Interference cancellation in multirate DSCDMA systems

Interference cancellation in multirate DS/CDMA systemsAnn-Louise Johansson and Arne SvenssonChalmers University of Technology, Dept. of Information Theory, S-412 96 Gothenburg, Swedene-mail: anne.johansson@it.chalmers.se, arne.svensson@it.chalmers.seAbstract : This paper presents multistage interference cancellation (IC) schemes for the uplink of multirate direct-sequence code division multiple access (DS/CDMA) systems.The performance is evaluated both analytically and via computer simulations for two multiple data rate schemes, mixed modulation and parallel channels, assuming flat Rayleigh fading and multipath environments.IntroductionIt is anticipated that an important feature of future mobile communication systems will be the ability to handle other services besides speech, e.g., facsimile, Hi-Fi audio and transmission of images; services that are not readily available today. To achieve this, a multiple-access method which is flexible, has the prospect of high sys-tem capacity, and the ability to handle variable data rates, is necessary. Direct-sequence code division multiple access (DS/CDMA) is believed to be a multiple-access method able to fulfil these requirements.The conventional multiuser detector is composed of a bank of matched filters, which is the optimal structure for a single user channel corrupted only by additive white Gaussian noise (AWGN) [1]. The presence of a num-ber of users in the system often introduces multiple-access interference (MAI), which may lead to an irreducible error probability [2]. Typically, MAI arise in an asynchronous system when the users’ signals are received with different time-lags, which usually increases the cross-correlation between the spreading sequences.Another important issue for CDMA systems, in addition to the MAI, is the near-far problem [3]. In the reverse link (mobile to base station), the signals are received with different powers. The current most referred solution to this problem is stringent power control [4], though over the last few years a lot of attention has been given to the area of multiuser detectors [2]. These have the prospect of both mitigating the near-far problem and cancelling the MAI. One subclass of multiuser detectors is the class of detectors that employ interference cancellation (IC)schemes. The structures of the IC schemes differ depending on how the users are detected and cancelled. In [5]-[7] the detection and cancellation is done in parallel for all the users and in [8]-[11] the detection and cancellation is done successively (or serially), one user at the time. There are also hybrid schemes where the detection and cancellation is done both successively and in parallel [13].In [11] and [14] two different methods to handle multiple data rates in DS/CDMA systems were analyzed. One way to do this is to allow different forms of modulation. The spreading factor is equal for the users in the system,hence, for transmission of a specific rate, the user chooses a modulation format, e.g., one of BPSK, QPSK, or 16-QAM. A second approach to implementing multiple data rates is to allow each user to transmit over one or sev-eral parallel channels. That is, the users employ several signature sequences and transmit the signals synchro-nously.For the flat Rayleigh fading channel we consider non-decision directed interference cancellation (NDDIC)[11] and for the multipath environment we consider a multistage hybrid IC scheme [12]. To take advantage of the diversity in the system we employ a RAKE receiver. The RAKE demands full knowledge of the channel parame-ters and therefore it seems natural to include decision directed (DD) IC, which also adopts this knowledge. How-ever, instead of employing a multistage DDIC scheme together with a RAKE [10], a less complex hybrid IC scheme with a combination of NDDIC and DDIC is presented.System ModelWe consider a model for a system with square lattice QAM signalling over slow, frequency-selective Rayleigh fading channels. The received signal is the sum of the reflections of all the signals embedded in AWGN with a two-sided power spectral density of . Assuming that there are users and paths per user, the received composite signal is(1)This work