Locally optimum detection in moving average non-Gaussian noise

Detection algorithms that are locally optimum Bayes, and also asymptotically optimum, are developed for both coherent and incoherent signaling for arbitrary interference and signal waveforms when the dependence in the noise samples is represented by a moving-average model. This leads to receiver structures, which are prewhitened versions of the locally optimum detectors in the independent case. A probability-of-error expression (in the ideal-observer symmetric case), the processing gain, and the minimum-detectable signal are derived in both cases. These demonstrate, by means of an expression comparing performance between this and the independent case, that for the same large sample size (n≫1), an improvement in performance is always achieved when the noise samples are dependent, without any additional complexity in receiver structure