Copyright 2000 IEEE. IEEE Global Communications Conference (Globecom 2000), November 27 - December 1, 2000, San Francisco, California, USA

Transcription

1 Copyright 2000 EEE. EEE Goba Communications Conference (Gobecom 2000), November 27 - December 1, 2000, San Francisco, Caifornia, USA Persona use of this materia is permitted. owever, permission to reprint/repubish this materia for advertising or promotiona purposes or for creating new coective works for resae or redistribution to servers or ists, or to reuse any copyrighted component of this work in other works must be obtained from the EEE.

2 Space-ime Combining in the Upink of URA/FDD Kaus Kopsa 1,2, Reinod Weinmann 3, Voker Braun 3, and Michae angemann 3 1 nstitut für Nachrichtentechnik und ochfrequenztechnik echnische Universität Wien, Gußhausstraße 25/389, A-1040 Wien, Austria Kaus.Kopsa@mobie.nt.tuwien.ac.at 2 FW, eecommunications Research Center Vienna Maderstraße 1/9, A-1040 Wien, Austria 3 Acate SEL AG, Corporate Research Center, Radio Communications Department ZFZ/R Lorenzstraße 10, D Stuttgart, Germany Abstract - We investigate various array combining agorithms considering the specia requirements of the URA/FDD upink. After formuating the agorithms, we assess their performance in a modified vehicuar B channe that is spatiay inhomogeneous. he bit error ratios in different situations are presented together with some considerations about the computationa compexity of the used methods. From bit error performance and required effort, we concude that the minimum mean squared error (MMSE) agorithm with reduced compexity is the most suitabe method for a rea operating system.. SYSEM LAYOU For bit error performance assessment, we used a baseband ink-eve simuation chain without error correction coding and power contro. he impementation is according to the W-CDMA standard of the URA/FDD upink [1]. An existing singe antenna simuation chain was equipped with a spatia channe mode and up to six antennas.. NRODUCON Co-channe interference in ceuar CDMA-systems is the main imiting factor of transmission quaity and capacity. he origin of this spurious power is mainy twofod. Echoes received via mutipe paths with distinct propagation deays ead to inter-chip interference (C). Secondy, incompetey vanishing crosscorreation between different user s spreading-codes eads to unwanted contributions, caed mutipe access interference (MA). Empoying a tempora RAKE receiver remedies C, since mutipath components can be separated and coherenty summed up to gain maximum signa power. f we additionay use smart antennas at base stations, it becomes possibe to expoit the directiona nature of the mobie radio channe against MA. By coherenty combining a the echoes at different antenna outputs, we are abe to accumuate received signa power, whie simutaneousy suppressing interference. he aim of this paper is to rate the performance and computationa effort of various combining agorithms in different propagation scenarios. Fig. 1: Simuation environment he transmitter generates data channe (DPDC) and contro channe (DPCC) from random data and known piot symbos, performs spreading, scrambing, and puse-shaping. he signa is then fed into the spatia channe, which is modeed as a separate vehicuar B [2] tapped deay ine from the transmitter to every receiving antenna. Each deay tap represents a distinct path with independent fading. n the spatia domain, we can seect two different fading scenarios. ndependent fading among the antennas modes diversity reception, whereas spatiay correated fading corresponds to cosey spaced antennas. n this case the Rayeigh-coefficients at the antennas differ ony by a constant phase factor determined by the path s direction of arriva (DOA). nterference is modeed as an additiona mobie station with constant DOA and variabe power. t uses a different scrambing code and a paths have a constant ange offset and a random time offset. After summing up a

