Low-complexity systolic V-BLAST architecture

Toshiaki Koike-Akino
2009 IEEE Transactions on Wireless Communications  
In multiple-input multiple-output systems, an ordered successive interference canceller, termed the vertical Bell laboratories layered space-time (V-BLAST) algorithm, offers good performance. This letter presents a low-complexity V-BLAST scheme suited for parallel implementation. The proposed scheme, using a greedy ordering, can achieve a performance comparable to that of V-BLAST with optimum ordering, while its computational complexity is lower than a linear detector. Index Terms-MIMO,
more » ... greedy ordering I. INTRODUCTION R ADIO communications systems using multiple antennas at both transmitter and receiver, to form a multiple-input multiple-output (MIMO) system, have attracted attention, as a promising technique for achieving a significant increase in spectrum efficiency [1, 2] . A large number of MIMO detectors, such as linear detectors based on minimum mean-square error (MMSE) and based on zero-forcing (ZF), have been studied so far. Although linear detectors are generally low in complexity, their performance can be poor, especially in MIMO systems that use a small number of receiving antenna branches. To improve performance, a so-called vertical Bell laboratories layered space-time (V-BLAST) algorithm has been introduced; this performs successive interference cancellations in the appropriate order [3] [4] [5] [6] [7] [8] . V-BLAST yields higher diversity gains and improves bit-error-rate (BER) performance. The computational complexity of V-BLAST, when compared to a linear detector, is generally increased from O[M 3 ] to O[M 4 ]. Here, O[·] denotes the order of complexity and M denotes the number of antenna pairs. In time-varying MIMO channels, since the V-BLAST requires re-ordering and weight updating, the complexity reduction is significant for any realistic application. Hitherto, some advanced algorithms have been derived to decrease the complexity from O[M 4 ] to O[M 3 ] (see references in [5-10] for example). In [5], Wai et al. have introduced a low-complexity ZF/V-BLAST using Gram-Schmidt orthonormalization (GSO) and sub-optimum ordering. However, ZF/V-BLAST does not offer such good performance as does MMSE/V-BLAST, especially at a low SNR regime. Hassibi has proposed a square-root algorithm to reduce the complexity of MMSE/V-BLAST [6]. However, it is only advantageous with a very large number of antenna pairs, as discussed in [8]. In [7], Benjebbour et al. have studied a Manuscript
doi:10.1109/twc.2009.070498 fatcat:mx63his2sbb77pc6uy35osm5ye