A neural network based method for detection of weak underwater signals

Abstract

Detection of weak underwater signals is an area of general interest in marine engineering. A weak signal detection scheme was developed; it combined nonlinear dynamical reconstruction techniques, radial basis function (RBF) neural networks and an extended Kalman filter (EKF). In this method chaos theory was used to model background noise. Noise was predicted by phase space reconstruction techniques and RBF neural networks in a synergistic manner. In the absence of a signal, prediction error stayed low and became relatively large when the input contained a signal. EKF was used to improve the convergence rate of the RBF neural network. Application of the scheme to different experimental data sets showed that the algorithm can detect signals hidden in strong noise even when the signal-to-noise ratio (SNR) is less than −40d B.