Analysis of convolutive source separation methods based on self-normalized weight updating terms

In this paper, we define two associated convolutive source separation methods, by applying the same algorithm to two separating structures. This practical algorithm performs a self-normalization of the updates of the separating filter weights, by adaptively estimating the mean powers of nonlinear functions of the separating system outputs. We first describe the main advantages of these methods and then analyze their convergence properties (locations and stability of all equilibrium points) with respect to those of the corresponding non-normalized methods.