A hybrid learning approach to blind deconvolution of MIMO systems

In this paper we present a hybrid learning method for blind deconvolution of linear MIMO systems. We propose a hybrid network that consists of a linear feedforward network followed by a linear feedback network, where each of synapses is represented by an FIR filter. The FIR synapses in the feedforward network are learned by the constant modulus algorithm (GMA) to recover source signals and at the same time, the FIR synapses in the feedback network are updated by spatio-temporal decorrelation algorithms so that different sources appear at different output nodes. As a spatio-temporal decorrelation task, we consider the extension of anti-Hebbian rule and the natural gradient-based learning algorithm. Useful behavior of the proposed hybrid network is verified by computer simulation results.