Informax algorithm based on linear ICA neural network for BSS problems

In the paper, given the condition that the source signals are statistically independent, an extended Informax blind source separation (BSS) algorithm is presented using a linear ICA neural network, based on the fundamental rules of information-maximization. Aiming at source signals with super-Gaussian and sub-Gaussian distributions, this algorithm can successfully separate each independent source signal when the mutual information between input and output signals is maximized. Using the algorithm the weight matrix or separation matrix can be obtained with fast convergence speed. Experimental results illustrate the good performance of the algorithm.