was supported by the Swedish National Board for Industrial Technical Development, NUTEK. Project 9303363-2.N 02⁄K P r t ()αk p ,p 1=P∑2E 0T---------d k I t τk p ,–()c k I t τk p ,–()ωc t φk p ,+()cos d k Q t τk p ,–()c k Q t τk p ,–()ωc t φk p ,+()sin +{}n t ()+k 1=K ∑=The information-bearing signal,, for all the users, is an infinite sequence of rectangular pulses of dura-tion T with amplitude. The amplitudes of the quadrature carriers for the user’s symbol element, and , generate together M equiprobable and independent symbols. The amplitudes are then affected by the channel gain of each path,, which are assumed to be i.i.d. random Rayleigh variables. Each user k hastwo signature sequences, and , which are used for spreading the signal in the in-phase (I) and the quadra-ture (Q) branch, respectively. They consist of sequences of antipodal, unit amplitude, rectangular pulses of dura-tion . The period of all the users’ signature sequences is , so there is one period per data symbol.The variable is the time delay and is the phase of the user’s path. These are, in the asynchro-nous case i.i.d. uniform random variables over and , respectively. Initially, the only parameters which are assumed to be known are the time delays of the various users.Multistage Interference CancellationThe multistage interference canceller (IC) for M-ary QAM and flat Rayleigh fading is composed of a bank of filters matched to the I and Q spreading sequences of each user. Initially the users are ranked in decreasing order of their respective received signal power. Then the output of the matched filter of the strongest user is used to estimate that particular user’s baseband signal, which is subsequently cancelled from the composite signal. In other words, the projection of the received signal in the direction of the spreading sequence of the strongest useris subtracted from the composite signal. This is illustrated in Fig. 1. Let denote the complex sum of and . The variable defines the composite baseband signal after cancellation of user and define the estimated signal of the user. The superscript defines the stage of the scheme. (In the first stage of the scheme,which is described above, is zero.) In this way we attempt to cancel the interference, which the remaining users are exposed to. We continue by cancelling the second strongest user, successively followed by all the other users. The cancellations are however not perfect. The filter output contains, besides the desired signal, also Gaus-sian noise and noise caused by MAI and hence, for each cancellation noise is projected in the directions of the other users in the system.If additional stages of the IC scheme is employed the MAI can be further reduced and, consequently, the esti-mates of the signals can be improved. In a multistage scheme we add the estimated signal from the previous stage(corresponding to in Fig. 1a) to the resulting composite signal and use the output of the matched filter to obtain a new estimate of the signal, which in turn is cancelled. The scheme is repeated for all the users for a desired number of stages. This is illustrated in Fig. 1b. The IC scheme reduces the MAI between the users but, on the other hand, the Gaussian noise is enhanced gradually through projection. However, it is shown in [10] that the multistage IC performs better than the decorrelator after a limited number of stages and that they are asymptoti-cally equivalent.Multistage Interference Cancellation in a Multipath EnvironmentIn a multipath environment we have to employ a RAKE receiver to take advantage of the diversity of the chan-nel. The RAKE receiver demands full knowledge of the channel parameters and it seems obvious that a DDIC scheme should be used instead of an NDDIC scheme. The complexity of the receiver can, however, be reduced by employing a combination of NDDIC and DDIC [12].The proposed RAKE receiver with a hybrid IC scheme works as follows. First we employ a few stages ofFig. 