Artificial neural networks for the generation and estimation of chaotic signals

Abstract

Dynamic feedback, inversion and LMS estimation have been established for the estimation of an information signal encoded onto a chaotic carrier. The poor resultant SNR/sub sig/ of the recovered signal limits the applicability of these methods. Two novel chaotic coding/decoding strategies based on artificial neural networks (ANN) and radial basis functions (RBF) have been developed and the resultant performance has been assessed. The results indicate that the nonlinear predictor (ANN-RBF-NLP) offers performance independent of the channel SNR (for SNR>10 dB) and offers 4 dB improved SNR/sub sig/ compared to the LMS. Pseudo-chaotic sequences generated using an ANN and estimated in a dynamic feedback manner (ANN-RBF-DF) have resulted in a system with an SNR/sub sig/ that is linearly dependent on the channel SNR and offering for example 20 dB improved SNR/sub sig/ compared to the LMS at a channel SNR of 40 dB.