Signal detection in compound-Gaussian noise: Neyman-Pearson versus GLRT

Abstract

We dealt with the detection of signal in presence of compound-Gaussian noise. First of all, we derived the optimum Neyman-Pearson detector for a known signal to establish the limiting performance. Since the NP detector does not implement a uniformly most powerful (UMP) test, we adopt an alternative approach, namely the generalized likelihood ratio test (GLRT): the result is a completely canonical structure. Nevertheless since the implementation of such a processor requires averaging infinitely many projections of the received waveform, we also presented a sub-optimum scheme, retaining most of the characteristics, especially the CFARness, of the ideal detector. The analysis of the above detectors shows that both of them largely outperform the conventional square-law receiver and that the sub-optimum processor is practically equivalent to the GLRT receiver.