A new variable step-size EASI algorithm based on mutual information

In view of slow convergence for fixed-step Equivariant Adaptive Separation for Independent (EASI )algorithm and the problem for variable step-size algorithm which bases on kurtosis is sensitive to outlier, a new variable step-size EASI algorithm is proposed, which applys the negative entropy maximization method of non-polynomial functions to the approximate calculation of the mutual information. Experiments results show that the proposed algorithm not only achieves faster convergence and smaller stead-state error than fixed-step EASI and other variable step-size algorithms, but also demonstrate better stability for the problem of outlier.