Wideband characterization of the urban PCS channel

Transcription

1 Wideband characerizaion of he urban PCS channel Aris L. Mousakas, Sridhar Arunachalam, Kam H. Wu and Howard Heller Wireless Advanced Technology Laboraory Lucen Technologies Bell Laboraories 67 Whippany Rd, Whippany, NJ 07981, USA ABSTRACT In his paper he analysis of urban wideband (3MHz) channel measuremens a 1.9GHz is presened. The mulipah behavior of he channel is characerized hrough he power saisics of resolvable pahs and heir delay-spread. The correlaion beween differen pahs is found o be very small, while he Ricean facors are moderae in size. The spaial correlaions beween anennas a he base saion are high and in agreemen wih he narrow angle-spread measured a he base saion. To obain spaial informaion of he channel a he mobile we employ a mehod similar o synheicaperure radar. Surprisingly we find high spaial correlaions a he mobile, indicaing narrow anglespreads. This may prove o be a limiing facor in he operaion of closely spaced muliple anenna (MIMO) mobiles. I. INTRODUCTION The evoluion of hird-generaion wireless elecommunicaions sysems demands he allocaion of wider bandwidhs. As a resul, exising propagaion models for narrower bandwidhs have o be modified, in order o describe accuraely he wireless channel a shorer ime scales. In addiion o aking ino accoun addiional resolvable pahs, i is imporan o analyze he power disribuion associaed wih he differen pahs, as well as he angular spread associaed wih each pah. These issues become especially imporan when inelligen anenna echnologies ( smar anennas ) are used in ransmission/recepion. To his end, exensive measuremens are necessary o provide a basis for an accurae model of he channel. In his work we repor on a recen experimenal effor in Newark, New Jersey o characerize a wideband (3MHz) oudoor urban channel in he PCS band (1.9 GHz). II. EXPERIMENTAL SETUP A base saion (BS) anenna was placed in hree locaions a or above roof level and a mobile monopole omnidirecional anenna was mouned a he roof of a van. The mobile ransmied a coninuously repeaed PNsequence of 63 symbols a a rae of 3MHz and a a frequency of MHz. The base saion anenna consised of an 8-elemen array wih columns spaced a 7.5cm 0.46λ, wih wo passive elemens on each side o minimize mismaches beween columns. The van was driven around each base saion locaion wih speeds beween 8-40km/h and measuremens were aken a abou 10 locaions, chosen o cover he beamwidh of he array (10 o ) uniformly up o a radius of abou km. Afer he daa was calibraed and filered, he signal from each anenna was separaed ino blocks of duraion 0.666msec and each block was muliplied wih a sring of PN-sequences wih differen delays, o give h µ (,τ), he channel a ha ime, wih delay τ and a BS anenna µ. III. FINGERS DELAY PROFILES We firs address he emporal disribuion of power. As he bandwidh of he signal increases, he number of resolvable pahs is expeced o increase. Several ineresing quesions arise, such as: Wha diversiy gains do he addiional fingers give and wha is heir delayspread? How correlaed are he various resolvable pahs wih each oher and wha saisics does he auocorrelaion of each finger follow? In order o address hese quesions, one mus specify wha is mean by a resolvable pah. Two definiions are used, as follows: 1. In he firs definiion (denoed loc-max or lm ), we ake he local maxima of he power delay profile P(,τ) as a funcion of delay τ for each ime. Three resricions are imposed. Firs we keep pahs wih power no less han 10% of he power of sronges finger. Second, heir power is greaer han 10 imes he noise floor. Third, we limi he maximum number of fingers o 10.. In he second mehod we ake he local maxima of he power delay profile P(,τ) averaged over he whole measuremen ime ( 10sec). We use his mehod when deermining correlaions beween fingers (denoed as average fingers ) over ime, because i focuses a he behavior of paricular delay imes. A maximum of hree local maxima of he average power delay profile are colleced, wih he condiion ha heir average power is no less han 30% of he maximum average power and a leas 10 imes larger han he average noise floor. A. Finger power disribuion To address he emporal diversiy gain from muliple resolvable pahs, we calculae he cumulaive disribuion funcion (CDF) of he oal power received from several fingers. This provides a way o analyze he exra power received from secondary fingers. Figures 1A-C are he

