Overcomplete Blind Source Separation Based on Second Order Statistics

Overcomplete blind source separation (BSS) has been a difficult problem for a long time. This paper proposed a novel overcomplete BSS algorithm based on second statistics. The first step is complete decorrelation processing by principal component analysis (PCA), then one strong correlation signal chosen as one input signal by the first canonical correlation analysis (CCA). At last, source signal estimation can be completed by multi-steps CCA. Simulation results show that this novel overcomplete BSS algorithm can be simply realized and the computation burden is small. This overcomplete BSS algorithm may be have a broad project application.