Some preliminary experiments with distribution-independent EVT-CFAR based on recorded radar data

The statistics of the radar clutter are often characterized by heavy-tailed distributions. Usually in CFAR design it is assumed that the clutter distribution belongs to some parameterized class of probability distributions. Contrary to that, according to the extreme value theory (EVT), the tail distribution of virtually any distribution can be uniquely modeled by the generalized Pareto distribution. Based on this property it is possible to approximate the tail of the radar clutter distribution and design a CFAR detector that would be distribution-independent. The parameters of the generalized Pareto distribution can be efficiently estimated using the method of L-moments. In this paper, the performance of the EVT-CFAR is compared to that of conventional cell-averaging CA-CFAR for recorded radar data with Gaussian quadrature clutter components and artificial K-distributed clutter intensity.