Abstract: On the basis of the model of multi-source mechanical vibration propagation and convolution mixtures, a blind source separation (BSS) algorithm for convolution mixture signals with multiple vibration sources based on time domain is proposed . This algorithm takes independence as judgment criterion, and carries out the learning of filter coefficient by adopting the mode of backward learning and rational simplification of filter coefficient, then realizes the separation of convolution mixture signals with multiple vibration sources based on time domain. The results of simulation test and multi-source mechanical vibration test indicate that the algorithm can achieve good effect in separating multi-source convolution mixture signals.

Key words: Blind source separation, Convolutive mixture, Mechanical vibration, Multi vibration sources, Time domain