A PARAFAC decomposition based algorithm for blind MIMO source separation

The paper deals with the problem of blind source separation after a MIMO convolutive mixture. We propose an algorithm for the simultaneous extraction of all the sources. It is based on the PARAFAC decomposition of a tensor built from the observations and from so called reference signals. In particular this algorithm allows to overcome the classical drawbacks of the deflation approach in the sequential separation scheme. The order of the PARAFAC decomposition depends on the mixture parameters, the extraction filter length and the number of sources. Then a selection among these PARAFAC factors is proposed, in order to obtain the different sources. A fixed point method improves then the estimation performances iteratively. Computer simulations illustrate the good behavior and the interest of our algorithm in comparison with other approaches.