Blind separation with unknown signs of the kurtoses of sources
Gang Wei · Journal of Circuits and Systems · 2005
Most blind source separation algorithms assume that the signs of the kurtoses of the source signals are known, according to which nonlinear functions are chosen to approximate the score functions. To tackle the cases where the signs of the kurtoses are unknown, we propose a new algorithm exploiting the parametric estimation of the score function. The proposed algorithm can separate super-Gaussian, Gaussian and sub-Gaussian signals from their mixtures. Validity and performance of the proposed algorithm are demonstrated by extensive computer simulations.