FLOWS 1ST-2001-32125 Deliverable No:D14 The LMS technigu can be implemented without division (though a normalised version of the tec which is most suitable for practical applications,does require divi ,an is most popula applications cause a bas plementing the LMS tech the LMS techni of the MS algorithm make its hardware implementation(for example,on FPGA platform or as an ASIC) difficult [Rag99,GFF00,CZB00]. A hardware implem ntation on FPGA or ASIC allows reduction of pov wer con umption of mobile fast convergence and low com still of for CDMA in th he inverse were prop ed:a eralised Levinson algorithm taki ing advantage of the block Toeplit hich and an equaliser xp the iterative ance e of th multiplications were presented.However.these techniques are sti complicated to implement. In [ZT02]it has been shown how systems of linear equations(normal the linear Least Square (LS)problem,in particular. it do opt applied to multiuse dete optim demor nlink w has beer requires s a re y sm omputational ed that ad reove the on of the equali rector of nor ns)can be done with no xplicit multiplication The performance of a fractional chip-equalised downlink receiver was investigated for UMTS sc f thend fast ading c channels by using simulation 01 number of bits representing the equaliser taps,channel power del rofile and fading rate (Dopple sprea for single- o-antenna Ihe propose the he ad ptation tge。e3gbe d that of the tion techn ique allo igh fully 2.6 MIMO in HIPERLAN/2 and the SATURN project 2.6.1 MIMO channel characterisation In another e METRA H=(R)PG(R)/2 (2.26) 17 December 2003 Page 37
FLOWS IST-2001-32125 Deliverable No: D14 The LMS technique can be implemented without division (though a normalised version of the technique, NLMS, which is most suitable for practical applications, does require divisions [HNK99, KNK99]), and is most popular in real-time applications. Because a basic DSP operation is MAC (multiplication and accumulation), there are many DSP programs implementing the LMS technique. However, the LMS technique possesses a low convergence speed, which significantly limits its applications and, in particular, prevents the use of the LMS adaptation in chip-equalised downlink receivers. In addition, multiplications used in the LMS algorithm make its hardware implementation (for example, on FPGA platform or as an ASIC) difficult [Rag99, GFF00, CZB00]. A hardware implementation on FPGA or ASIC allows reduction of power consumption of mobile terminals which is very important in practice. Therefore, equalisation techniques possessing a fast convergence and low complexity are still of great interest. Several reduced complexity implementations of equalisers for CDMA downlink were investigated in [Mai01]. Channel estimation was based on a continuously available pilot, and three algorithms for the matrix inverse were proposed: a generalised Levinson algorithm taking advantage of the block Toeplitz structure; an equaliser which is approximated by parallel equalisers with a polyphase structure; and an equaliser exploiting the iterative Gauss-Seidel technique. The performance of these approaches was compared, and estimates of complexity in terms of the number of complex multiplications were presented. However, these techniques are still complicated to implement. In [ZT02] it has been shown how systems of linear equations (normal equations), appearing in the linear Least Square (LS) problem, in particular, describing the MMSE equalisation, can be solved by means of coordinate descent optimisation. Such optimisation is particularly attractive because it does not require any knowledge of the gradient of the function to be optimised. When applied to multiuser detection the coordinate descent optimisation demonstrates both high performance and low complexity [LLPW00, ZT01a]. This approach is applied to the linear equalisation problem in downlink WCDMA transmission; it has been demonstrated that it requires a relatively small computational load. Moreover, the calculation of the equaliser taps after calculation of auto- and cross-correlations (providing the coefficient matrix and rightside vector of normal equations) can be done with no explicit multiplication. The detection performance of a fractional chip-equalised downlink receiver was investigated for UMTS scenarios in multipath slow and fast fading channels by using simulation [ZTT01]. Specifically, the dependence of the bit-error rate on the number of active users, length of the equaliser filter, number of bits representing the equaliser taps, channel power delay profile, and fading rate (Doppler spread) for single- and two-antenna receivers were considered. The proposed equalisation technique provides results close to that based on the direct matrix inversion and the computational load of the adaptation stage is negligible compared to that of the matched filtering and calculation of the autocorrelation of the received signal. The proposed equalisation technique allows high detection performance even in fully loaded multiuser scenarios and a simple implementation of a UMTS mobile receiver. These techniques can be also used in downlink MIMO equalisation schemes. 2.6 MIMO in HIPERLAN/2 and the SATURN project 2.6.1 MIMO channel characterisation In another IST research and development project, SATURN (similarly to the METRA project [Sea01]), the following model of the MIMO channel matrix is used [MGI02] H = (Rr) 1/2 G (Rt) 1/2 (2.26) 17 December 2003 Page 37
