Trimmed-Mean Detector for Noncoherent Distributions

摘要

Sea-clutter measurements using high-resolution radar indicate that clutter cross-section returns do not usually follow a Rayleigh distribution. The log-normal and contaminated-normal descriptions of sea clutter have been considered, and detection curves for these distributions have been generated for both the mean and median detectors - the median being better for the log-normal distribution and the mean being slightly better for the contaminated-normal. In this report, the trimmed-mean detector is proposed as a detector against these clutter distributions. By using a Monte Carlo technique, equations are obtained for the threshold values, and curves of probabilities of detection (P sub d) are drawn for false-alarm probabilities (P sub fa) ranging from 0.01 to 10 to the minus 9th power. When one compares the mean, median, and trimmed-mean detectors, the conclusion is that the trimmed-mean is the better detector against either of the two distributions considered. In fact, for the log-normal distribution, when 10 samples are available, the trimmed-mean requires 5 dB less S/N ratio than the mean in order to obtain P sub d = 0.9 and P sub fa = 10 to the minus 8th power. Unfortunately, whereas the mean and median are easily implemented, the implementation of the trimmed-mean is very difficult because of the necessity of ordering the samples. (Author)