Convolutive sources separation combining optimized sub-band filter with instantaneous separator

Blind source separation (BSS) is a general and promising technique for signal processing, which can be used to recover the contributions of different physical sources that only from a finite set of observations recorded by sensor. This method is attractive for many fields of applied sciences and engineering including medicine, telecommunication, audio processing, noise reduction or data processing, health condition monitoring and fault diagnosis of machine. However, the results by the existing BSS algorithms are often not enough for the subsequent signals analysis, on account of the limited assumption on the mixture models. In this paper, special improvements on the existing BSS algorithms were made, which made it possible to restore sources waveform more accurately. The validity of the new method was verified by several selected experiments, which further showed the potential of such technique in practice.