Robust source separation using ranks

Robustness against deviations from nominal source PDF assumptions is very desirable in blind source separation (BSS) algorithms. A new approach for robust BSS is proposed. We modify the EASI (equivariant adaptive separation by independence) algorithms to use ranks of observed signals. Two different methods for evaluation of ranks are introduced. Our modified algorithm can be applied to both real-valued data and complex-valued data. Design guidelines are discussed for the nonlinear rank weighting functions in the modified algorithm. Simulation results and some examples are given, showing very good performance.