The blind deconvolution of the multi-channel based on the higher order statistics
Janghoon Yang, Chrysostomos L. Nikias · 2002
We have proposed a source separation algorithm of the convolutive mixtures based on the maximization of the auto-kurtosis and minimization of the cross-kurtosis with the constraint on the output power. As an iterative method, we suggest a nontrivial extension of the generalized eigenvector algorithm for blind equalization (GEnEVA) to the blind deconvolution of the multi-input multi-output (MIMO) systems. The application of the proposed algorithm on the 64-QAM signal separations shows that it can achieve excellent performance and it is robust to the broad range of the noise levels.