3 paths, independent white Gaussian noise is added at each antenna eement. After puse-shaping, the received antenna signas are fed into the three bocks shown in Fig. 2. Fig. 2: Weight estimation, path searcher, and combiner he path searcher uses a incoming antenna signas to identify the deay times of the dominating paths and deivers them to weight estimator and combiner. Using the methods described in Section 4, the weight estimator computes a weight vector for every detected path. Finay, the space-time combiner mutipies the incoming antenna signas with the compex conjugated weight vector, removes the deay time differences and adds up a contributions to obtain a singe data stream for succeeding despreading. After passing the decision device, the received data is compared to the transmitted to obtain the bit error ratio.. SGNAL MODEL We transmit one data channe (DPDC) on the - branch and the contro channe (DPCC) on the - branch. hese symbo streams are denoted as s and s. ogether with the channeization codes c and c, the compex vaued scrambing code c, and the ampitudes A and A, the transmitted signa of the desired user is given as ( A s ( k ) c ( n) + j A s ( k ) c ( n) ) c( ) d( n) = n, (1) where n and k / denote the chip and symbo index, respectivey. We used a spreading factor of SF =128 for data and SF =256 for contro information, corresponding to 20 and 10 bits per sot, respectivey. he compex vaued scrambing code has aways the fixed ength of 256 chips. n our singe-user detection case, the base band signa received at the M antenna eements is a discrete time convoution sum of the transmitted data with the channe impuse response L 1 = 0 h ( n) = a δ ( n τ ). (2) t consists of L dominant paths, where the integer τ denotes the deay of the -th path in mutipes of the chip duration. he Mx1vector a represents the array response vector of the -th path incuding path oss, fading, and puse shaping. he whoe received signa incuding spatiay and temporay white Gaussian noise n, and interference s N is then given as L 1 x ( n) = d( n τ ) h( τ ) + sn ( n) + n( n). (3) = 0 n the despreading process we can distinguish between the signas of different paths by choosing an according offset. Correating x(n) with the spreading codes yieds the signas y, (k ) and y, (k ), where the index indicates the dependency on the -th path and the symbo indices k and k represent the different bit rates. With these quantities we are abe to define the covariance matrices R xx = E{ xx }, (4) R yy = E{( y + y )( y + y ) } (5),,,,, that we wi need in the foowing section. n practica impementations, the expectation vaue has to be approximated by the time average. V. ALGORMS Weight computation agorithms can be roughy divided into two casses. Maximum ratio combining (MRC) agorithms use an estimate of the channe impuse response as weight vector and are optima ony in the presence of spatiay and temporay white noise. f, on the other hand, significant coored interference is present, we have to use interference suppression methods to achieve optima performance. ere, the impuse response vector is pre-mutipied by the inverse of a suitaby chosen covariance matrix to decrease the infuence of interference. A. Piot Based MRC Since URA empoys piot symbos in the DPCC, the simpest way to obtain an estimate of the channe impuse response is to correate the de-spread contro information with the known piot symbos [3] w = E{ y, s, piot}. (6) As mentioned above, we wi approximate the expectation vaue by a sampe mean over one sot. his robust

