Improving performance in pulse radar detection using Bayesian regularization for neural network training

Abstract

A better approach for training a multi-layered feedforward network for pulse compression is presented. The Bayesian regularization technique used for training the network for pulse radar detection results in superior performance in terms of signal-to-sidelobe ratio compared to the Backpropagation algorithm. The presented method also has better range resolution performance in terms of resistance to lower input code magnitude ratios. 13-bit Barker code, 31-bit m-sequence and 63-bit m-sequence are used as the signal codes. (C) 2004