A NOTE ON NON-GAUSSIAN ADAPTIVE ARRAY DETECTION AND SIGNAL PARAMETER-ESTIMATION

Kelly's generalized likelihood ratio test (GLRT) statistic is reexamin ed under a broad class of data distributions known as complex multivar iate elliptically contoured (MEC), which include the complex Gaussian as a special case. We show that, mathematically, Kelly's GLRT test sta tistic is again obtained when the data matrix is assumed MEC distribut ed. The maximum-likelihood (ML) estimate for the signal parameters-ali as the sample-covariance-based (SCB) minimum variance distortionless r esponse beamformer output and, in general, the SCB linearly constraine d minimum variance beamformer output-is like,vise shown to be the same . These results have significant robustness implications to adaptive d etection/estimation/beamforming in non-Gaussian environments.