Signal mixture decomposition methods and algorithms. II. Application of м-gradient to analysis of independent components

The problem of signal mixture decomposition (with reconstruction of its component forms) is analyzed without any information concerning the proportions and type of mixing. The method is based on using information criteria and adaptive algorithm for learning neural network. Several forms of distributions were considered for original signals reaching the sensors of information-measurement system. Variations of neural transfer functions corresponding to several types of distributions are given. The approach to signal mixture separation includes application of m-gradient in the analysis scheme for independent component analysis.