Underdetermined blind source separation based on Continuous Density Hidden Markov Models

In this paper, a novel method is developed to solve the problem of underdetermined blind source separation, where the number of mixtures is smaller than that of sources. Generalized Gaussian Distributions (GGDs) are used to model the source signals and generative Continuous Density Hidden Markov Models (CDHMMs) are derived to track the nonstationarity inside the source signals. Each source signal can switch between several states such that the separation performance can be significantly improved. The model parameters are trained through the Expectation Maximization (EM) algorithm and the source signals are estimated via the Maximum a Posteriori (MAP) approach. Compared with the results of L₁-norm solution, our proposed algorithm has obtained much better output signal-to-noise ratio (SNR) and the separation results are more realistic.