2 plos of he CDFs of powers received from he maximum finger, as well as he, 4, 10 (all) sronges fingers. We see ha he gains are generally locaion-dependan. In Fig. 1A (Newark Garage SW) he use four fingers gives a 5dB gain a he 1% ail of he disribuion. In conras, in he oher wo locaions he second finger gives abou db exra power wih no exra gain from more fingers. RAKE receiver. Among several ways used in he lieraure o characerize delay-spread[1,, 3], here we use wo such mehods[1] and focus on he delay-spread of he resolvable fingers. Firs, we deermine he convenional delay-spread τ, given by he spread of arrival ime of resolvable pahs, independen of heir power. The average value < τ>=0.95µsec, corresponds o disances of abou 300m. This is he expeced ypical disance[1] beween scaering ceners for a mobile a abou 1km away from he base-saion, which was he case in our measuremens. To focus more on he relevan pahs, i.e. he ones wih he highes received power, one has o weigh he delays wih heir corresponding power. In his case, he delay-spread[1] τp has an average of < τp>=0.85µsec. The 1% difference beween < τp> and < τ> is consisen wih he lieraure[1]. The associaed CDFs of τ and τp over all BS locaions appear in Fig.. C. Correlaion beween differen fingers Figure 1A Muliple resolvable pahs are useful no only for effecively increasing he received signal power, bu also because hey provide robusness hrough emporal diversiy, because hey are expeced o fade independenly. To address his issue, we analyze he correlaions beween differen resolvable pahs. We use he average finger mehod described above (mehod ), o ensure ha we are looking a hree pahs wih high average power. The normalized correlaion C(τ,τ ) beween wo fingers wih delays τ and τ is given by * å h1 (,τ )h1 (,τ ') (1) C (τ,τ ') = å h1 (,τ ) å h1 (,τ ') Figure 1B where he sum over is over he whole ime inerval of he measuremen (approx. 10sec) and h1(, τ) is he channel a BS anenna 1 a ime and delay τ. A meric ha describes he degree of correlaion beween differen fingers is he deerminan of C(τ,τ ) in erms of he delay/finger indices τ and τ. Fig. 3 is he plo of he CDF of de(c(τ,τ )) over all locaions, confirming low correlaions beween fingers. Figure 1C Figures 1A-C: Disribuion of power received from maximum finger, maximum fingers, 4 maximum fingers and all fingers. B. Delay-spread disribuion Delay-spread is an imporan propery of he channel, for general modeling purposes, as well as he design of a Figure : CDFs of power weighed and non- power weighed delay-spreads for loc-max fingers.

3 spread gives low anenna correlaion and vice versa. Assuming saisically independen incoming waves, he correlaion Rµν beween anennas µ and ν a he receiver is given by[4] Rµν = Figure 3: CDF of he deerminan of finger cross-correlaions (in db). For comparison, for 3 fingers wih C1= C3=0.14, de[c] =- 0.5dB D. Rice K-facors The flucuaions of he signal power from a paricular pah provide informaion of he coherence of he ampliude of he pah, which can be quanified by he Rice K-facor, he raio of he coheren o incoheren power.[] This parameer can be obained by fiing he measured probabiliy densiy funcion (PDF) of he individual finger power. In Figure 4 he CDF of Kfacors over all measured locaions is ploed. We see ha mos locaions have small K s, wih an average of abou.3 and a median value of 1.3. IV. SPATIAL SIGNATURE AT BASE STATION We now urn o he analysis of he spaial signaure of he channel a he BS. For simpliciy we only analyze he maximum average finger. Firs, we calculae he correlaions beween differen anennas and plo he CDFs of he correlaions for differen anenna separaions in Fig. 5. As expeced, he anennas a he base saion are srongly correlaed. Figure 4: CDF of Ricean K-facors over all locaions (K = raio of coheren o average incoheren power) Kave =.3 and Kmedian=1.3. Alernaively, one can aemp o exrac he angle of arrival of he incoming rays and heir angle-spread. Usually he wo approaches are relaed: Large angle ò dφ ò dφ χ µ (φ ) χν (φ )* p(φ ) () χ µ (φ ) p(φ )ò dφ χν (φ ) p (φ ) where χµ(φ) is he response of anenna µ in direcion φ and p(φ) is he average incoming power from he same direcion. The angle-spread is hidden in he power disribuion p(φ). I is cusomary o assume