A Block-Adaptive Subspace Method Using Oblique Projections for Blind Separation of Convolutive Mixtures

This paper presents a block-adaptive subspace algorithm via oblique projection for blind source separation (BSS) problem of convolutive mixtures. In the proposed algorithm, the problem is reformulated into the one of instantaneous mixtures through oblique projections within one block, and then the separation matrix and other model parameters are updated by any static separation approach in a block-adaptive scheme. Compared with other work, the proposed algorithm can obtain lower computational complexity, faster convergence and higher robustness. Simulation results of modulation signals and real speech sources validate the proposed algorithm.

• Type: Conference
• Year: 2005
• Where: ISNN
• Authors: Chunyi Peng, Xianda Zhang, Qutang Cai
• Keywords: Blind Source Separation, Block-adaptive Subspace Algorithm, ISNN 2005, Oblique Projections