A robust preprocessing method for noisy blind source separation

Most of the blind source separation (BSS) models assume that the independent sources are observed through some unknown mixing process in the absence of noise. However, the additive noise may be present in many practical applications. The bias caused by noise will limit the separation performance of the BSS algorithm. The existing modified algorithms based on quasi-whitening only can improve the signal-to-interference ratio (SIR) of the separated signals, but cannot effectively suppress the additive noise. This paper addresses this difficult problem. We introduce a robust preprocessing approach, which allows one to find the minimum mean squared error (MMSE) separation matrix up to an orthogonal or unitary factor. Combining the fast independent component analysis (FastICA) algorithm, the mixed signals can be robust separated. Simulation results show that the proposed algorithm improve the signal-to-interference-plus-noise ratio (SINR) of the separated signals.