Orthogonal Subspace-Based Blind Separation

An orthogonal subspace-based method is proposed for the blind separation of convolutive mixtures of nonstationary colored signals. The proposed method relies on eigenvalue decomposition (EVD) of a specially constructed matrix, which contains inform from two orthogonal subspaces. A multiple-input multiple-output (MIMO) blind deconvolution problem can be converted to multiple single-input multiple-output (SIMO) blind deconvolution problems by using the EVD. Numerical simulation demonstrates that the method can successfully separate audio and speech signals from their convolutive mixtures.