic. Δd Δd 3 Δd Δd 3 Δd Δd 3 Fig. 7. hotograph of th masurmnt nvironmnt. Each masurmnt sris was takn moving th transmitting antnna in cm stps along a 9.9 m long straight lin, rsulting in N = masurmnts (s Fig. ). Entfrnungsdiffrnzn Rang diffrncs Δd Δdij ij [m] Schätzung Echt Indx of masurmnt Fig. 9. UWB rang diffrncs among all antnnas with th straight lin showing th ral and th crosss showing th stimatd rang diffrncs.

5 As it can b obsrvd, stimatd valus and ral valus form a good match. Th fw outlirs, which can b found for xampl in Δ d, could b asily compnsatd in subsqunt localization algorithms. Th root man squar rror < > RMS of thss masurmnts is.9 m with a standard dviation σ of. m. Th corrsponding stimatd cumulativ rror distribution function cdf ( <> ) is dpictd in Fig.. Th rlativly fw outlirs can b rcognizd hr in th flat part of cdf ( <> ) for valus abov 97%..9 Cumulativ distribution function cdf( <> ) σ σ 3σ Cumulativ distribution function cdf( <> ) σ σ 3σ 3 3 Entfrnungsfhlr <> [m] Fig.. Wirlss LAN cumulativ distribution function. Tabl I summarizs th prformanc mtrics of th two masurmnt sts: TABLE I ACCURACY ERFORMANCE Mtric Ultra Widband Wirlss Lan Entfrnungsfhlr <> [m] Fig.. UWB cumulativ distribution function. Fig. shows th stimatd rang diffrncs using th sam masurmnt stup as bfor, but with th Wirlss LAN transmittr as mobil dvic. As it can b obsrvd, th rsults ar much wors. Th root mans squar rror < > RMS of thss masurmnts is.9 m with a standard dviation σ of 7. m. Th conglomration of stimats at th top and at th bottom is du to th limitation of stimats to thir gomtrically maximal possibl ral valus. Entfrnungsdiffrnzn Rang diffrncs Δdij [m] 3 - Δd Δd Δd 3 - Δd Schätzung - - Echt Indx of masurmnt - Δd - - Δd 3 3 Fig.. Wirlss LAN rang diffrncs among all antnnas with th straight lin showing th ral and th crosss showing th stimatd rang diffrncs. Th corrsponding stimatd cumulativ rror distribution function cdf ( <> ) dpictd in Fig. also shows th svrly lowr accuracy compard to Ultra-Widband. - < > RMS.9 m.9 m μ. m 7.7 m σ. m 7. m It can b concludd for lin-of-sight-conditions in multipath nvironmnts, that Ultra-Widband offrs a significantly highr ranging accuracy than Wirlss LAN. B. Non Lin of Sight Masurmnts For th cas of NLOS masurmnts, th masurmnt systm has bn changd. An absorbing wall in front of antnna was usd to block th dirct path to this antnna at all positions of th mobil dvic, rsulting in NLOS conditions (s Fig. 3). Fig. 3. Masurmnt stup with absorbing wall in plac, for non-lin-ofsight masurmnts. In ordr to avoid cross channl ffcts, rsulting from th cross-corrlation of two rspctiv wirlss channls, a wird tmplat of th transmittd signal was usd instad. Th

6 tmplat was rcordd on channl using a powr splittr (s. Fig. ). s ( n) i i But whil th rror histogram in th cas of Wirlss LAN simply sms to widn (s Fig. ), Ultra Widband allows th dtction of dominating paths, rsulting in rror clustrs, which ar quasi paralll to th ral mobil dvic distanc (s lowr lft of Fig. ). Th tim dlay information containd in ths quasi paralll rror clustrs, which ar strongly spatially corrlatd to th dirct LOS componnt, is still valuabl for advancd localization and tracking algorithms to gain positioning information. Fig.. Systm stup for Non Lin Of Sight masurmnts. Two masurmnt sts wr conductd for this scnario: Th first st without any blocking objct, th scond st with th absorbr wall in front of antnna. Th rsults ar dpictd in Fig.. Entfrnung d [m] UWB LOS d Schätzung stimatd Echt ral UWB NLOS d WLAN LOS d WLAN NLOS d V. CONCLUSION Ultra-Widband signals and IEEE.b Wirlss LAN signals hav bn compard rgarding thir capabilitis for advancd indoor localization. Th two tchnologis wr analyzd rgarding thir suitability towards tim-dlaystimation by thortical bounds as wll as by mans of practical tim-of-flight rang stimation in LOS and NLOS conditions in an industrial nvironmnt. As prdictd by th thortical bounds, in LOS conditions, Ultra-Widband offrs a significantly highr ranging accuracy. In NLOS conditions, multipath componnts, containing information valuabl to advancd localization algorithms, can only b rsolvd using Ultra-Widband signals. Th significantly bttr ranging accuracy as wll as th suprior capabilitis in non-lin-of-sight conditions dmonstrat th much highr potntial of Ultra-Widband compard to Wirlss LAN to nabl advancd localization systms in indoor nvironmnts. Indx of dr masurmnt Mssung Fig.. Comparison of LOS and NLOS tim dlay stimats. As it can b obsrvd from Fig., th variation from LOS to NLOS lads to dgradation of accuracy for both tchnologis. Anzahl UWB LOS d UWB NLOS d WLAN LOS d Entfrnungsfhlr <> [m] Fig.. Histograms of LOS and NLOS distanc rrors. WLAN NLOS d REFERENCES [] S. Gzici, T. Zhi, G. B. Giannakis, H. Kobayashi, A. F. Molisch, H. V. oor, and Z. Sahinoglu, "Localization via ultra-widband radios: a look at positioning aspcts for futur snsor ntworks," Signal rocssing Magazin, IEEE, vol., pp. 7-,. [] ARC Wirlss Solutions Inc., ARC Frdom Antnna. What Ridg, Colorado, USA,. [3] S.. Lohmir, R. Rajaraman, and V. C. Ramasami, "Dvlopmnt of an ultra-widband radar systm for vhicl dtction at railway crossings," prsntd at Ultra Widband Systms and Tchnologis,. Digst of aprs. IEEE Confrnc on,. [] J. Schrodr, S. Gallr, and K. Kyamakya, "A Low-Cost Exprimntal Ultra-Widband ositioning Systm," prsntd at IEEE Intrnational Confrnc on Ultra-Widband, Zurich,. [] ANSI/IEEE Std., "art : Wirlss LAN Mdium Accss Control (MAC) and hysical Layr (HY) Spcifications," 999 Edition (R3). [] L. Xiaoming and H. Torp, "Intrpolation mthods for tim-dlay stimation using cross-corrlation mthod for blood vlocity masurmnt," Ultrasonics, Frrolctrics and Frquncy Control, IEEE Transactions on, vol., pp. 77-9, 999. [7] I. Cspds, Y. Huang, J. Ophir, and S. Spratt, "Mthod for stimation of subsampl tim dlays of digitizd cho signals," Ultrason. Imaging, vol. 7, pp. -7, 99.