A simple yet effective statistic is proposed or detecting transients buried in partially unkniwn ambient noise. The transient model is the frequency scattered increased variance observations. Our motivation is to derive a test statistic that can tell whethe or not the whole observation sequence follows a signle distribution whcih corresponds to the ambient only case; in other words, to test the homogenetiy ofthe data set. The statistic we presetn is derived as te likelihood ratio test of overdispersion when the underlying observation sequence folows a double exponential distribution. Some nice properties are given, such as its CFAR ability. Numerical testing focuses on the ocmpaison of this scheme with the popular power-law detector as well as its CFAR extension.
展开▼
