Bayesian blind marginal separation of convolutively mixed discrete sources

We formulate the discrete source separation problem in a Bayesian framework. We show that it is possible to integrate analytically the so called nuisance parameters, leading to an analytic expression of the marginal posterior distribution of the symbols conditional upon the observations. We present two algorithms, a deterministic algorithm and stochastic algorithm, that allow one to optimize the marginal posterior distribution. We present simulation results and draw some conclusions