Blind Source Separation for MIMO-AR Mixtures Using GMM

Abstract

The problem of blind source separation (BSS) of multiple-input multiple-output (MIMO) autoregressive (AR) mixture is addressed in this paper. A new time-domain method for system identification and BSS for MIMO-AR models in proposed based on the Gaussian mixture model (GMM) for sources distribution. The algorithm is based on generalized expectation-maximization (GEM) for joint estimation of the AR model parameters and the GMM parameters of the sources. The method is tested via simulations of synthetic and audio signals mixed by a MIMO-AR model. The results show that the proposed algorithm outperforms the well-known multidimensional linear predictive coding (LPC), and it enables to achieve higher signal-to-interference ratio (SIR) in the BSS problem.