The Best Blind MMSE Estimator


This is a variant of the well known FastICA algorithm that is proposed to be used for blind source separation in off-line (block processing) setup and a noisy environment. The algorithm combines Symmetric FastICA with test of saddle points to achieve fast global convergence and one-unit refinement to obtain high noise rejection ability. It’s computational complexity is similar to that of the original FastICA, i.e. it is fast. It is shown that the bias of the algorithm due to additive noise is reduced; it is proportional to σ^3, where σ^2 is the variance of the additive noise, while the bias of the other methods (namely of all methods using the orthogonality constraint, and even of recently proposed EFICA algorithm) is asymptotically proportional to σ^2. Thanks to the reduced bias, the novel algorithm exhibits a significantly lower symbol-error rate when it is applied to blindly separate mixtures of finite alphabet signals that are typical for communication systems.

Corresponding papers:

  • Z. Koldovský and P. Tichavský, “Blind Instantaneous Noisy Mixture Separation with Best Interference-plus-noise Rejection”, Proceedings of 7th International Conference on Independent Component Analysis (ICA2007), pp. 730-737, Sept. 2007. (here)
  • Z. Koldovský and P. Tichavský, “Asymptotic Analysis of Bias of FastICA-based Algorithms in Presence of Additive Noise”, technical report nr. 2181, ÚTIA, AV ČR, Jan 2007.

Matlab codes: 1FICA for real signals; for complex signals FicaCPLX available at personal pages of Petr Tichavský

Copyright: Zbyněk Koldovský, Petr Tichavský