Blind separation of nonstationary and temporally correlated sources from noisy mixtures

We present a new method of blind source separation that is robust with respect to additive white noise. Our method exploits the nonstationarity and temporal structure of sources. The method needs only multiple time-delayed correlation matrices of the observation data at several different time-windowed frames to estimate the mixing matrix. We present an implementation based on the joint diagonalization. Extensive simulations verify the high performance of the proposed method, especially in a low SNR environment