Robust detection of random signals in exponential mixture noise

Exponential mixture probability density functions (PDFs) are shown to be useful models of the intensity of high resolution low-pulse-rate radar clutter. In this environment, using known parameters, incoherent detection algorithms based upon these noise models have significantly improved performance in comparison with detection algorithms based on exponential PDFs. To implement exponential mixture based detection algorithms, parameters must be estimated from noise only data and applied to the data under test. Certain parameters vary over short range and time segments, and performance is often degraded due to uncertainty in the true parameter values. For the algorithms presented, each parameter is assumed to be known within a certain interval, and valves of the parameters needed by the processor are selected to prevent an excessive number of false alarms. One technique selects certain percentiles for each parameter, and another minimizes the maximum false alarm rate. In addition a high variance state measured globally may be added to the processor. The performance of these algorithms are compared with a CFAR processor using radar data.