Simple adaptive algorithms for blind source separation of noisy mixtures

In most practical blind source separation (BSS) applications, the measured mixtures contain additive noise that limits the performance of most existing BSS algorithms. In this paper, we present several new methods for blindly extracting sources from noisy linear mixtures. The methods combine subspace tracking and source separation in an elegant fashion. Both density-modeling-based and decorrelation-based approaches are described. We also show how to modify the methods so that minimum mean-square-error (MMSE) or Wiener estimation of the unknown sources is performed. Simulations verify the robust and accurate behavior of the methods.