Analysis of the convergence properties of a self-normalized source separation neural network

An extended source separation neural network was derived by Cichocki et al. (1995) from the classical Herault-Jutten network. It was claimed to have several advantages, but its convergence properties were not described. In this paper, we exhaustively define the equilibrium points of the standard version of this network and analyze their stability. We prove that the stationary independent sources that this network can separate are the globally sub-gaussian signals. As the Herault-Jutten network applies to the same sources, we show that the advantages of the new network are not counterbalanced by a reduced field of application, which confirms its attractiveness.