Optimization of a neural network applied to pulsed radar detection

The purpose of this paper is to present the results of the optimization of some features of a Neural Network applied to the binary detection problem. We present how to design the structure and the training sets, and how to modify the BackPropagation algorithm to improve the results of the network for the binary detection task. The Neural Network, so designed, presents an optimal range of pulse integration and a performance very close to the theoretical limits, even under input distributions different from those used for training.