Expectation-Maximization Algorithm for Noisy Independent Component Analysis

Expectation-maximization (EM) algorithm is applied in the noisy independent component analysis (ICA) model, i.e., the source signals are assumed statistical independent and formulated in a Bayesian estimation framework. A Bayesian approach with EM algorithm for noisy ICA is proposed. In the noisy ICA model, supposing the means and variances of source signals are uniform, the proposed EM algorithm can efficiently estimate the model parameters of the mixing matrix and hyperparameters under a certain model, and then estimate the sources by processing the mixing matrix and hyperparameters alternatively. Simulation results show that the proposed method can perform blind source separation (BSS) with the noisy ICA model.