Mcmc Methods for Discrete Source Separation

Source separation consists in recovering signals mixed by an unknown transmission channel. Likelihood and information theory or higher order statistics can be used to perform the separation. This paper proposes a Bayesian approach to the problem of an instantaneous linear mixing, considering the source signals are discrete valued. The Bayesian inference enables to take in account jointly prior information and the information available on the observation signals. This approach implies complex calculations which can be achieved through Monte Carlo Markov Chain (MCMC) simulation methods. The separation method for binary inputs was exposed in and is now extended to PSK source signals.