Transmission Adaptation for Broadband Wireless MIMO-OFDM Systems with Limited Feedback

Transcription

1 Transmission Adaptation for Broadband Wireless MIMO-OFDM Systems with Limited Feedba Harri Pennanen and Jouo Leinonen Centre for Wireless Communiations P.O. Box 4500, University of Oulu, Finland Juha Ylitalo Eletrobit Ltd Tutijantie 7, Oulu, Finland ABSTRACT This paper studies lin adaptation tehnique for multiple-input multiple-output (MIMO) orthogonal frequeny division multiplexing (OFDM) systems with limited feedba. In order to utilize highly flutuating MIMO hannel effiiently and reliably, spatial mode swithing between beamforming, diversity and multiplexing modes is needed. We propose mutual information effetive SINR mapping (MIESM) method to spatial mode and modulation sheme seletion problem in frequeny seletive MIMO-OFDM hannels. The use of the effetive reeived SINR allows aurate lin error predition leading to the enhaned lin performane. Simulation results indiated that a low omplexity adaptive system whih applies diversity mode and multiplexing mode attain good performane in MIMO. Slight performane improvement an be provided by using simple beamforming with two fixed beams in addition to diversity and multiplexing modes. 1. INTRODUCTION Inreased demand of data and multimedia servies evoes the need for wireless appliations that provide higher data rates and enhaned reliability. To meet these demands the most promising wireless tehnologies inlude multiple-input multiple-output (MIMO), orthogonal frequeny division multiplexing (OFDM) and their ombinations. In order to further improve the system performane, radio lin adaptation is a promising expedient. Adaptation an be provided in time, frequeny and/or spae. Optimal transeiver design for MIMO has been studied in several papers, see [1] and referenes therein. However, in frequeny division duplexing (FDD) tehnique based system optimal MIMO an never be ahieved in pratie due to the limited feedba hannel apaity. MIMO optimization with imperfet hannel state information (CSI) at the transmitter has been reeived a lot of attention in existing literature []-[6]. Usually mean ovariane [5] or mean hannel [6] is assumed to be nown, but atual required number of feedba bits is not presented. Diversity and spatial multiplexing (SMUX) tehniques using quantized feedba information have been investigated in [7] and [8], respetively. However, user hannel may vary from line-of-sight (LOS) situation to non line-of-sight (NLOS) hannel leading diffiult odeboo optimization problem. Spatial mode swithing between beamforming, diversity and spatial multiplexing modes provides pratial method to follow real MIMO hannel [9]. Problem of the adaptation method presented in [9] is the fat that determination of the threshold values for spatial seletivity fator and SNR values requires lot of lin level simulations. Moreover, two dimensional loo up table is required to find the most suitable transmission mode in the method proposed in [9]. In addition, the feedba amount inreases notably in OFDM systems due to the frequeny seletive fading. It has been showed in [10] that using the average SNR over the subarriers of the previous transmission time as a lin adaptation method in OFDM systems may lead to the poor lin performane. It is ruial to selet the proper type of the average quality indiator, sine inaurate spatial and modulation mode seletion leads to the redued spetral effiieny or inreased error rate performane. In this paper we onsider MIMO-OFDM systems with stritly limited feedba. Exat frequeny domain adaptation an not be employed due to the limited feedba onstraint. Thus, a single spatial and modulation sheme is seleted for the entire transmission bandwidth and the adaptation is based on the average hannel quality indiator. We propose Mutual information effetive SINR mapping (MIESM) [11] for lin adaptation. MIESM has traditionally been used to approximate lin level performane in system level simulations [11]. Due to its exellent apability to predit error rate performane of large variety of MIMO-OFDM hannels, we show that MIESM is appropriate method to selet the most suitable spatial