a Gaussian angle-of-arrival disribuion for he power, wih anglespread defined as he sandard deviaion of he Gaussian. In his case, for small enough, ideal anenna paerns and he direcion of arrival θ0 no oo close o end-fire (±900) he correlaion magniude beween anennas can be expressed as a funcion of one parameer[5,6] é æ πd ( µ ν ) cos θ 0 ö ù (3) R µν = exp ê 0.5ç ú λ è ø úû êë namely πd(µ-ν) sin(θ0)/λ, where θ0 is he mean angle of arrival measured from he broadside of he array. Exracing he underlying angle-spread from he daa is no rivial, due o he presence of noise and he finie number (eigh) of anennas in he BS array, especially for angle-spreads narrower han he inheren minimum angle-widh of he anenna array (approximaely 10). To ge around hese problems we look a he flucuaions of he apparen angle of arrival (AOA) a he base saion. Fig. 6 depics he CDFs of he apparen angle-spread a differen BS locaions. The variance of he AOAs is more pronounced in he Newark Garage SW and Newark Park locaions (rms. of apparen angle-spreads are 110 and 180, respecively) raher han he NJIT locaion (60). This is due o he fac ha, unlike NJIT, he oher wo locaions have high-rising buildings near he BS locaion, which ac as reflecors. Figure 5: CDFs of correlaions beween anennas separaed by 0.5 λ, λ, λ, 3λ, for all locaions and all 3 average fingers. To ge a more direc esimae of he underlying anglespread, we need o assume a paricular model for he incoming waves. Based on his model one can relae he sandard deviaion of he apparen AOA of he incoming waves o hrough simulaions or by analyic mehods

4 presened elsewhere.[7] Thus by measuring he observed flucuaions of he apparen AOA one can obain. Figure 6: CDFs of apparen angle-spreads a 3 locaions where measuremens were aken. The red curve, (NJIT) has he smalles angle-spread, since he BS array had no higher buildings in he immediae viciniy, unlike he oher wo locaions. wih close delay imes bu differen angles of arrival inerfere wih each oher. To demonsrae his effec we focus on one paricular sample. The disribuion of angle of arrival (AOA) displays a clearly bimodal behavior (see Fig. 8). For simpliciy, he angular dependence of he power disribuion is aken o be a sum of wo Gaussians wih a variable power raio and angle-spread, cenered a he wo main arrival angles aken direcly from he daa (solid curve in Fig. 8). The disribuion of AOAs was deermined hrough simulaion for various power raios and angle-spreads and he opimal one is ploed in Fig. 8 (dashed curve). The opimal power raio (0.4) is close o he observed one. Using he opimal power raio and angle-spread (.50) we can calculae he expeced correlaions beween anennas, which are shown in Fig. 9. The calculaed curve very accuraely capures he beaing of he correlaion as a funcion of disance, which is due o he differen AOA of he wo main peaks. Figure 8: PDF of AOA (solid curve) and fi o bimodal Gaussian disribuion wih power raio 0.4 and individual angle-spreads 30 (dashed curve) Figure 7: Correlaion beween anennas a BS, separaed by d= λ/ and λ (blue and red crosses) respecively, versus normalized angle-spread (see Eq.3). The solid curve is he plo of Eq.3. In abou /3 of he locaions where measuremens were aken, he angular disribuion of power can be described a single peak wih a cerain widh. In hese cases we used a single Gaussian disribuion of incoming rays wih sandard deviaion around is AOA. Using his model we obained he rue angle spread, wih an rms. value of rms=90. I is insrucive o check he anglespreads obained by his mehod versus he corresponding correlaions beween anennas for each locaion. The simple model described in Eq. 3 was used and Fig. 7 is he resuling scaer-plo for he above cases. One can see ha he agreemen beween he approximae heory and he daa is saisfacory (see also ref [8]). For he res of he locaions i was no possible o obain any agreemen wih a single peak Gaussian, since he incoming waves followed mulimodal disribuions, corresponding o narrow beams from differen direcions. This effec can be undersood by he fac ha fingers Figure 9: Correlaions beween anennas 1 and -8 (solid curve) for same locaion as Fig. 8. The dashed curve resuls from he disribuion fi in Fig. 8. V. SPATIO-TEMPORAL CORRELATION AT MOBILE Nex we analyze he behavior of he channel a he mobile hrough he emporal auo-correlaion funcion G(τ) of he signal for he average fingers. G(τ) is simply