1. (a) Interference cancellation unit. (b) Multistage non-decision directed interference cancellation.d kI Q /t ()A k l,I Q /k th l th A k l ,I A k l,Q αk p ,c k I c kQ T c N T T c ⁄=τk p ,φk p ,k th p th 0T ),[02π),[τk p ,c k c kI c k Q r k 1–i k 1–d k i k th d k0d k i 1–++MF kr k 1–i d ki 1–+++–r ki c kd ki IC 1IC 2IC KIC 1IC 1IC 2IC 2IC KIC Kr 01r 11r K 1–1r K1r K2d K1r K 1–2d 21r 12d 11d 1i 1–d 1i d 2i 1–r 1i d 2i d Ki 1–r K 1–i d Ki r Ki (a)(b)NDDIC where, in contrast to [11], the demodulation of the signal is noncoherent [10]. This does not change any-thing for the IC scheme, since knowledge of the phase is only needed for the coherent detection. Besides that, the only difference in the case of multipaths is that all the paths of each user are treated as separate users. They are ordered independently and cancelled one by one according to received signal power, in the manner shown in Fig.1b. The receiver can therefore be visualized as a bank of matched filters, one filter for each path, which evi-dently is equivalent to sampling each of the original filters times. After a few NDDIC stages the MAI in each path has been reduced and we estimate the channel parameters of the paths using, e.g., pilot symbols. These estimates are used to rank the users according to their total received power.In the last stage of the hybrid IC a RAKE is combined with DDIC to benefit from both removal of the Gauss-ian noise and the diversity. A block diagram of a DDIC unit is shown in Fig. 2a, where the resulting baseband sig-nal before the DDIC unit is . Let denote the signal estimate of the user’s path. The variable is one of the signal estimates, which from the outputs of the NDDIC unit belong to the user.This is illustrated in Fig. 2b.For the DDIC scheme, in Fig. 2a, a similar procedure as described for the NDDIC is initially followed. First new estimates of the signals corresponding to the strongest user’s paths are obtained. This is done successively since, if the received multipath signal is regenerated and a conventional RAKE [1] is employed, the intersymbol interference (ISI), which is partly removed in the NDDIC stages, would be introduced again. Hence, in the mod-ified RAKE the signals are fed separately to the combiner. After the symbols have been detected, the estimated channel parameters are used to regenerate the signals corresponding to the user’s all paths, which are then can-celled from the composite baseband signal. The scheme is repeated successively for all the users. This is shown in the right part of Fig. 2b. If more than one stage of RAKE and DDIC is implemented the regenerated signals,corresponding to the various paths, are fed to the subsequent stage in the same manner as for the first stage.Performance Analysis and SimulationsThe graphs below contain, together with simulation results, analytical performance estimates. These estimates have been derived using a Gaussian approximation of the MAI. This approximation seems to be feasible for values up to at least 20 dB both for systems employing the NDDIC scheme and for systems employing the hybrid IC scheme, although it may be considered crude. Especially for the hybrid scheme, since we consider DDIC together with a RAKE in the last stage. In the analysis we have considered NDDIC for that stage too, with the difference that we have regarded known channel parameters and maximal ratio combining of the paths of the various users. Consequently, the pdf’s of the ordered sums of signals, belonging to each user, are used in the anal-ysis of the last stage. With that exception, the analysis is equivalent to the one presented in [11].We have considered asynchronous systems and slow, flat Rayleigh fading or frequency-selective Rayleigh fad-ing with two equally strong, independently fading paths per user. In all the simulations the different IC schemes operates block-wise on the data where two guard symbols have been employed in the beginning and the end of each block to avoid edge effects. It is assumed that the channel do not change over the time of the transmission of a block, which corresponds to slow vehicle speed. Pilot symbols are also employed in the simulations when esti-mation of the channel parameters is considered. In those cases, the estimate is obtained from the average of two pilot symbols in the beginning as well as in the end of each block of data. Known channel parameters have been used for the ranking in the simulations of the systems with only NDDIC. The ranking in the NDDIC part of the Fig. 2. Decision directed interference cancellation in combination with a modified RAKE.