4 and simpe way to obtain the w inherenty yieds the absoute channe phase, enabing coherent detection. Averaging data over more than one sot is beneficia ony if the channe stays constant during this time. B. Principa Components (PC) MRC n the piot based method, antenna weights are computed separatey and independenty of each other, whie the averaging process is ony extended over the number of piot symbos in the contro channe. Possiby, principa components methods offer an improvement in performance since they are abe to use a symbos of data and contro channe and consider the correations between the antenna eements. On the other hand, the required eigenvaue decomposition wi make principa components methods more sensitive to noise. With (3), (4) and under the assumptions [4] that noise is temporay and spatiay white, the chip sequence of the desired user is white, the transmitted signa is independent of noise and interference, and a channes are L with a finite duration, we can write the chipcovariance matrix as L A = 0 R = 2( A + ) a a + R. (7) xx he sum contains a the mutipath components of the desired signa, whereas R nn denotes the noise and interference covariance matrix. f we additionay presume that noise and interference in y, and y, are mutuay independent, and that the code chips are binary i.i.d. random variabes, we can write the despread covariance matrix as R L = ( A A ) aa 2( A A C k = 0, k 2 yy, ) ak where the constant nn a k + R nn (8) SF SF C = (9) SF + SF describes the attenuation of noise, interference and mutipath components. For the considered spreading factors we can expect that this attenuation is strong enough that the matrix R yy, is dominated by the term a a. hus, the eigenvector beonging to the dominant eigenvaue of R yy, provides an estimate for the desired array response vector of the -th path [3]. But to obtain the absoute channe phase for coherent detection, we have to additionay correate the eigenvector with the known piot symbos. n the presence of spatiay coored interference we coud aso take the dominant eigenvector of the matrix C R yy, R xx [3] as a weight vector, which is a better estimate of the channe impuse response (coored noise (cn) principa components). But even this approach cannot totay remedy the inherent sub-optimaity of MRC agorithms in the presence of coored interference. C. Minimum Mean Squared Error (MMSE) he Wiener fitering interference suppression approach minimizes the mean squared error between the symbo decision and the transmitted data and is hence superior to MRC in the presence of coored interference. he desired weight vector is given as [5] R 1 a w = xx ˆ, (10) where â is the estimated array response vector. he pre-mutipication by the inverse of the chip covariance matrix suppresses coored noise and interference. o estimate R xx accuratey enough, it might not be necessary to use a chips for computing the sampe mean. By taking just a fraction of the avaiabe chips, we can reduce the computationa compexity significanty (reduced compexity MMSE). D. Optimum Combining (OC) A second interference suppression agorithm is caed optimum combining [6]. t maximizes the signa to noise and interference ratio (SNR) at the combiner output. With (7) and (8), we find that L = R xx R yy, 1 2( A + A ) akak + Rnn C k= 0, k (11) contains ony the unwanted mutipath components, noise, and interference. he weights given as w = 1 ( R xx R yy ) aˆ, thus optimay enhance the SNR [6]. E. 2D Agorithms, (12) Up to now, we computed the weights separatey for each deay. Athough the correation receiver is abe to separate the different paths, we may nevertheess expect

5 an additiona benefit by considering a mutipath components jointy. Using the vectors [4] [ x ( n + τ ) ( n ) ( n ] 0 x + τ1 x + L [ ] ) = y,0 ( k ) y,1( k ) y, L ( k ) = [ y ( k ) y ( k ) y ( k ] x n L ), L( ) = τ 1 y k L ),, L ( 1 y k L ),, L (,0,1, L 1 (13) to create the above mentioned covariance matrices, we can empoy the same weight computation agorithms as in the sequentia space-time case. n the region of ow SNR, noise is the dominating source of error, hence MRC and interference suppression agorithms perform neary equay. Above about 2dB SNR (after despreading), the infuence of interference becomes obvious. he BER of MRC methods saturates, whie interference suppression methods sti offer improvement as noise goes down. n Fig. 4, the advantage of interference suppression agorithms becomes even more cear, as a 32 times stronger interferer is present (SF int =4), whie a other parameters are the same as in Fig. 3. Saturation of the MRC agorithms starts at about 2dB SNR and they can hardy obtain a BER beow 1%. Both MMSE agorithms oose about 1.5dB compared to Fig. 3, but they sti just need about 3dB SNR for a BER of 0.1% V. SMULAONS n simuations with various parameter settings we assessed the performance of the presented agorithms. As a reference, we further incuded a singe antenna receiver. Fig. 3 shows resuts with variabe white noise and one intra-ce interferer with spreading factor SF int =16. o compensate for the ack in spreading gain, the interferer is transmitting with a constant power of 8 times the desired user. Fading was uncorreated among the six antennas, mobie veocity was set to 50km/h and we simuated 10 6 data symbos. BER singe antenna PC MRC 2D PC MRC PC MRC cn MRC piot MMSE reduced MMSE SNR Fig. 3: White noise and one interferer with SF int =16 BER singe antenna 2D PC MRC PC MRC PC MRC cn MRC piot MMSE reduced MMSE SNR Fig. 4: White noise and one interferer with SF int =4 o keep Fig. 3 and 4 more readabe, we presented ony one 2D agorithm and negected optimum combining. As you can see in Fig. 5 (correated fading, variabe powered interference, no noise, mobie speed 50km/h), the increased computationa effort of optimum combining does not ead to a performance gain against MMSE. BER singe antenna MRC piot MMSE OC C/ Fig. 5: Variabe interferer power and negigibe noise

