Distribution-free radar detection in compound-gaussian clutter

The authors introduce a new class of receivers, which turn out to be distribution-free. At first, they consider the case of completely known target signal which, although unrealistic in radar applications, provides an upper bound to the attainable performance. They show that, once the spherically invariant random processes (SIRP) model is adopted for modelling clutter, the classical Neyman-Pearson test can be approximated, for high number of integrated pulses, by a distribution-free test, which is one and the same independent of the clutter amplitude distribution. Next, they focus on the case of signal with unknown parameters, which has been already considered for K- and Weibull distributed clutter, respectively. The detection structures presented therein are based on the implementation of a generalised likelihood ratio test. They also show that the generalised likelihood ratio converges to one and the same statistic for increasingly high pulse number, regardless the marginal distribution of the compound-Gaussian clutter process. Hence, a distribution-free detector can be conceived, implementing the asymptotically optimum test even for finite sample sizes