Semi-blind identification of wideband MIMO channels via stochastic sampling

In this paper we address the problem of wide-band multiple-input multiple-output (MIMO) channel (multidimensional time invariant FIR filter) identification using Markov chains Monte Carlo methods. Towards this end we develop a novel stochastic sampling technique that produces a sequence of multidimensional channel samples. The method is semi-blind in the sense that it uses a very short training sequence. In such a framework the problem is no longer analytically tractable; hence we resort to stochastic sampling technique. The developed technique samples the channel, the variance of the noise and the symbols in order to build an ergodic Markov chain whose equilibrium distribution is the distribution of interest. The estimates of the MIMO channel and the noise variance are inferred from marginal posterior distributions, which are by-products of the output of the algorithm.