Multichannel blind equalization is an important task for numerous applications such as speech separation

dereverberation

communication

signal processing and control

etc.In this paper

we reconstruct a cost function based on second order statistics.To avoid trivial solution

an additional item is added to the cost function

which assures that the equalizer is irreducible.Then

an algorithm is derived with natural gradient search method for multichannel blind source separation and deconvolution of convolved signal mixtures.To avoid divergence of the algorithm

an iterative converging condition is also given

and is used to adjust the learning rate.Simulations indicate the ability of the algorithm to perform blind equalization under the weaker condition (the FIR system is equalizable).