A NEURAL-NET FOR BLIND SEPARATION OF NONSTATIONARY SIGNALS

This paper proposes a neural network that recovers some original rando m signals from their linear mixtures observed by the same number of se nsors. The network acquires the function with a learning process witho ut using any particular information about the statistical properties o f the sources and the coefficients of the linear transformation except the fact that the source signals are statistically independent and no nstationary. The learning rule for the network's parameters is derived from the steepest descent minimization of a time-dependent cost funct ion that takes the minimum only when the network outputs are uncorrela ted with each other.