Quadratic MIMO contrast functions for blind source separation in a convolutive context

Abstract

This paper considers the problem of blind separation of sources mixed by a MIMO convolutive system. For both i.i.d. and non i.i.d. sources, quadratic separation criteria previously designed for the extraction of a single source are extended to parallel extraction in the MIMO case. These criteria are based on the use of so-called reference signals and a condition is given under which we obtain MIMO contrast functions. Simulations demonstrate that a particular choice of a set of reference signals ensures the contrast property. The performance offered by these criteria is investigated through simulations: it is shown that the proposed contrast functions avoid accumulation errors, contrary to deflation methods. © Springer-Verlag Berlin Heidelberg 2006.