MLP-based approximation to the Neyman Pearson detector in a terrestrial passive bistatic radar scenario

In this paper, the design of Neural Network (NN) based solutions for detecting ground targets using passive radar systems exploiting Digital Video Broadcasting transmitters as illuminators of opportunity, is tackled. Real radar data acquired by a technological demonstrator developed at the University of Alcala was used, to determine suitable statistical models of the interference. To exploit the expected NN based detector performance improvement, a novel approach was proposed to define the observation space for the detection problem. Observation vectors composed of complex radar returns belonging to different Coherent Processing Intervals (CPIs) were considered. For CPIs of 250ms, statistical analyses showed that the problem was an example of detection of Swerling II targets in white Gaussian interference. NN based detectors were designed for approximating the Likelihood Ratio detector (Neyman-Pearson solution). Results were a new prove of NN approximation capabilities, which could be exploited in other passive bistatic radar scenarios.