A fast converging sequential blind source separation algorithm for cyclostationary sources

A fast converging natural gradient algorithm (NGA) for the sequential blind separation of cyclostationary sources is proposed. The approach employs an adaptive learning rate which changes in response to the changes in the dynamics of the sources. This way the convergence and the robustness to the initial choice of parameters are much improved over the standard algorithm. The additional computational complexity of the proposed algorithm is negligible as compared to the cyclostationary NGA method. Simulations results support the analysis