Skip to Main Content
We consider the problem of identification of a n/spl times/n convolutive mixture. The sources are independent and non-white, with known statistics. We have recently proposed a second-order statistics based identification approach of Diamantaras, Petropulu and Chen (see IEEE Trans. on Signal Processing, 1999), that utilizes eigenvalue decomposition of second-order spectral correlations. In this paper we conduct a preliminary analysis of the method of Diamantaras et al., that provides some insight on the selection of parameters involved, and indicates how the channels/signals characteristics affect the reconstruction result.