Parallel blind source separation by kurtosis maximization with successive prewhitening

Blind source separation (BSS) has been of great interest in many areas such as wireless communications and biomedical imaging. Since many of the existing BSS methods are based on a successive cancellation procedure for extracting all the unknown sources, they suffer not only from the error propagation accumulated at each stage but also from a long processing latency. Chen et al. recently proposed two effective prewhitening non-cancellation multistage (PNCMS) parallel blind source separation (BSS) algorithms, one using the fast kurtosis maximization algorithm (FKMA), called the PNCMS-FKMA(p), and the other using the turbo source extraction algorithm (TSEA), called the PNCMS-TSEA(p), which not only have significantly reduced processing latency but also have improved source extraction performance. In this paper, we further propose two computationally improved BSS algorithms, called the SPNCMS-TSEA(p) and SPNCMS-FKMA(p), by incorporating prewhitening processing at each source extraction stage. This successive prewhitening processing can literally cut down the dimension of the multisensor data, and therefore the computational complexity of TSEA/FKMA is decreased from one stage to another. The efficacy of the proposed algorithms is verified by computer simulations.