2 mode and modulation sheme. Spatial mode seletion between diversity and multiplexing modes ombined with adaptive modulation are used to ahieve high spetral effiieny for LOS and NLOS hannels with aeptable reliability. Performane of adaptive MIMO- OFDM systems suitable for pratial environment with limited feedba is studied via simulations This paper is organized as follows. In Setion, the adaptive MIMO-OFDM system model is introdued. In setion 3, the basi priniples of the MIESM based lin adaptation method is desribed. Setion 4 illustrates some simulation results and setion 5 represents the onlusions.. SYSTEM MODEL A simplified blo diagram of the adaptive MIMO- OFDM system is shown in Fig. 1. First, the randomly generated bit sequene is turbo enoded. The oded bits are then mapped to the orresponding symbols by modulator. Four different modulation orders, whih are presented in Table 1, are used. The output symbols of the modulator are then onveyed to the transmitter spatial proessing blo, where the signal follows the operations of the seleted spatial mode. After that the OFDM transmitter proessing is performed similarly for all the parallel symbol streams orresponding to a ertain transmit antenna. The OFDM transmit proessing inludes inverse fast Fourier transformation (IFFT) and the yli prefix (CP) addition. It is assumed that CP is larger than the delay spread of the hannel, thus inter symbol interferene (ISI) is prevented. Finally, the OFDM modulated signals are transmitted to the hannel. In the reeiver side, the OFDM demodulation is exeuted before onveying the parallel signals to the spatial proessing blo. The OFDM reeiver proessing onsist of the CP removal and fast Fourier transformation (FFT). It is assumed that the subarriers are orthogonal with eah other and the OFDM symbols are perfetly synhronized, thus no inter arrier interferene (ICI) is present. In the reeiver spatial proessing blo the signal is spatially operated aording to the employed spatial mode, and then onveyed to the demodulator. After the demodulator the soft bit sequene is turbo deoded and the resulting bits are onveyed to the data sin. Spatial mode seletor blo alulates effetive SINRs for eah spatial and modulation shemes. The best transmission sheme with aeptable frame error rate (FER) is seleted. Information of the seleted transmission mode is onveyed ba to the transmitter via fast feedba lin. The studied adaptive system an swith between three various spatial modes inluding fixed-beam spatial Turbo enoder Turbo deoder Modulator Demodulator Transmitter spatial proessing Reeiver spatial proessing Transmission mode seletor OFDM transmitter proessing OFDM transmitter proessing Fast feedba lin OFDM reeiver proessing OFDM reeiver proessing Fast feedba lin Fig. 1. A generi blo diagram of the MIESM based adaptive MIMO-OFDM system. multiplexing (FB SMUX) [1], fixed-beam spae-time blo oding (FB STBC) [13] and fixed-beam beamforming (FB BF) [14]. The basi idea behind the fixed beam based multiplexing and diversity tehniques is that the information is transmitted aross the generated fixed beams. Fixed beams are generated by the Butler matrix given by W = [ w , w ] = j j, (1) where a olumn represents beamforming vetor of a stream. The benefit of the fixed beam tehniques is that the transmit energy is foused on the narrow area produing an additional array gain. The obtained array gain inreases as the number of beams inreases. The best beamforming vetor for beamforming mode an be seleted aording to the uplin power measurements [15]. FB SMUX and FB STBC do not require any beam seletion measurements, sine they use both of the generated fixed beams for transmission. A half-wavelength antenna spaing at the both end of the ommuniation lin is assumed, sine side lopes are unwanted. The main simulation parameters of the urrent adaptive MIMO-OFDM system are presented in Table 1. For FB SMUX, we an desribe the reeived signal vetor of the th subarrier as r = H Ws + n = B s + n, () N R 1 where r C is the reeived signal vetor, N R N H T C N T is the hannel matrix, T N W C is N 1 the Butler matrix, s C T is the transmitted signal N R 1 vetor, n C is the noise omponent vetor and N R N B T denotes the virtual beamformer hannel C

