Adaptive detection of multichannel signals without training data

Abstract

Adaptive detection of multichannel signals in Gaussian background is studied in this paper, for the case where the existence of training data is not assumed. Four new detectors are designed for this detection problem, by using the generalized likelihood ratio test, Rao test, Wald test and a reduced-dimension (RD) approach; their probabilities of false alarm (PFAs) and detection (PDs) are analytically deduced. These PFAs indicate that the four new detectors possess the constant false alarm rate properties against the noise covariance matrix. Experimental results show that the RD-based detector achieves larger (smaller) PDs than the other three new detectors if limited (sufficient) test data are available. When mismatched signals are encountered, the RD- and Rao-based detectors perform more robust and more sensitive, respectively, than the other two new detectors.