Neural networks for blind decorrelation of signals

This work analyzes and derives a robust locally adaptive network structure based on a posteriori output signals that remains stable for any step-size value and extends a class of adaptive networks for second-order blind decorrelation of instantaneous signal mixtures.

Abstract

We analyze and extend a class of adaptive networks for second-order blind decorrelation of instantaneous signal mixtures. First, we compare the performance of the single-layer neural network employing global knowledge of the adaptive coefficients with a similar structure whose coefficients are adapted via local output connections. Through statistical analyzes, the convergence behaviors and stability bounds for the algorithms' step sizes are studied and derived. Second, we analyze the behaviors of locally adaptive multilayer decorrelation networks and quantify their performances for poorly conditioned signal mixtures. Third, we derive a robust locally adaptive network structure based on a posteriori output signals that remains stable for any step-size value. Finally, we present an extension of the locally adaptive network for linear-phase temporal and spatial whitening of multichannel signals. Simulations verify the analyses and indicate the usefulness of the locally adaptive networks for decorrelating signals in space and time.