3 Table 1: Main parameters of the system. Antenna set-up -by- Number of users 1 Number of subarriers 51 Channel model SCME Channel bandwidth 0 MHz Senario Urban miro Channel options NLOS, LOS Channel oding Turbo Code rate ½ Deoder iterations 8 Modulation methods BPSK, QPSK, 16QAM, 64QAM FER onstraint 0.1 Antenna spaing at BS ½ λ Antenna spaing at MS ½ λ Spatial modes BF, STBC, SMUX Reeiver types MRC, LMMSE Effetive SINR threshold values (db) -1.31(BPSK), 1.55(QPSK), 6.76(16QAM), 11.6(64QAM) matrix. The input-output relationship per subarrier for FB STBC is given by s s,1, r = B + n. (3) s, s,1 For FB BF, the reeived signal of the th subarrier an be expressed as N R 1 r = +, (4) H w d s n where w d C denotes the beamforming vetor being the d th olumn of the W and d = {1, }. 3. MODULATION AND SPATIAL MODE ADAPTATION This setion desribes the basi priniples and the mathematial baground of the MIESM based lin adaptation tehnique. Effetive SINR values are alulated for eah spatial and modulation shemes for every instantaneous hannel onditions at the reeiver. Effetive SINR results are ompared with the thresholds. The transmission sheme that oversteps its threshold while owning the greatest spetral effiieny is seleted for the next transmission. Thresholds are predefined so that a desired FER onstraint is met. The most appropriate transmission mode is onveyed ba to the transmitter via fast feedba lin. The aim of the effetive SINR metri is to ompress a set of post-detetion SINR values γ into a single effetive SINR value γ via loo-up table. Then this eff effetive SINR value is mapped to the orresponding error probability, suh as FER. The MIESM ompression method an be expressed as [11] γ eff N 1 1 = F F ( γ ), (5) N = 1 where F denotes the funtion of the modulation onstrained mutual information in additive white Gaussian noise (AWGN) hannel. Mutual information versus SNR results are alulated beforehand and employed as a loo-up table during the simulations. The funtion F an be presented as [16] F ( γ ) = log ( M ) M M 1 1 p q s + n s n E log exp, (6) p= 0 q= 0 σ 1 M where M is the size of the modulation alphabet, represents the p th possible onstellation point and n is the noise omponent with variane σ. The harater of the mutual information loo-up table depends on the used modulation method. Hene, loo-up tables should be alulated separately for eah modulation methods. The mapping of a set of post-detetion SINR values into a single effetive SINR value using F as a loo-up table is illustrated in [16]. First, eah γ is mapped into the orresponding mutual information (MI) domain values via loo-up table. Then the average is alulated over the orresponding MI domain values. This average value is then mapped ba to the SINR domain through the inverse of the mutual information loo-up table (dash line). The ahieved SINR domain value is the effetive SINR value γ eff. Next the alulation of γ is presented for different transmission tehniques. For FB SMUX with the linear MMSE reeiver, γ of the th streams is, H, H N 0 b, + P T g, b, p s PT g b, γ =, (7) where denotes the L -norm of a vetor spae, denotes the absolute value, P T is the transmit power, N R 1 N 0 is the noise power spetral density, g, C is the th olumn of the matrix MMSE filter and N R 1 b, C is the th olumn of the virtual beamformer hannel matrix B. For STBC and beamforming, the post-detetion SINR is redued to post-detetion SNR, sine there is no

4 Average MIdomain value Solid lines: Post-detetion SINR values per subarrier =1,..8 and the orrersponding MI-domain values 10 0 BF, SCME NLOS BF, SCME LOS BF, UC hannel STBC, SCME NLOS STBC, SCME LOS STBC, UC hannel SMUX, SCME NLOS SMUX, SCME LOS SMUX, UC hannel SISO AWGN Mutual information(mi) domain FER 10 1 Dash line: Effetive SINR value and the orresponding MI-domain value Effetive SINR value Post-detetion SINR domain Fig.. The method for ompressing the set of the orresponding γ. eff γ into multistream interferene (MSI) between transmitted symbols. Hene, γ for FB STBC after maximum ratio ombining (MRC) is defined by PT = b, ij N 0 j= 1 i= 1 γ, (8) where b, ij is the oeffiient of the virtual beamformer hannel matrix B. Post-detetion SNR for FB BF after MRC is expressed as PT γ = H w d. (9) N 0 4. SIMULATION RESULTS In this setion, FER versus effetive SINR results, FER performane and spetral effiieny results of the MIESM based adaptive MIMO-OFDM system are presented. Spetral effiieny is measured in bits per seond per hertz (bits/s/hz) and it is alulated by taing the FER into aount and observing the number of suessfully reeived frames in a given time interval and bandwidth. Simulations were performed aording to the parameters in Table 1. Non-line-of-sight (NLOS) and line-of-sight (LOS) hannels in urban miro environment of the geometri spatial hannel model extended (SCME), developed by the 3 rd generation partnership projet (3GPP), were employed to generate the hannel oeffiients [17]. Fig. 3 represents the FER versus effetive SINR urves, from whih the MIESM thresholds are determined for a predefined FER onstraint of 0.1. A single simulation point represents a single MIMO- OFDM hannel realization and FER is averaged over Effetive SINR [db] Fig. 3. FER vs. effetive SINR values for variety of different hannel models and transmission tehniques. 