A novel permutation algorithm in frequency-domain Blind Source Separation

This paper presents a novel permutation algorithm for solving the permutation ambiguity of frequency-domain BSS(Blind Source Separation). The proposed algorithm harnesses the characteristics that signals at adjacent frequency bins have a high similarity and utilizes Dynamic Time Warping (DTW) to compare this similarity in order to obtain an adjusting matrix at adjacent frequency bin. In real experimental environment, ASSNR(Average Segmental SNR) was utilized to evaluate the separation quality and the proposed method is found to improve the separation over the traditional criterions.