Automated detection in multiple-target environments using the censored mean-level detector

This paper presents performance results for a class of robust, constant-false-alarm-rate (CFAR) detectors known as censored mean-level detectors (CMLD). CMLD have a generalized maximum likelihood detector structure that has been modified to provide robust performance in multiple target environments. In the CMLD, a censoring of selected order statistics is used in the noise power estimate to achieve the desired robustness. In the absence of interfering targets, the performance of the CMLD approaches that of the optimal parametric energy detector as M, the number of noise reference samples per test sample, approaches infinity; but even for M as small as 32, the SNR performance is typically within 0.1 dB of optimum. The two realizations of the CMLD compared in this paper are those in which either the censoring is performed before the time averaging of the noise reference samples (CBA) or the censoring is performed after the time averaging (CAA). The results indicate that censoring up to 75% of the noise reference samples significantly improves the CMLD's robustness against interference while making no significant change in the performance in the absence of interference. In addition, the results indicate that the CAA-CMLD is much more robust and computationally efficient than the CBA-CMLD.