Characterization of clutter heterogeneity and estimation of its covariance matrix

Heterogeneous clutter environments are frequently encountered in radar processing and constitute a main source of performance degradation for most adaptive detection schemes which use the sample covariance matrix (SCM). This degradation is mainly due to the fact the SCM is no longer a consistent estimate of the clutter covariance matrix in the cell under test (CUT). In a recent paper we proposed a knowledge-aided Bayesian framework for such heterogeneous environments, and derived the associated minimum mean-square error (MMSE) estimate of the CUT covariance matrix. Both the degree of heterogeneity and the degree of a priori knowledge were adjusted by known scalar variables denoted as nu and mu. In this paper, we extend these results to the more practical case where these scalars are unknown and have to be estimated together with the CUT covariance matrix. We extend the Gibbs sampler strategy of our previous work to this new problem. This allows us in particular to estimate the degree of heterogeneity nu of the clutter, and hence to characterize the environment. We show that the MSE for estimation of the CUT covariance matrix is only marginally increased compared to the case of known nu and mu. The new scheme is also successfully applied to real radar data where we show that it can distinguish between homogeneous and nonhomogeneous areas in a range-azimuth map.