000 AWGN realizations. The results are simulated for several different transmission tehniques and MIMO hannels in order to prove the fat that the MIESM method provides a reliable approximation of the FER performane for a variety of transmission shemes and instantaneous hannel realizations. The applied transmission tehniques inlude FB BF, FB STBC and FB SMUX. The used hannel models are SISO AWGN, NLOS (SCME), LOS (SCME) and exponentially deaying unorrelated hannel with 65 ns root mean square delay spread. As was expeted, the simulation results indiate that the MIESM method provides reliable lin error predition for a variety of transmission tehniques and MIMO hannel realizations. In Table 1, the MIESM threshold values are represented for BPSK, QPSK, 16-QAM and 64-QAM. The FER performane of the adaptive beamforming diversity and multiplexing (ABDM) system and the nonadaptive MIMO-OFDM systems in the NLOS (SCME) hannel is presented in Fig. 4. Results show evidently that after going below the FER onstraint of 0.1 for the first time, the FER urve of the ABDM remains there. Atually, the urve goes frequently up and down, but remains well below the FER level of This is due to the fat that there is only finite number of spatial, modulation and oding shemes. Fig. 5 presents the spetral effiieny of the ABDM and non-adaptive MIMO-OFDM systems in the NLOS (SCME) hannel. It an be seen that the ABDM has always the greatest spetral effiieny ompared to the non-adaptive tehniques. Hene, ABDM produes a gain in spetral effiieny over non-adaptive systems, as was expeted. It is worthwhile to notie that for ABDM the FER always remains below the required level while spetral effiieny is inreasing. Adaptive system analyzed previously employed three different transmission shemes. However, we an

5 ABDM, NLOS ABM, NLOS ADM, NLOS ABDM, LOS ABM, LOS ADM, LOS FER 10 FER FB SMUX BPSK FB SMUX QPSK FB SMUX 16QAM FB SMUX 64QAM FB STBC BPSK FB STBC QPSK FB STBC 16QAM FB STBC 64QAM FB BF BPSK FB BF QPSK FB BF 16QAM FB BF 64QAM ABDM Fig. 4. FER performane of adaptive and non-adaptive MIMO-OFDM systems in the NLOS hannel. Fig. 6. FER performane of various adaptive MIMO- OFDM systems in the NLOS and LOS hannels. Spetral effiieny [bits/s/hz] FB BF BPSK FB BF QPSK FB BF 16QAM FB BF 64QAM FB STBC BPSK FB STBC QPSK FB STBC 16QAM FB STBC 64QM FB SMUX BPSK FB SMUX QBPSK FB SMUX 16QAM FB SMUX 64QAM ABDM Spetral effiieny [bits/s/hz] ABDM, NLOS ABM, NLOS ADM, NLOS ABDM, LOS ABM, LOS ADM, LOS Fig. 5. Spetral effiienies of adaptive and nonadaptive MIMO-OFDM systems in the NLOS hannel. assume that similar performane results may be ahieved with the system using only two different transmission tehniques inluding FB BF and FB SMUX. Adaptive system using BF and SMUX is denoted as ABM. This will derease the system omplexity by reduing the amount of required feedba bits from four to three as well as reduing the effetive SINR alulations in the reeiver side. To further derease the system omplexity, so that the beam seletion is not needed, an adaptive diversity multiplexing (ADM) system using FB STBC and FB SMUX an be employed. Fig. 6 shows the FER performane results of ABDM, ABM and ADM systems for NLOS and LOS hannels. ABDM and ABM have similar performane ompared to eah other and they outperform ADM for approximately 1 db at low SNR region. It an be notied that in SCME the NLOS and LOS hannels are surprisingly similar. The spetral effiienies of the adaptive systems in NLOS and LOS hannels are showed in Fig. 7. It an be Fig. 7. Spetral effiienies of various adaptive MIMO- OFDM systems in the NLOS and LOS hannels. seen that ABDM and ABM have equal spetral effiienies. ADM has worse spetral effiieny than ABDM and ABM at low and medium