Weak Signal Detection Based on Chaotic Prediction

Abstract

In this paper, a weak signal detection method based on radial basis function (RBF) neural networks is discussed. The principle of weak signal detection with a background noise predictor is that the predictor trained by chaotic time series has a small prediction error, and the prediction error becomes relative large when the input contains a source or target. By exploiting the short-term predictability of the input signal, a one-step-ahead prediction model is proposed as the basis of designing an RBF neural network. To enhance the detection performance in noisy background, the extended Kalman filter (EKF) is applied to perform the training and better parameter estimates can be acquired compared to the conventional RBF network training method. The performance of detection for low signal-to-noise ratio (SNR) is analyzed. Computer simulations show that the proposed method is effective for weak signal detection.