Independent residual analysis for temporally correlated signals

Abstract

In this paper, we study the blind source separation problem of temporally correlated signals via exploring both the temporal structure and high-order statistics of source signals. First, we formulate the problem as independent residual analysis and present a simple cost function. Efficient learning algorithm is developed for the demixing matrix and the corresponding stability analysis is also provided. The formulation provides much more flexibility for us to identify learning algorithms with good learning performance and stability. Furthermore, the approach unifies the conventional high-order statistical method and the second-order statistical method. From stability analysis, we infer that if the temporal filters of sources are mutually different, the second order statistical algorithm will be sufficient to separate the sources from their linear mixtures. © Springer-Verlag Berlin Heidelberg 2003.