《通信原理》课程教学资源（辅导资料）Selected MIMO Techniques and their Performance.pdf_P31-P35

FLOWS IST-2001-32125 Deliverable No: D14 2.4.2 Capacity of time-varying MIMO channels Most wireless communication systems form channel estimates and use them as if they were true known channels. The achievable rate of a typical digital communication system following the principle of synchronised detection was derived in [BFM01] under the assumption that the system operates on a flat fading time-varying MIMO channel and uses an interleaver to combat bursty error events. It has been shown that the LMMSE channel estimation directly follows from the derivation. The achievable rate is optimised with respect to the amount of training information needed. Some results from the paper [BFM01] are as follows. For the (1,1) system all achievable rates exhibit a distinct maximum for channel sampling periods strictly smaller than Nyquist sampling (ν < 0.5/F, where ν is the channel sampling period and F is the normalised Doppler frequency). There are two opposite effects here. For small ν, the channel estimates have a small MSE, however, a large portion of the achievable rate is sacrificed due to the insertion of pilot symbols. For large ν approaching Nyquist sampling the achievable rate is reduced not primarily due to inserted pilot symbols, but the increased MSE of the channel estimates. For example, in downlink UMTS, the symbol rate is 15000 Hz, the slot rate is 1500 Hz, and the maximum Doppler frequency is 220 Hz (V = 120 km/h), so that the channel sampling is ν = 10 and the normalised Doppler frequency F = 0.015. For 16-QAM, such parameters lead to a loss in capacity of approximately 22% at SNR=6 dB; the value ν is approximately optimal for this case. If SNR=12 dB, the loss in capacity is about 14%, while at SNR=0 dB the loss is about 40%. For the (4,4) system the achievable rate for SNR=6 dB is about 4.2 bits/s/Hz instead of about 7 bits/s/Hz, i.e. the loss is 47%. This could be improved if the parameter ν was ν ≈ 20; then the loss would be only 40%. Optimal rates are achieved for the (4,4) system at significantly higher channel sampling period ν than for the (1,1) system. It becomes clear that for a large amount of antennas and a rapidly varying channel, too many pilots are needed in order to adequately estimate the channel, and in such cases significant degradation in terms of achievable rate are inevitable. This shows that synchronisation for MIMO channels is much more difficult than for SISO channels. 2.4.3 MIMO techniques for time-varying channels Single-carrier communication systems in MIMO channels face a very complex equalisation problem. Multicarrier modulation schemes (like MIMO OFDM systems) greatly lessen the equalisation complexity problem. MIMO OFDM systems employing space-time coding approach the outage channel capacity, while not requiring high complexity signal processing in the receiver. In order to additionally reduce the complexity of multi-antenna transmission, it was proposed to divide transmit antennas in groups and use individual space-time encoders for each group [LWS01]. In the receiver, an interference cancellation scheme attempts to separate the received signal due to one of the encoders from the received signals due to the other encoders. After this cancellation, MLSE decoding is employed, followed by successive interference cancellation. A (4,4) system employing two 16-state 2-antenna space time codes with channel estimation (requiring about 10% of the throughput) in the TU frequency-selective channel with Doppler scattering provides a 2-dB improvement with respect to previously known results. In [BLWY01] a 16-state 4-antenna code has been proposed that achieves an additional 2-dB improvement with lower complexity and a 256-state code performed within 3 dB of outage capacity with channel estimation (or 2 dB with known channels). The proposed 4-antenna codes provide higher performance improvement with increasing Doppler frequency. Results presented in [BLWY01] show that the MIMO OFDM system can operate properly until the Doppler spread is less than 0.01 with respect to the symbol rate. 17 December 2003 Page 32

FLOWS IST-2001-32125 Deliverable No: D14 Multicarrier (e.g. OFDM-based) modulation schemes can be efficiently combined with multiple antennas for wireless channels. This allows a relatively simple implementation in time-variant channels because for each subcarrier the channel is seen as a flat fading channel. However, the time-invariance assumption may not be valid in high-mobility applications or in applications with frequency synchronisation errors. In such applications, loss of subcarrier orthogonality at the receiver results in intercarrier interference (ICI). A time-domain filtering-based ICI mitigation technique is proposed in [DADS02]. This technique cascades a linear receive filter with the time-varying channel so that the overall channel is approximately time-invariant. This, however, requires solving a generalised eigenvalue problem, which can be complicated for real-time implementation. A space-time coded OFDM (2,1) system with Alamouti code was investigated when the channel is perfectly known to the receiver. It has been shown that for a specific scenario with