Covariance matrix estimator in non-Gaussian clutter and the mismatch performance analysis

In the non-Gaussian clutter modeled as a spherically invariant random vector, a covariance matrix estimator is introduced. Moreover the performance of the corresponding adaptive normalized matched filter (ANMF) is assessed in the mismatch case that the estimated clutter group size is unequal to the actual one. The results show that the ANMF holds fully constant false alarm rate (CFAR) property when the actual clutter group size is multiples of the estimated one, whereas it is only CFAR with respect to the clutter covariance matrix structure for other cases. It also shows that, different mismatch cases cause different performance loss for the ANMF, and the worst performance results from the case that no a priori information about the texture correlation is available.