A narrowband approach to blind source separation in convolutive MIMO mixtures

In this contribution, we adopt a novel and simple narrowband approach to blind separation of mutually independent and temporally i.i.d. sources in convolutive MIMO mixtures. After remodelling the observation at the sensor array as an instantaneous mixture where the delayed replicas of the desired signals from multipath propagation are seen as separate sources independent of each other, we estimate them all in an analysis step by a conventional narrowband blind source separation (BSS) technique. In a synthesis step, we match the delayed replicas of each source based on cross-correlation then combine them after proper time & phase alignments and weighting. In the process, this multipath matching and combining procedure identifies the convolutive MIMO channel and is able to provide accurate expressions for direct blind deconvolution (BD) equalizers. Simulations support the efficiency of the new narrowband approach to BD.