Efficient Implementation of MIMO Decoders

Abstract

Multiple-Input Multiple-Output (MIMO) technology has emerged as a promising technology for achieving the high data rates of next generation wireless communication systems. MIMO systems improve either the bit-error rate (BER) performance by using spatial diversity or the date rate via spatial multiplexing. However, Maximum-likelihood (ML) detection for high order MIMO systems faces a major challenge in computational complexity that grows exponentially with the number of transmit and receive antennas. This limits the practicality of these systems from an implementation point of view, particularly for mobile battery-operated devices. This reality motivated researchers to consider other suboptimal approaches for MIMO decoding, such as Zero Forcing (ZF), Minimum Mean Square Error (MMSE) and VBLAST (Proakis & Salehi, 1994; Guo & Nilsson, 2003; Myllyla et al., 2005). All of these suboptimal approaches vary in performance and complexity. Recently, the sphere decoding (SD) algorithm which is a tree-based search algorithm enabled the implementation of efficient MIMO decoders that achieve near MLD performance together with reduced complexity (Burg et al., 2005; Barbero et al., 2005; Khairy et al., 2009). Instead of the exhaustive search over all possible combinations of the transmitted symbols, the SD algorithm reduces the complexity by searching only over a finite number of symbols within a circle of radius R centered at the received vector. While the SD approach provides a near ML solution, the runtime measured by the required operations to find the optimum solution is highly dependent on the received signal to noise ratio and the channel conditions. Consequently, the SD algorithm experiences variable throughput problems as the search radius R for each symbol varies according to the noise levels and the channel coefficients. Moreover, the sequential search results in hardware implementations that are not fully pipelined. To alleviate these problems, the fixed sphere decoding (FSD) algorithm was recently proposed (Barbero & Thompson, 2006 b). The FSD algorithm achieves a fixed throughput performance and enables fully-pipelined hardware by performing fixed number of operations per detected symbol, independent of the noise level. The FSD algorithm is performed in two stages where the first one is a pre-processing stage for the received