MLPs for Detecting Radar Targets in Gaussian Clutter

A neural network based coherent detector is proposed for detecting gaussian targets in gaussian clutter. Target and clutter ACF are supposed gaussian with different powers and one lag correlation coefficients. While clutter mean Doppler frequency is set to 1, the influence of target mean Doppler frequency is considered. The neural detector performance is compared to the Neyman-Pearson one. For evaluating the neural detector performance, Montecarlo Simulation and Importance Sampling Techniques are used in order to assure a low relative error with a suitable computational charge. Results show that a low complexity neural network can implement very good approximations of the Neyamn-Pearson detector for the case of study. In the presented cases, the MLP performance tends to decrease when the TSIR (Training Signal to Interference Ratio) decreases to very low values, but it is more robust when the correlation characteristics of target and clutter are varied. Key-Words: Neyman-Pearson detector, gaussian interference, radar detection, neural network, Importance Sampling, Levenberg-Marquardt