about 3 kHz of the subcarrier spacing, the frequencies 2.4 GHz and 5 GHz, four-path channel, and a mobile speed of 100 km/h this technique allows improvement of the SINR by about 7 dB. Note that for HIPERLAN/2 the carrier spacing of 312.5 kHz [ETS01] is much higher and therefore 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 thus requiring accurate frequency synchronisation. Frequency estimation techniques and lower accuracy bounds for frequency-selective channels are investigated in [MM00, BZT01b, BZT02, BZT01a]. An optimal channel estimator for time-varying fading channels should account for the statistics of the channel variation using Kalman filtering. Using Kalman estimation for channel tracking and adaptive MMSE-DFE equalisation in time-varying frequency-selective MIMO channel was investigated in [KFSW00]. Since it is nontrivial to obtain the information on the channel variation, and due to the high complexity of Kalman filtering, a more simple interpolationbased estimator with orthogonal training sequences was proposed in [SC02] in application to a BLAST system operating in a flat fading channel. It is possible to track the channel variations by interpolating between consecutive channel estimates; zero-order prediction, linear, parabolic, and cubic interpolations are considered. The channel estimation error contains two terms, one is due to noise, and another is due to the temporal channel variation. There are also tradeoffs between minimising the errors and minimising the training overhead. In [SC02] the optimal training interval, training length and the maximum throughput are found by simulation as functions of the Doppler frequency and the number of antennas. Without channel tracking (zero-order prediction), when the Doppler frequency increases, the throughput decreases significantly; the channel tracking improves the throughput. The increase in the number of antennas requires the optimal training length to be also increased. In [ZT01b] a procedure for parabolic polynomial approximation of time-varying channels is proposed, and the paper [ZTA02] describes different polynomial interpolation algorithms for tracking the channel variation and estimates their accuracy. Results presented in [ZT01b, ZTA02] can be used for choosing the training period depending on the Doppler spread. Reliable communications in multiuser mobile systems operating in time-variant multipath channels requires advanced signal processing techniques to be used. It is stated in [Ver00] that the combined multipath and Doppler diversity is a power measure to overcome fading in multipath channels with Doppler scattering. This type of diversity was first proposed in application to underwater data transmission, which is known to be characterised by the rich multipath nature and fast-varying parameters; since 1987 receivers exploiting this type of diversity were investigated in many experiments [ZK95, ZK96], in particular, with OFDM signals [ZK00]. The results have shown that a time-frequency Rake receiver demonstrates significant performance improvement relative to conventional Rake receivers. The use of the principle of multipathDoppler diversity in mobile wireless radio communications was proposed in [SA97]; since this work, time-frequency Rake receivers were proposed [SA98, SA99], pilot-based channel estimation was investigated [BS02], and OFDM-like signals for doubly dispersive channels were proposed [LKS01]. Application of the multipath-Doppler diversity principle to MIMO systems is expected to significantly improve their performance in time-varying environments. 17 December 2003 Page 33

FLOWS IST-2001-32125 Deliverable No: D14 2.5 MIMO in UMTS and METRA project The Multi-Element Transmit and Receive Antennas (METRA) project, an IST research and development project, carried out from January 2000 to June 2001 [FGM+02, Hei01b]. The main objective of METRA was to study the feasibility of introducing multi-element adaptive antennas into the user equipment and the base station in UMTS. In this project, a stochastic channel model was developed, based on realistic multi-antenna channel measurements, different MIMO configurations and processing schemes were developed for both the FDD and TDD modes, and their link performance was assessed. 2.5.1 MIMO channel characterisation Different pico- and microcell environments have been investigated. It is shown that, in most investigated cases, a small spacing (half a wavelength) is enough to significantly decorrelate antenna elements. The frequency-selective fading channel model is more general and realistic with respect to the frequency-flat fading channel model. In METRA, results on capacity of frequency-selective MIMO channels were presented. It has been shown that the frequencyselectivity can increase the capacity at a given outage probability. 