(b) Hybrid interferencecancellation scheme.P KP K P r 'k 1–d p k ,'k th p th d p k ,'P KP k th E b N 0⁄d j k,''∑d P k,''d 2k,''d 1k ,''IC 1IC 2IC P RAKE++Channeld 1k,'d 2k,'d P k,'r 'k 1–b kc kr 'kRAKE 1+ DDICRAKE 2+ DDICRAKE K + DDIC123KL12Ki Stages ofNDDIC Group the Paths &Orderr 01r KPi d KPi d 1i d 2i d 3i d 1'd 2'd K 'r 'Kr 'K 1–r '2r '1b 1b 2b K(a)(b)hybrid IC scheme is performed using the output of the matched filters, while the ranking before the stage with RAKE and DDIC is performed using known or estimated channel parameters.All the simulated systems are chip-rate sampled, which limits the possible time-lags between users to be mul-tiples of chip-times. This is a restriction that should be modified for simulations of more realistic scenarios. It should also be noted that because of long simulation runs, the confidence level for high values of is not sufficiently low to give completely accurate results.Mixed Modulation SystemOne way to handle multiple data rates is to allow different modulation formats. We refer to this as a mixed modulation system. The analytical and simulation results are presented in Fig. 3. The systems include BPSK,QPSK, and 16-QAM users and the average SNR/bit is equal for all the users independent of modulation format.The graphs depict the average bit error rate (BER) of a system with 20 BPSK, 10 QPSK and 5 16-QAM users.(We assume Gray encoding of the symbols.) Gold sequences of length 127 are used for the spreading. In Fig. 3a the results employing NDDIC schemes in flat fading is shown and in Fig. 3b the performance of a hybrid IC scheme in two-path Rayleigh fading is presented. The average BER for each kind of user derived from the simu-lations is also shown in the graphs. The graphs depict that the results of the simulations agree well with the ana-lytical results for up to about 20 dB except for the single stage NDDIC scheme. This can be explained by the difference in received power of the various users which makes the Gaussian approximation of the MAI less accurate. It can also be observed that the BER of the 16-QAM users dominates the average BER of the system.Parallel Channel SystemThe other multiple data rate scheme studied is parallel channels, which implies that each user transmits over one or several synchronous channels. We then know that the received signals will be affected by the same chan-nel parameters and we can also make use of spreading sequences with good cross-correlation properties. In Fig. 4the performance of a system with 15 QPSK users, transmitting over two channels each, are shown. Fig. 4a depicts analytical and simulation results for flat Rayleigh fading. The simulations support well the analytical esti-mates. In Fig. 4b the performance of the hybrid IC scheme is presented considering two equally strong multip-aths. The performance is simulated for both known and estimated channel parameters.The graphs show that the performance of the parallel channel system is close to single user performance in both flat and frequency-selective fading and the degradation due to estimated channel parameters is about 1 paring the performance of the two multirate schemes for the same throughput, it can be seen that the perfor-mance of the parallel channels surpasses that of the mixed modulation system.ConclusionsWe have presented non-decision directed IC schemes for flat Rayleigh fading channels and a novel multistage hybrid IC scheme for multipath environments, which is evaluated together with two different multiple data rate schemes, mixed modulations and parallel channels. The hybrid IC includes a combination of non-decisionFig. 