Blind source separation from noisy data using bayesian estimation and gibbs priors

This paper deals with the blind separation and reconstruction of source signals from their mixture with unknown coefficient, in the practical ase where noise affects the mixtures themselves. We address the blind source separation problem within the ICA approach. We propose a MAP estimation method based on simulated annealing to recover both the mixing matrix and the sources, and experimentally verify that a model for the source signals which accounts for time correlation is able to increase robustness of the estimates against noise in the data.