Blind source separation for overcomplete mixtures with noise

Blind Source Separation (BSS) algorithms based on Independent Component Analysis (ICA) generally make several standard assumptions in ICA. The number of sources is assumed to be equal to the number of sensors and the presence of noise is negligible or Gaussian. However, these assumptions do not always hold in certain applications and may result in less than optimum performance from the algorithms. The aim of this paper is therefore to propose an algorithm which is capable of separating signals where the model involves fewer observed signals than the unknown source signals. Furthermore, the presence of noise is to be addressed and a noise reduction procedure is merged with the separation process. A set of simulations and comparison of results are presented to justify the accuracy and advantage of the proposed algorithm. Key-Words: Blind Source Separation, Overcomplete Mixtures, Noise Reduction

[1]  P. Comon,et al.  Blind identification and source separation in 2/spl times/3 under-determined mixtures , 2004, IEEE Transactions on Signal Processing.

[2]  A. Cichocki,et al.  MEASURING SPARSENESS OF NOISY SIGNALS , 2003 .

[3]  Satnam Dlay,et al.  Robust nonlinear signal separation using regularised maximum likelihood neural network , 2003 .

[4]  Bhaskar D. Rao,et al.  Subset selection in noise based on diversity measure minimization , 2003, IEEE Trans. Signal Process..

[5]  Laurent Albera,et al.  Fourth order blind identification of underdetermined mixtures of sources (FOBIUM) , 2003, 2003 IEEE International Conference on Acoustics, Speech, and Signal Processing, 2003. Proceedings. (ICASSP '03)..

[6]  Shun-ichi Amari,et al.  Natural Gradient Learning for Over- and Under-Complete Bases in ICA , 1999, Neural Computation.

[7]  Bhaskar D. Rao,et al.  Sparse Bayesian learning for basis selection , 2004, IEEE Transactions on Signal Processing.

[8]  Yehoshua Y. Zeevi,et al.  Separation of semireflective layers using sparse ICA , 2003, 2003 IEEE International Conference on Acoustics, Speech, and Signal Processing, 2003. Proceedings. (ICASSP '03)..