FLOWS 1ST-2001-32125 Deliverable No:D14 where G is a stochastic x n matrix with i.i.d.complex Gaussian zero-mean unit variance then extended to channels with multipath propagation as H(T)=>ak(R-(k))2G&(R:(k))26(). (2.27) 0 Real experiments performed in the project have shown that the model(2.27)is in a good H(T)=>axGk6(T-T). (2.28) k=0 2.6.2 Exploiting SMART antenna technology The docun ion of C/N and link throughput when operating in con nart Antenna at the AP.The use of sectorised antennas has bee n proposed as a means (a)achieving su verage,(b)sav ng transm pow and (c)red dire em has beeni ed chane data for both sectorised no dominant multipathco up tent. 2.6.3 Performance analysis and validation of MIMO MGSSO-k mun n dicates rge IR when passing f n an actua to chann system smaller inc MIMO systems and configurat tions e ing an antenna array only at the Access Point (AP) that for a small r number of ar un en in a synthet w the cha e assumpt Further of this idea ented in [MGIO2]Her ral sIso-like channels with independent symbol streams een sed to use in order to provide data rates The effect of cor rrelation can be decrease cted significantly by the correlation:for example.the first 17 December 2003 Page 38
FLOWS IST-2001-32125 Deliverable No: D14 where G is a stochastic nT × nR matrix with i.i.d. complex Gaussian zero-mean unit variance elements. The transmit Rt and receive Rr side correlation matrices define the correlation between transmit and receive array elements, respectively. The narrowband model (2.26) is then extended to channels with multipath propagation as H(τ ) = L k=0 αk (Rr(k))1/2 Gk (Rt(k))1/2 δ(τ − τk). (2.27) Real experiments performed in the project have shown that the model (2.27) is in a good agreement with results of the measurements. The spatially uncorrelated (“ideal”) channel model is obtained from (2.27) by assuming Rt(k) = Rr(k) = I: H(τ ) = L k=0 αkGkδ(τ − τk). (2.28) 2.6.2 Exploiting SMART antenna technology The document [DAK+01] describes the performance assessment of HIPERLAN/2 in terms of packet error rate as a function of C/N and link throughput when operating in conjunction with a Smart Antenna at the AP. The use of sectorised antennas has been proposed as a means of (a) achieving superior coverage, (b) saving transmit power, and (c) reducing interference relative to an omni-directional antenna. The physical layer performance of a HIPERLAN/2 system has been investigated by simulation using measured channel data for both sectorised and omni-directional antenna structures. A gain of 3.5-7.25 dB is achieved in all scenarios via the antenna gain. The performance is also improved by spatial filtering of the radio channel. In channels with a dominant multipath component, the correct choice of sectors improves the Rician statistics; gains of up to 6.7 dB occur. In channels with a high angular spread and no dominant multipath component, a small improvement will occur from six-sector antennas. Benefits are expected to be higher for mobile terminals. 2.6.3 Performance analysis and validation of MIMO In [MGI01] SISO-like communication employing only the spatial channel associated to the highest singular value of the channel matrix for each OFDM bin has been analysed. The performance evaluation indicates a large gain in SNR and SIR when passing from an actual SISO to the SISO-like channel derived from a (2,2) MIMO system. A smaller increment gain is achieved when passing from the (2,2) to (4,4) MIMO system. A comparison between MIMO systems and configurations employing an antenna array only at the Access Point (AP) indicates that for a small number of array elements the latter can be more cost effective. Note that the investigation undertaken in [MGI01] is concerning a synthetic channel model which is time-invariant and uses the assumption that the transmitter and receiver know the channel. Further development of this idea was presented in [MGI02]. Here, several SISO-like channels with independent symbol streams has been proposed to use in order to provide data rates beyond the maximum data rate of 54 Mbps. It is indicated that spatial correlation degrades the subchannels having smaller gains. The effect of correlation can be decreased by using only those spatial channels not affected significantly by the correlation; for example, the first 17 December 2003 Page 38
