Blind source separation for convolutive mixtures using a non-uniform oversampled filter bank

Adaptive subband structures have been proposedwith the objective of increasing the convergence speed and/or reducing the computational complexity of adaptation algorithms for applications which require a large number of adaptive coefficients. In this paper we propose a blind source separation method for convolutive mixtures which employs a real-coefficient non-uniform filter bank and a new normalization scheme for the adaptation algorithm. Since the separation filters in the subbands work at reduced sampling rates, the proposed method presents smaller computational complexity and faster convergence rate when compared to the corresponding fullband algorithm.