Asymmetric PCA neural networks for adaptive blind source separation
Konstantinos Diamantaras · 2002
We show that second order cross-coupled Hebbian rule used for asymmetric principal component analysis is capable of blindly and adaptively separating uncorrelated sources. Our method enjoys the following advantages over similar higher-order models such as those performing independent component analysis: 1) the strong independence assumption about the source signals is reduced to the weaker uncorrelation assumption; 2) there is no constraint on the sources PDFs, i.e., we remove the assumption that at most one signal is Gaussian; 3) the higher order statistical optimization methods are replaced with second order methods with no local minima; and 4) the kurtosis of the sources becomes irrelevant. Simulation experiments shows that the model successfully separates source images with kurtoses of different signs.