Space-Time Coding for Large Antenna Arrays

Abstract

Multiple-input multiple-output (MIMO) systems can greatly improve the capacity and performance of wireless communications. In particular, space-time coding techniques have received much attention in recent years as an efficient approach to achieving the performance gains offered by MIMO channels. Thus far, most work on space-time coding has focused on systems with small antenna arrays or high signal-to-noise ratios (SNRs), for which it has been shown that codes should be designed according to the rank and determinant criteria. For such scenarios, coherent space-time coding and differential space-time modulation (DSTM) schemes have been designed, for systems with or without channel knowledge at the receiver, respectively. In recent years, there has been some work on coherent space-time coding for large arrays, which indicates that the code design metric should be chosen diffently from that for small arrays. In this dissertation, we study the design of space-time coding for large arrays. We focus on three aspects: performance analysis, code construction and decoding algorithms.

We first analyze the asymptotic performance of differential space-time modulation. A new upper bound on the pairwise-error probability is derived for large arrays. This bound suggests that Euclidean distance is an appropriate design criterion for DSTM with large numbers of antennas, which is similar to the design of coherent space-time coding for the large-array regime. For two transmit antennas and four or more receive antennas, we use the new design criterion to obtain several new unitary codes with large minimum Euclidean distance. The proposed codes outperform some existing codes, for example, the well-known Alamouti code, for large receive arrays.

Although the codes designed according to the new design criterion achieve good performance, most of them require maximum-likelihood (ML) decoding, which is undesirable for high-rate codes. On the other hand, the Alamouti code, which is designed for high-SNR regime, enables simple linear ML decoding. It is of interest to design codes that perform well for large arrays, but which also allow simple decoding at the receiver. We first consider the design of unitary codes, for use with and without channel knowledge at the receiver. For two transmit antennas, we consider a structure which is a modification of the Alamouti code. We optimize the new code with respect to the Euclidean distance criterion. We then show that the new code allows us to use two suboptimal decoders that have complexity comparable to the Alamouti decoder. The analytical bit-error performance and the constellation-constrained capacity are derived for the suboptimal decoders. For coherent detection, the coding structure is extended to non-unitary constellations. We also extend the new code to more than two transmit antennas.

Conventional DSTM assumes that the channel remains constant for two adjacent transmission blocks, which is questionable for some time-varying channels. In this dissertation, we investigate the performance of the new code when fast-fading is encountered. We show that multiple-symbol decision-feedback differential detection (DFDD) can be used to reduce the performance degradation of the new code in fast-fading channels. We also consider the use of suboptimal decoders in DFDD to further reduce the decoding complexity.