Approximate CFAR signal detection low rank non-Gaussian interference

Abstract

We have derived a new generalized likelihood ratio test for detecting a signal in unknown, strong non-Gaussian low rank interference plus white Gaussian noise. An important advantage of the new test is that it needs essentially no knowledge of the non-Gaussian distribution, except that the coherent part of the interference is or can be approximated as low rank. Using perturbation expansions of the test statistic, we establish the connection of the proposed GLRT detector to the UMPI test and show that it is approximately CFAR. Computer simulations indicate that the new detector significantly outperforms traditional adaptive methods in non-Gaussian interference and is insensitive to signal and rank mismatch.