5 he Fourier ransform of he Doppler spread frequency profile. There are various models for G (τ) in he lieraure[,9], mos noably ha of Clarke: G ( τ ) = J 0 (πf Dτ ) (4) where f D is he Doppler frequency relaed o he vehicle speed υ (f D =υ/λ) and J 0 is he zeroh order Bessel funcion. The underlying assumpion in his model is ha all rays emied from he mobile in all direcions arrive a he base saion wih he same srengh (bu random phases). Using he same assumpions, he spaial correlaions a he mobile ake exacly he same form, namely J 0 (πd/λ). In general, he form of he emporal correlaions of he signal depends on he disribuion of scaerers around he mobile in he same way he spaial correlaions a he mobile do. This is a poin ha has no been emphasized enough in he lieraure. One can exploi he relaion beween emporal and spaial flucuaions a he mobile. In paricular we will use a mehod similar o synheic aperure radar[1] o obain he spaial correlaions a he mobile. The approach is similar o he one discussed above for he case of he BS array. To ge an esimae of he spaial correlaions beween poins separaed by d a he mobile, we use he emporal auocorrelaion funcion G (τ) evaluaed a ime τ=d/υ, where υ is he measured velociy of he mobile. Even hough his approach gives a lower bound for he spaial correlaions, since i does no ake ino accoun he change of he environmen iself in a msec scale (e.g. a ruck passing by), he obained correlaions are relaively high (see Fig. 10) a shor disances, suggesing narrow angle-spreads. This is somewha surprising, bu if one akes ino accoun he canyon srucure of he srees in an urban environmen, i cerainly sounds reasonable. The exisence of high spaial correlaions and narrow angle-spreads around he mobile becomes imporan when muliple anennas are closely placed a he mobile. The analysis performed here is only a firs aack o he problem. However, given ha unil now wide anglespreads have always been assumed a he mobile, fuure daa acquisiion should focus on his issue. VI. SUMMARY In conclusion, we have performed a deailed quaniaive analysis of he PCS urban channel wih a bandwidh of 3MHz. The mulipah behavior of he channel was sudied in erms of he saisics of he power of resolvable pahs. The channel was found o be Ricean wih median Ricean facor around 1.3. The spaial signaure of he channel a he BS was found o behave as a sum of narrow peaks from differen direcions. This was aribued o he inerference of differen fingers wih shor relaive delays. The spaio-emporal channel behavior a he mobile was analyzed using a mehod called synheic aperure analysis. This analysis was no as accurae as desired due o he limiing facor of he single mobile anenna. However, we found ha, surprisingly, he angle-spread a he mobile, albei wider han ha a he BS, was sill narrow enough in several cases o cause high correlaions beween differen poins. This finding meris more accurae measuremens in he fuure. Figure 10: CDFs of correlaions beween anennas separaed by λ/ a mobile. The large correlaions a small disances sugges narrow angle-spreads The auhors wish o acknowledge useful discussions wih R.D. Benning, R.M. Buehrer, P. Monogioudis and W. Wynn. REFERENCES [1] K.I. Pedersen e al, A sochasic model of he emporal and azimuhal dispersion seen a he base saion in oudoor propagaion environmens, IEEE Trans. Vehic. Technol, vol. 49, 000, pp [] J.D. Parsons, The mobile radio propagaion channel, nd ediion, John Wiley & Sons, (000). [3] J.C. Liberi and T.S. Rappapor, Smar anennas for wireless communicaions: IS-95 and hird generaion CDMA applicaions, Prenice Hall (1999). [4] A.L. Mousakas e al, Communicaion hrough a diffusive medium: coherence and capaciy, Science vol. 87, 000, pp [5] H. Carrasco Espinosa e al, Performance evaluaion of space-ime block coding using a realisic mobile radio channel model IEEE Proc.VTC000, 000,pp [6] R.M. Buehrer, Generalized equaions for spaial correlaion for low o moderae angle-spread Proc. 10 h Virginia Tech Symp. Wireless Personal Communicaions, 000. [7] A.L. Mousakas e al, o be published. [8] S. Arunachalam e al, Spaial Channel Model and Measuremens for IMT-000 Sysems, submied o IEEE VTC Spring 001. [9] W. Jakes (ed.), Microwave Mobile Communicaions Wiley-Inerscience, (1974). [10] J.P. Fich, Synheic Aperure Radar, New York, Springer Verlag (1988).