FLOWS IST-2001-32125 Deliverable No:D14 spatial channels in a (4.4)correlated configuration have about the same gains as the first two atial channels of th nheorelate orrelatio ng the aP antenna e ments can be reduced b diffe ration.ol the tained by increasing the of elem erforms a nmetric one.The nh spatial subchannel actically ineffective in dec the transmitter pow for throughputs below 54(n- 1)Mbps;for example,up to 108 M only the ached two use thre Ihe cos to 化gTea transmission mode on all spatial subchannels is also considered in [MG102]. 2.6.4 MIMO synchronisation scribes a nov el t of null (virtual)subcarriers and does not require the knowle virtual subca rs due to inte quen estin he nstrates a high p rormance in different 2.7 Conclusions In this chapter. e have considered MIMO communicatio technig s known from the litera- ne-varving channels We can conclude the following. For MISC e add 10 cha th gai systems, Space. cost e antennas.while the channel capacity approaches The lavered spac e-time architecture allo omplexity to w linearly. gationy. tennas are f the cha spa n o the systems more suited for real-time implementation 17 December 2003 Page39
FLOWS IST-2001-32125 Deliverable No: D14 spatial channels in a (4,4) correlated configuration have about the same gains as the first two spatial channels of the uncorrelated configuration. Note that the gain of the first channel is quite the same for both the correlated and uncorrelated case; this is also true for asymmetric configurations. The effect of correlation among the AP antenna elements can be reduced by increasing the number of mobile terminal (MT) antenna elements beyond the number of used subchannels. In general, when using different transmission modes in distinct spatial subchannels, an asymmetric configuration, obtained by increasing the number of elements only at the AP and taking only the minimum number of MT elements to achieve the maximum desired throughput, outperforms a symmetric one. The nth spatial subchannel is practically ineffective in decreasing the transmitter power for throughputs below 54(n − 1) Mbps; for example, up to 108 Mbps only the first two subchannels are useful. The cost effectiveness of using an nR-element MT is approached only when the required throughput gets close to 54nR Mbps. Using the same transmission mode on all spatial subchannels is also considered in [MGI02]. 2.6.4 MIMO synchronisation The document [MG02] describes a novel time and carrier frequency synchronisation algorithm in OFDM MIMO systems in application to HIPERLAN/2. The algorithm requires the presence of null (virtual) subcarriers and does not require the knowledge of channel impulse responses or a preamble sequence. It measures the excess of energy on the virtual subcarriers due to inter carrier interference and intersymbol interference; a minimum of the energy indicates the time and carrier frequency estimates. The algorithm demonstrates a high performance in different multipath environments, while being simple for real-time implementation. 2.7 Conclusions In this chapter, we have considered MIMO communication techniques known from the literature by dividing them in accordance with the channel model used for their derivation. We have presented MIMO techniques for flat fading, frequency-selective, and time-varying channels. We can conclude the following. For MISO systems, there is no additional channel capacity to be gained. For SIMO systems, the channel capacity scales only logarithmically with the number of receive antennas. For MIMO systems, the asymptotic capacity grows linearly with the number of transmit antennas. Space-time block codes provide full diversity at low computational costs. When the number of transmit antennas is increased, their capacity approaches the capacity of one AWGN channel with SNR scaled by the number nR of receive antennas, while the channel capacity approaches the capacity of nR parallel AWGN channels. The layered space-time architecture allows signal processing complexity to grow linearly, rather than exponentially, with the promised capacity increase. However, such systems suffer from error propagation, especially, when equal number of antennas are employed at both ends. Advanced layered space-time systems can be developed by employing the QR-decomposition of the channel matrix; in particular, this allows overcoming the problem of error propagation. The computational load of layered space-time systems is still high and application of such matrix techniques as the QR-decomposition and the Cholesky decomposition can result in systems more suited for real-time implementation. 17 December 2003 Page 39
FLOWS 1ST-2001-32125 Deliverable No:D14 5or5cenam where the perf equivale nt versty nT The combination of ive fre se and multiple antenna nber of nd the of freedo m prove e of MIMC dulation with igher througput thanINR-basedper pow of measurement of MIMO Re channel for 5-GHband (ERLAN)inmicro (b)the of the char nd (e)t measured apacity exhibits no significant increase if the inter-elements spacing is increased above half of the wavelength(2.9 cm). n MIMO systems, preeabeoractcfmp2henan8E mple,for HIPER est singular alue of the channel matrix The shows that si mncbe obtained n the infomatio oeeming the prormanc of MM channele based mainly on the tion channels is not we presented in the literatur number of antenna too many e rate estimation topic requires significant efforts in future investigations channels.MIMO OFDM syste Doppler scattering due to the mob ility sh t MIMO b significant.Ho ever frequency ne oput. E ssible to track the purpose,poly tehe IST research and development project METRinestigated applications of MIMO d ewere deve and tesdiv ansmit a 17 December 2003 Page 40
