Abstract
Chi et al. recently proposed two effective non-cancellation multistage (NCMS) blind source separation algorithms, one using the turbo source extraction algorithm (TSEA), called the NCMS-TSEA, and the other using the fast kurtosis maximization algorithm (FKMA), called the NCMS-FKMA. Their computational complexity and performance heavily depend on the dimension of multi-sensor data, i.e., number of sensors. This thesis proposes the inclusion of the prewhitening processing in the NCMS-TSEA and NCMS-FKMA before performing source extraction. We come up with two improved algorithms, referred to as PNCMS-TSEA and PNCMS-FKMA with significant computational savings on one hand, and some performance improvements on the other hand (owing to dimension reduction and noise reduction by prewhitening processing), especially when the number of sensors is much larger than the number of sources. Two implementation structures for the proposed PNCMS-TSEA and PNCMS-FKMA are considered. One is parallel structure (denoted as PNCMS-TSEA(p) and PNCMS-FKMA(p)) and the other is sequential structure (denoted as PNCMS-TSEA(s) and PNCMS-FKMA(s)). The performances of PNCMS-TSEA(p) (PNCMS-FKMA(p)) and PNCMS-TSEA(s) (PNCMS-FKMA(s)) are the same while the former is well suited to software and hardware implementations thanks to much smaller processing latency. Some simulation results are presented to verify the efficacy and computational efficiency of the proposed algorithms.