A CFAR processor for the detection of unknown random signals in nonstationary correlated noise

Signal detection in a nonstationary environment has been widely studied in the literature, either in the context of known deterministic signals in unknown correlated noise or in the context of unknown random signals in white noise. However, the detection of unknown random signals in unknown correlated noise has not been adequately described although it is of practical importance. Application of the conventional normalization-plus-integration technique to such situations results in performance degradation and failure to achieve a constant false-alarm rate (CFAR). An efficient and computationally simple CFAR detection procedure is proposed in this paper, based on the comparison of the covariance structure of the data vector under test with that of the interference. The theoretical detection performance of the new test, in a closed form, is presented. It is shown that for interference with moderate and high autocorrelation, the new test provides about 6 to 12 dB improvement over the conventional normalization technique.