FLOWS IST-2001-32125 Deliverable No: D14 Antenna selection is an inexpensive way to obtain diversity over a multiple antenna fading channel. With as little as one extra antenna it can dramatically improve the performance of linear receivers. For scenarios, where the performance can be improved by optional receive diversity, transmit diversity may be used, when the same information is transmitted from all transmit antennas. This is equivalent to transmission over a SISO channel with no diversity gain. To obtain a diversity gain, modulation diversity or delay diversity can be additionally used. The combination of adaptive modulation, aggressive frequency reuse and multiple antenna transmission can significantly increase the data throughput in multiuser systems. Increasing the number of receivers beyond the number of transmitters will add more degrees of freedom for interference cancellation and thus will improve the performance of MIMO systems by providing diversity against fading and cochannel interference. Using adaptive modulation without power control provides a significantly higher throughput than using SINR-based power control. Results of measurement of MIMO channel capacity for 5.2-GHz band (HIPERLAN) in microcellular environments have shown that: (a) the capacities are up to 30% lower than would be expected; (b) the frequency-selectivity of the channel adds additional diversity; and (c) the measured capacity exhibits no significant increase if the inter-elements spacing is increased above half of the wavelength (2.9 cm). Similarly to receivers in flat-fading channels, layered space-time receivers are of interest for frequency-selective channels. MIMO systems based on channel equalisations are attractive for implementation. In MIMO systems, ordered successive interference cancellers (OSIC) outperform decision-feedback equalisers in BER performance. Partially connected OSIC receivers are preferable for practical implementation. Schemes with a feedback channel can offer a better performance. For example, for HIPERLAN/2 it was proposed to form for every OFDM subcarrier a SISO-like link, based on the highest singular value of the channel matrix. The ”water-pouring” approach shows that significant performance improvement can be obtained by adaptively allocating the information bits and power over the sub-channels. Results concerning the performance of MIMO channels are based mainly on the assumption of perfect knowledge of the channel. However, the channel estimation problem in application to MIMO channels is not well presented in the literature. For a large number of antennas and a rapidly varying channel, too many pilots are needed in order to adequately estimate the channel, and significant degradation in terms of achievable rate are inevitable. The channel estimation topic requires significant efforts in future investigations. In time-varying channels, MIMO OFDM systems can operate properly until the Doppler spread is less 0.01 relative to the symbol rate (carrier spacing). For HIPERLAN/2, the influence of Doppler scattering due to the user mobility should not be significant. However, frequency synchronisation errors may make the equivalent MIMO channel time-variant. The channel tracking improves the throughput. It is possible to track the channel variations by interpolating between consecutive channel estimates. For this purpose, polynomial interpolation can be used. The IST research and development project METRA has investigated applications of MIMO techniques in UMTS. In this project, a stochastic MIMO channel model and different MIMO schemes were developed. A layered MIMO scheme and a punctured scheme were developed and tested for the FDD mode; in lower-order MIMO channels, puncturing and transmit and receive diversity is superior to layered techniques in terms of complexity and performance. For the TDD mode, such schemes as beamforming, Alamouti space-time codes, BLAST, and STBC were investigated. In another IST research and development project, SATURN, MIMO techniques in HIPERLAN/2 systems are considered. SISO-like channels associated with highest singular values of 17 December 2003 Page 40
FLOWS IST-2001-32125 Deliverable No:D14 the channel matrix are investigated.It is determined that n subchannel is ineffective in de 17 December 2003 Page 41
FLOWS IST-2001-32125 Deliverable No: D14 the channel matrix are investigated. It is determined that nth subchannel is ineffective in decreasing the transmitter power for throughput below 54(n − 1) Mbps; for example, up to 108 Mbps only the first two subchannels are useful. This project also deals with the MIMO synchronisation, in particular, time and frequency synchronisation techniques suited for real-time implementation. 17 December 2003 Page 41