Research On Clutter Model Identification Technology Based On Anderson Darling Fitting Goodness Test

Constant false alarm rate processing is very important in automatic radardetection, and has become one major project in radar signal processing. Differentdistribution of radar clutter corresponds to different optimum CFAR processor, thus,identification of radar clutter model is employed to get better CFAR performanceunder changing environment.Method based on empirical distribution statistic is widely used for theidentification of radar clutter model, such as2, KS, and Anderson-Darling (AD)goodness-of-fit test. This essay is focusing on AD goodness-of-fit test because it isfree from grouping and more practical and effective under small sample sizethan2and KS tests. Several radar clutter models like Rayleigh, Weibull, Lognormaland K distribution are studied in this article, and methods for parameter estimation aregiven firstly as it is one important process in AD test. Moment-based method is usedto estimate parameters of Weibull and K distributions, because method of maximumlikelihood estimation is not convenient to estimate parameters of Weibull withnonlinear equation, and estimation formula is impossible to observe for K distribution.Comparison shows that fractional-based methods with lower moments are moreaccurate. But it is difficult to get accurate estimation for K distribution when theshape parameter generating samples is bigger.The principle of AD test is comparing two degree statistic with the critical value,if the critical value is bigger then assumption can be accepted. It is not benefit forradar target detection with the usual way to generate critical value as it is complex andneeds large amount of time. Thus, we focusing on the study of critical value for fourradar models, and methods to generate tables of critical value under different numberof samples and significant points are also given. Critical tables under completelyknown and partly known parameters have been brought up by researchers, but inpractice, parameters must be estimated from given samples. Thus, changes of criticalvalue for different clutter model under different parameters are given in graphs.Monte Carlo simulation results show that, critical value for Rayleigh and Lognormaldistribution is invariable and it changes a little for Weibull distribution under small shape parameter, and it changes greatly for K distribution. In order to save time forAD test, we choose curve fitting to get asymptotic critical value for Weibull and Kdistributions. Comparisons of AD test with asymptotic critical value and usual methodto generate it are carried out to verify that the way in which using asymptotic criticalvalue is workable.Monte Carlo simulations show that it maybe accepted as two or three cluttermodels after the clutter samples are tested under different assumptions. In order todistinguish them from several decisions and get the right clutter model, we choosemethod of fusion combining parameters estimated from samples and the probabilitiesto get the final decision. Simulations also show that decisions getting from thismethod can observe better performance. Thus, we can choose optimum CFARprocessors corresponding to the clutter model and get accurate decision whether therehas a target or not.