3. (a) NDDIC in a mixed modulation system with 35 users. Analytical and simulation results for a flat Rayleigh fading channel. (b) Simulation of asynchronous mixed modulation system with 35 users, 2 paths per user and perfect channel estimates. The performance of the conventional detector and the hybrid IC is shown together with analytical results. The hybrid IC has 4 NDDIC stages and 1 DDIC stage with RAKE.E b N 0⁄E b N 0⁄E b /N 0 (dB)5101520253010−410−310−210−110A v e r a g eB i t E r r o r P r o b a b i l i t yProcessing Gain = 127Random Sequences Mix (20/10/5)Conventional Sim. 1, 2 & 5 St.Ave. BPSK Ave. QPSK Ave. 16−QAM Single BPSKE b /N 0 (dB)5101510−410−310−210−110A v e r a g eB i t E r r o r P r o b a b i l i t yProcessing Gain = 127Gold Sequences Mix (20/10/5)Conventional Hybrid Sim. (4+1) Ave. BPSK Ave. QPSK Ave. 16−QAM Hybrid Analysis Single BPSK, 2 raysdirected and decision directed IC combined with a modified RAKE receiver.Analytical results using the Gaussian approximation are presented and they agree relatively well with the results from computer simulations. Performance close to the single user bound is obtained for perfectly known channels. In the case of channel estimation, the degradation of the performance is relatively small ( dB).Comparing the performance of the two evaluated multirate schemes, parallel channels are preferable for high rate users.In future work the restriction to slow fading will be relaxed and other methods for channel estimation, besides pilot symbols, will be investigated.AcknowledgementThe authors would like to acknowledge Karim Jamal at Ericsson Radio Systems for stimulating discussions and for his invaluable assistance in obtaining the simulation results.References:[1]J.G. Proakis,Digital Communications , 3rd ed, McGraw-Hill, 1995.[2] A. Duel-Hallen, et al., “Multiuser detection for CDMA systems”IEEE Personal Communications , vol. 2, no. 2, pp. 46-58, 1995.[3]R. Lupas and S. Verdu, “Near-far resistance of multiuser detectors in asynchronous channels,”IEEE Trans. on Communications,vol.COM-38, pp. 497-507, April 1990.[4]K. S. Gilhousen, et al., “On the capacity of a cellular CDMA system,”IEEE Trans. on Vehicular Technology , vol. 40, pp. 303-311, May1991.[5]M. Varanasi and B. Aazhang, “Multistage detection in asynchronous code-division multiple access communications,”IEEE Trans. onCommunications,vol. COM-38, pp. 509-519, April 1990.[6]S. Striglis, et al., “A multistage RAKE receiver for improved capacity of CDMA systems,”Proceedings,VTC’94 (Stockholm, Sweden),June 1994, pp. 789-793.[7]Y .C. Yoon, et al., “A spread-spectrum multiaccess system with cochannel interference cancellation for multipath fading channels,”IEEEJournal on Sel. Areas in Communications,vol. 11, pp. 1067-1075, Sept. 1993.[8]P. Patel and J. Holtzman, “Analysis of a simple successive interference cancellation scheme in a DS/CDMA System,”IEEE Journal onSel. Areas in Communications,vol. 12, pp. 796-807, June 1994.[9]M. Ewerbring, et al., “CDMA with interference cancellation: A technique for high capacity wireless systems,”Proceedings,ICC’93(Geneva, Switzerland), May 1993, pp. 1901-1906.[10]K. Jamal and E. Dahlman, “Multi-stage serial interference cancellation for DS-CDMA,”Proceedings, VTC‘96 (Atlanta, Georgia),April1996.[11]A. Johansson and A. Svensson, “Multi-stage interference cancellation in multi-rate DS/CDMA systems,”Proceedings, PIMRC‘95 (Tor-onto, Canada),Sept. 1995, pp. 965-969.[12]A. Johansson and A. Svensson, “Multistage interference cancellation for multirate DS/CDMA on a mobile radio channel,”Proceedings,VTC‘96 (Atlanta, Georgia),April 1996.[13]Y . Li and R. Steel, “Serial interference cancellation method for CDMA,”Electronics Letters , vol. 30, pp. 1581-1583, Sept. 1994.[14]T. Ottosson and A. Svensson, “Multi-rate schemes in DS/CDMA systems,”Proceedings, VTC‘95(Chicago, Illinois), July 1995, pp.1006-1010.Fig. 4. (a) NDDIC in a system with parallel channels. Analytical and simulation results of asynchronous systems with 15 QPSK users, 2 paths per user, signalling over flat Rayleigh fading channels. (b) Simulations of asynchronous systems with 15 QPSK users, 2 parallel channels per user and 2 paths per channel. The performance of the conventional detector, the hybrid IC for known channel parameters, and the hybrid IC for estimated channel parameters, is shown. The hybrid IC has 4 NDDIC stages and 1 DDIC stage with RAKE.E b /N 0 (dB)5101520253010−410−310−210−110A v e r a g eB i t E r r o r P r o b a b i l i t yProcessing Gain = 128Orth. Gold Sequences QPSK, 15 users, P=2Conventional Sim. 1,2 & 5 St. Analysis 1 & 2 St. Single BPSKE b /N 0 (dB)510152010−410−310−210−110A v e r a g eB i t E r r o r P r o b a b i l i t yProcessing Gain = 128Orth. Gold SequencesQPSK, 15 users, P=2,Conventional Hybrid (known ch. 4+1)Hybrid (est. ch. 4+1) Single BPSK, 2 rays1≈。