An Extension of MISEP for Post–Nonlinear–Linear Mixture Separation

Mutual information separation (MISEP) is a versatile independent component analysis (ICA) algorithm that can be used to handle linear and nonlinear mixtures. By incorporating the
a

priori
information of mixtures, an extended MISEP method is proposed in this brief to recover the source signals from the post-nonlinear-linear (PNL-L) mixtures. One group of multilayer perceptrons and two linear networks are used as the unmixing system, and another group of multilayer perceptrons is used as the auxiliary network. The learning algorithm of the system parameters is obtained by maximizing the output entropy with the gradient ascent method. Experimental results demonstrate that the proposed method is effective and efficient for PNL-L mixture separation.