On blind separation of convolutive mixtures of independent linear signals

This paper is concerned with the problem of blind separation of independent signals (sources) from their linear convolutive mixtures. The problem consists of recovering the sources up to the shaping filters from the observations of the MIMO system output. The various signals an assumed to be linear but not necessarily i.i.d. (independent and identically distributed). An iterative, normalized higher-order cumulant maximization based approach is developed using the third-order and/or fourth-order normalized cumulants of the "beamformed" data. The approach is source-iterative, i.e., the sources an extracted (at each sensor) and cancelled one-by-one. The proposed solution provides a decomposition of the given data at each sensor into its independent signal components. The proposed approach is an extension/application of a previously proposed approach for MIMO system identification where the system is driven by unobserved i.i.d. inputs.