6 A second consequence of Fig. 5 is that MRC is very sensitive to strong interference. his refects the fact that an MRC agorithm is designed for spatiay white noise. Correated fading additionay reduces the gain of MRC against the singe antenna receiver in this scenario. From Fig. 3, 4, and simuations incuding other 2D methods we saw that 2D agorithms generay show no significant performance improvement. his may be due to the high spreading factors we used, enabing good suppression of muti-path components. V. COMPUAONAL COMPLEXY Besides the BER performance, the required computationa effort of the agorithms is important for practica impementation. abe shows the number of foating point operations 1 per sot for the parameter seection and weight computation methods of Fig. 3 and 4. ABLE Required computations for used agorithms agorithm operations singe antenna 430 MRC with piot 2600 reduced compexity MMSE MMSE OC principa components MRC D principa components MRC principa components MRC (cn) Piot based maximum ratio combining requires about six times as many operations as the singe antenna receiver, whie a the other methods are ceary more compex. he dominant source of effort is neither eigenvaue decomposition nor matrix inversion, but a processing that has to be performed on chip-eve. For instance, the creation of the chip covariance matrix R xx is the most compex part of the MMSE agorithm, since we have to average over 2560 data vectors. ere we can save neary a factor of 20 by just using every 32 nd sampe for averaging (reduced compexity MMSE). By far the most expensive methods are a principa components agorithms, because the data channe has to be 1 One foating point operation is either addition, mutipication, subtraction or division. despread to obtain R yy,. he coored noise (cn) principa components agorithm additionay needs the chipmatrix, making it the most compex of a considered methods. V. CONCLUSONS n simuations, we find the MMSE agorithm performing best with strong interference and just ike the other methods with spatiay white noise. ts computationa compexity is higher by more than a factor of 100 compared to the piot based MRC agorithm. But if we use MMSE with reduced compexity, we can save more than a factor of 10 in effort with a performance oss of ony about 0.25dB. With our parameter seection, optimum combining, principa components MRC and a 2D agorithms do not provide sufficient performance benefits to justify their increased compexity. herefore we concude that the reduced compexity MMSE agorithm is the most suitabe one for practica impementation. f however, the system is not abe to hande its computationa demands, we suggest to use the simpe and robust piot based MRC agorithm. ACKNOWLEDGMEN We woud ike to thank the team headed by Eie Bejjani at CRC Nanterre (France) for providing us with the singe antenna simuation chain, and Prof. Ernst Bonek for continuous encouragement. REFERENCES [1] 3GPP, UE radio transmission and reception (FDD) 3G S version 3.0.0, October [2] ES, Seection procedures for the choice of radio transmission technoogies of the UMS R version 3.2.0, Sophia Antipois, France, Apri [3] J. S. hompson, P. M. Grant, B. Mugrew, Performance of antenna array receiver agorithms for CDMA, Signa Processing, vo. 68, no. 1, pp , Juy [4]. Liu, M. D. Zotowski, Bind equaization in antenna array CDMA systems, EEE ransactions on Signa Processing, vo. 45, no. 1, pp , January [5] S. aykin, Adaptive fiter theory, Prentice-a, 590 p., [6] J.. Winters, Optimum combining in digita mobie radio with cochanne interference, EEE Journa on Seected Areas in Communications, vo. 2, no. 4, pp , Juy 1984.