2.5.2 Transmit techniques in the FDD mode of UMTS Two versions of a layered MIMO scheme and a punctured scheme have been tested in a (2,2) MIMO channel. In the first layered scheme, a frame of data is split into two streams that are separately convolutionally encoded and transmitted from two antennas. The antennas transmit two different pilot sequences so that the receiver is able to estimate both channels. The receiver selects the stronger transmit antenna and the selected transmit signal is detected using a LMMSE receiver. The dual-antenna LMMSE receiver is able to suppress the interfering signal from the other transmit antenna very effectively; however, this results in loss of receive diversity. The detected signal is re-encoded, re-interleaved, reconstructed, and cancelled from the received signals. The other transmit signal is then detected using another dual-antenna LMMSE receiver. The first layered scheme does not offer any transmit diversity. In the second layered scheme, transmit diversity is achieved by switching the code word symbols between the two transmit antennas at the symbol rate. Two dual-antenna LMMSE receivers detect each of the transmit signals, and the two symbol streams are reconstructed by removing the effect of the antenna switching. After decoding, both streams are re-encoded, re-interleaved, reconstructed, and cancelled from the received signals. After the cancellation, the transmit signals are once again detected by two dual-antenna LMMSE receivers providing final decisions. The puncture scheme also achieves twice the data rate of conventional single-antenna transmission and offers full transmit diversity by applying space-time transmit diversity (STTD) with the Alamouti space-time block code and full receive diversity by using a dual-antenna Rake receiver. This scheme follows the current UMTS FDD specification and its complexity is small with respect to the layered schemes. The analysis of the frame error rate (FER) performance in a two-path pedestrian channel (3 km/h) has shown that the punctured scheme offers twice the data rate with approximately the 17 December 2003 Page 34

FLOWS IST-2001-32125 Deliverable No: D14 same transmit power than the best layered scheme, but with significantly lower complexity. The lack of diversity degrades the performance of the first layered scheme; however, its performance is similar to single-antenna transmission which offers only half of the data rate. In a five-path vehicular channel (50 km/h), both the layered schemes have approximately the same performance; this is due to a high order of multipath diversity. The punctured scheme also possesses about the same performance; if a dual-antenna LMMSE receiver was used instead of the dual-antenna Rake receiver, the punctured scheme would be expected to outperform the other techniques. A conclusion is that in lower-order MIMO channels such as (2,2) MIMO, puncturing and transmit and receive diversity is superior to layered BLAST-like MIMO techniques in terms of complexity and performance. 2.5.3 Transmit techniques in the TDD mode of UMTS For the downlink TDD mode, two groups techniques were considered: standard-friendly and standard-nonfriendly. The first group consists of a narrowband beamforming and space-time block coding schemes. In the narrowband beamforming approach, a different spatial filter is used for each user. Since each user sees a different channel, one midamble per user has to be used. In the second approach, a combination of the Alamouti space-time codes for the dedicated physical channels (DPHCHs) and beamforming are considered. The second (standard-nonfriendly) group consists of a layered scheme (BLAST) and a four-antenna STBC with a code of rate 3/4 were considered. In the BLAST, the data are demultiplexed into four different substreams, each transmitted from a different antenna, and a convolutional code with rate 1/4 is applied; after multiplexing detected data symbols in the receiver, a Viterbi detector is used to recover the original transmitted data. Uplink-oriented and downlink-oriented receive techniques have been considered. In the uplink, the base station knows the spreading codes and spreading sequences of all users, therefore multiuser techniques are considered. Among them are the multi-user denoised matched filter (DMF) and multi-user LMMSE receiver. The former is a single-user matched filter taking into account covariance of the intercell interference. The latter requires a multi-user channel estimation. In the downlink, the user equipment knows only the spreading code and training sequences corresponding to its own channel, so single-user techniques are applied. Among them are the single-user DMF, single-user LMMSE, detection of space-time block codes (STBCs), and reception of BLAST streams. The STBCs are detected with a multi-user LMMSE receiver, where the signal arriving from each antenna is treated as a separate user. The reception of BLAST streams is made using the multi-user LMMSE receiver, where the signal transmitted from each antenna is treated as a different user. 2.5.4 Other detection techniques for UMTS High performance receivers are critical to maximise capacity in mobile communication systems. An important challenge is for downlink (mobile terminal) receivers in 3rd generation systems such as UMTS. Such a downlink receiver is required to offer high capacity, e.g. for Internet download, coupled with low computational complexity and corresponding low power consump- tion. Conventional single-user RAKE receivers are considered to be the main candidates for implementation in mobile terminals, but their performance is limited with a high number of active users due to a strong multiaccess interference (MAI). Optimal multiuser receivers are complex for implementation and require knowledge of spreading sequences of all active users, whereas in UMTS each mobile user knows only its own spreading sequence [HT00]. 17 December 2003 Page 35

