An adaptive approach to subband domain convolutive blind source separation

An adaptive approach of Blind Source Separation (BSS) based on subband state space modeling has been proposed in this paper. This algorithm is explicitly designed for convolutive mixtures of communication signals with automatic wavelet shrinkage denoising. The proposed approach is more reliable due to the combination of subband blind source algorithm and wavelet de-noising. It ultimately resolves the drawbacks of BSR algorithm in noisy environment of communication signals. Moreover, the speed of algorithm has been improved by employing shorter de-mixing filter in each sub band domain. The proposed methodology has improved efficiency in term of speed and accuracy as compared to the previous approaches. Better separation ability has been achieved by using an adaptive step size based on non-linear function in each subband. Particular settings have been applied to “de-mixing filter length” in various subbands to enhance the separation capability. Wavelet shrinkage de-noising has been applied to the recovered output of each subband to improve the algorithmic performance in noisy environment. “SURE Shrink” model is used for nonlinear soft thresholding of approximate coefficients of discrete wavelet transforms. It results improved performance in noisy environment. 16QAM communication signals with same carrier frequency are considered for Computer simulation in MATLAB for the verification of the improved results of the proposed scheme.