Hybrid learning approach to blind deconvolution of linear MIMO systems

A hybrid network is presented which performs blind deconvolutions of linear MIMO systems. The hybrid network consists of a feedforward network followed by a feedback network, where each of the synapses is represented by an FIR filter. The FIR synapses in the feedforward network are learned by a Godard cost based algorithm and the FIR synapses in the feedback network are updated by a spatio-temporal decorrelation algorithm so that different sources are recovered at different output nodes. An efficient spatio-temporal decorrelation algorithm based on the natural gradient is presented. The validity of the proposed method is confirmed by computer simulations