FLOWS IST-2001-32125 Deliverable No: D14 The downlink transmission possesses an inherent synchronism which can be used in the receiver for both simplification of implementation and improvement of detection performance. In [LH00] it was proposed to divide the reception into chip equalisation and decorrelation; any changes of the frequency-selective channel affect only the equaliser, which is easy to update, rather than the correlation matrix, which is complicated to invert. In that scheme, the decorrelating matrix is precomputed, which significantly simplifies the receiver. Additional simplicity arises due to orthogonal spreading sequences making the decorrelating matrix diagonal: after equalisation the multiuser detection can be performed by single-user matched filtering. A downlink chip-equalised receiver with orthogonal Walsh spreading sequences was proposed in [BF95]. There have been shown by simulation that, in a fully loaded system operating in a static multipath channel, such a receiver significantly outperforms a conventional RAKE receiver and the performance can be close to that of single-user transmission. For orthogonal spreading sequences, the chip-equalised receiver allows each mobile user to perform multiuser detection without knowledge of spreading sequences of all active users. The chip equalisation restores the orthogonality of user signals, lost due to the multipath propagation, which leads to suppression of the multi-access interference (MAI). Due to equalisation on the chip level, it can also be applied in systems using long scrambling codes, as in UMTS. Thus, in UMTS, the chip-equalised downlink receiver is an attractive alternative to the conventional RAKE receiver. Since the downlink receiver is implemented in a mobile user equipment, a simple implementation is particularly important. The chip-equalised receiver has received further investigation in the literature. In [Kle97] it has been proposed to use zero-forcing (ZF) and minimum mean-square error (MMSE) equalisers; This work has demonstrated the advantages of the receiver over the RAKE receiver in terms of detection performance and simplification. ZF equalisation as an alternative to the RAKE receiver was also proposed in [GS98]. The work [KZL00] compares the RAKE receiver with the chip-equalised receiver using ZF or MMSE equalisers with taps calculated by matrix inversions for perfectly known channels; this has shown that the MMSE equaliser outperforms both the RAKE receiver and ZF equalisation in single- and two-antenna receivers. A similar conclusion was in [HLaJ99] where fractionally spaced ZF and MMSE equalisers in Rayleigh fading channels under assumption of a perfectly known channel impulse response are investigated; this equalisation is also based on matrix inversion. It is emphasized in [HLaJ99] that development and analysis of adaptive chip-equalised receivers are important. In order to simplify the implementation and avoid matrix inversion, [KH99] proposes exploiting an adaptation rule derived from the bootstrap principle; further improvements of this approach, based on Grif- fiths’ adaptation (a modification of the LMS (Least Mean Squares) algorithm), were proposed in [KH00] and [Hei01a]. The simulations have demonstrated that these equalisation schemes result in a high detection performance in both single- and two-antenna receivers; however, this assumed perfect knowledge of channel impulse responses. In [SG00] a maximum SINR (signal-to-interference plus noise ratio) downlink equalisation for oversampled received signals in a multi-antenna receiver is derived, which also is based on knowledge of the channel impulse response. A fractionally spaced chip equaliser based on RLS (Recursive Least Squares) adaptation in time-invariant channels is proposed and investigated in [MS01]; comparison with the RAKE receiver demonstrates advantages of the chip-equalisation for high-order modulation schemes (QPSK and 16-QAM) in high loaded multiuser scenarios. Numerical results from the above show that a downlink receiver based on chip-equalisation can offer a significant performance improvement in comparison with the conventional RAKE receiver. Preferably, chip-equalisation should be based on the MMSE solution which requires solving a linear system of equations, for example, based on matrix inversion. The latter is complex for real-time implementation, whereas adaptive equalisation techniques, such as LMS and RLS, are more attractive for implementation. The RLS algorithm possesses a fast convergence (i.e. it results in a solution which is close to that based on matrix inversion), however, this technique has a high complexity and is computationally unstable - in particular, it requires multiplication and division operations. The latter do not facilitate hardware implementation and make real-time software implementation, e.g., on DSP (Digital Signal Processor) platforms, difficult. 17 December 2003 Page 36