Using neural networks for nonlinear and chaotic signal processing

The authors present preliminary results on the dynamic behavior of widely used feedforward neural filters and outline possible signal processing applications. It is shown that a feedforward neural network possesses chaotic dynamics, which are investigated via bifurcation plots and the evaluation of the Lyapunov exponents. Nonlinear predictors based on neural networks can be used to model and predict chaotic time series and at the same time provide an accurate method of evaluating the characteristic Lyapunov exponents of the underlying dynamical process. It is shown how synchronized chaotic neural filters can be used for information masking and signal reconstruction.<<ETX>>