An improved solution to the subband blind source separation permutation problem based on optimized filter banks

Subband based blind source separation (BSS) has a great potential in solving the complicated convolutive mixing problems. However, its performance is largely affected by the permutation ambiguity problem during the synthesis stage. Researchers have suggested methods to correct the permutation by using the correlation information between adjacent frequencies/subbands. In this paper, we propose an improved solution to this permutation problem based on a novel filter banks design method, which uses a model that includes inter-subband correlations as part of the optimization criterion. Simulation results show that a much better subband permutation alignment has been achieved, leading to an improved overall separation result.