The neural networks method for detecting weak signals under chaotic background

Abstract

A method for detecting weak signals embebed in chaotic noise by neural networks based on the theory of phase space reconstruction of the complicated nonlinear system is presented. One-step predictive model for chaotic background is built by neural network that possess powerful cap ability of learning and nonlinear processing. Then the weak transient signal or periodic signal which is embedded in the chaotic background can be detected from the predictive error. And the detecting ability of this method when the chaotic background is mixed with white noiseis studied. The difference in the detecting principle for the transient signal and periodic signal is pointed out. The experiment which takes the Lorenz system as chaotic background shows this method can effectively detect very weak signals embedded in the chaotic background.