Soft detection of spatially diverse wireless communications via stochastic sampling

This thesis investigates the application of the Metropolis-Hastings algorithm to the problem of symbol detection for Multiple Input Multiple Output (MIMO) wireless communication systems. The most common methods employed for MIMO detection approximate the maximum likelihood (ML) decision rule, by solving the closest lattice point problem. By contrast, Markov chain Monte Carlo (MCMC) methods such as the Metropolis-Hastings algorithm provide an efficient approximation for the maximum a-posteriori (MAP) decision rule. The key difference is that the MAP rule naturally incorporates prior information, in fact the ML rule is equivalent to MAP when no prior information is available. The incorporation of prior information makes MCMC based detectors ideal for use in an iterative receiver structure where the MIMO detector and channel decoder exchange information in a loop. The performance of a range of MIMO detectors based on the Metropolis-Hastings algorithm is characterised using mutual information transfer curves that illustrate how the detectors would behave when paired with a decoder, and the impacts of altering various parameters on the characteristics are demonstrated. The use of multiple MCMC sampling chains gives rise to perhaps the best characteristics. The feasibility of using multiple sampling chains in a practical MCMC based MIMO detector silicon chip is investigated by implementing various functional units that may be used as the building blocks for many of the detectors previously examined. The area requirements and maximum clock frequency of these functional units are used to assess their relative complexities. Finally a subset of the functional units are configured into a complete sampling chain core. The results of synthesizing that core are promising and demonstrate that MCMC based MIMO detectors are practical to implement and further work and optimisation is warranted.