SNRs. However, the differene is only 0.3 bits/s/hz. 5. CONCLUSIONS This paper studied the adaptive MIMO-OFDM systems whih swith between different spatial and modulation shemes aording to limited feedba information. The MIESM was introdued to be exellent method to selet the most suitable transmission modes to the urrent hannel ondition. The aim of the MIESM based adaptive tehnique was to enhane the spetral effiieny of the system with a predefined FER onstraint. Simulation results showed evidently that all adaptive systems provided signifiant gain in spetral effiieny ompared to non-adaptive systems. The results showed that the system using STBC and SMUX modes is attrative for -by- MIMO due to its low

6 omplexity and good performane. Results indiated that beamforming with two possible fixed beams option does not help signifiantly adaptive system using STBC and SMUX. Further study is needed to determine most suitable spatial modes for MIMO-OFDM systems with limited feedba and more than two antennas. Also performane study in measured MIMO hannels are left for future wor. 6. ACKNOWLEDGEMENT This wor has been arried out in the framewor of the projet Future Radio Aess Conept: Tehnology and performane Assessment (FRACTA), whih is funded by the Finnish Funding Ageny for Tehnology and Innovation (Tees). 7. REFERENCES [1] A. J. Paulraj, D. A. Gore, R. U. Nabar and H. Bölsei, An Overview of MIMO Communiations A Key to Gigabit Wireless, Proeedings of the IEEE, vol. 9, no., pp , Feb 004. [] S. A. Jafar, A. Goldsmith, "Transmitter Optimization and Optimality of Beamforming for Multiple Antenna Systems," IEEE Trans. Wireless Commun., vol. 3, no. 4, pp , Jul 001. [3] H. Sampath, A. Paulraj, "Linear Preoding for Spae-Time Coded Systems with Known Fading Correlations," IEEE Commun. Letters., vol. 6, no. 6, pp , Jun 00. [4] G. Barria, U. Madhow, "Spae-Time Preoding for Mean and Covarine Feedba: Appliation to Wideband OFDM," IEEE Trans. Commun.., vol. 54, no. 1, pp , Jan 006. [5] S. Zhou, G. Giannais, "Optimal Transmitter Eigen- Beamforming and Spae-Time Blo Coding Based on Channel Correlations," IEEE Trans. Inform. Theory, vol. 49, no. 7, pp , Jul 003. [6] S. Zhou, G. Giannais, "Optimal Transmitter Eigen- Beamforming and Spae-Time Blo Coding Based on Channel Mean Feedba," IEEE Trans. Signal Proessing, vol. 50, no. 10, pp , Ot 00. [7] D. J. Love, R. W. Heath, Jr, "Limited Feedba Diversity Tehniques for Correlated Channels," IEEE Trans. Veh. Tehnol., vol. 55, no., pp , Mar 006. [8] D. J. Love, R. W. Heath, Jr, "Limited Feedba Unitary Preoding for Spatial Multiplexing Systems," IEEE Trans. Inform. Theory, vol. 51, no. 8, pp , Aug 005. [9] A. Forenza, A. Pandharipande, H. Kim and R. W. Heath Jr, Adaptive MIMO transmission sheme: exploiting the spatial seletivity of wireless hannels, in Pro. of IEEE Vehiular Tehnology Conferene, Stoholm, Sweden, May 005, pp [10] M. Lampe and H. Rohling, "PER-predition for PHY mode seletion in OFDM ommuniation systems", in Pro. IEEE GLOBECOM 003, San Franiso, USA, De 003, pp [11] IST WINNER, D.7 assessment of advaned beamforming and MIMO tehnologies, Teh. Rep., 005. [1] J. Ylitalo, Fixed-Beam MIMO Sheme, in Pro. of 006 International Symposium on Personal, Indoor and Mobile Communiations (PIMRC 06), Helsini, Finland, Sep. 006, pp [13] W. Meng, L. Gu, and C. Li, "The Combined Beamforming and Spae-Time Blo Coding Tehnique for Downlin Transmission", in Pro. of IEEE International Conferene on Wireless Networs, Communiations, and Mobile Computing, Hawaii, USA, June 005, pp [14] E. Tiirola and J. Ylitalo, Performane evaluation of fixed-beam beamforming in WCDMA downlin, in Pro. of IEEE Vehiular Tehnology Conferene, Toyo, Japan, May 000, pp [15] J. Ramiro-Moreno, K. I. Pedersen and P. E. Mogensen, Capaity Gain of Beamforming Tehniques in a WCDMA System Under Channelization Code Constraints, IEEE Trans. Wireless Commun., vol. 3, no. 4, pp , Jul 004. [16] G. Ungerboe, "Channel Coding with Multilevel/Phase Signals," IEEE Trans. Inform. Theory, vol. 8, no. 1, pp , Jan 198. [17] 3rd Generation Partnership Projet (3GPP); Tehnial Speifiation Group Radio Aess Networ, Spatial hannel model for multiple input multiple output (MIMO) simulations (3G TS version (release 6)), Teh. Rep., 3rd Generation Partnership Projet (3GPP), 003. (3G TS version (release 6)), Teh. Rep., 3rd Generation Partnership Projet (3GPP), 003.