Performance of adaptive radar in range-heterogeneous clutter

The performance of adaptive radar in range-heterogeneous clutter is a topic of practical interest, since radars are often trained over nonhomogeneous stretches of clutter. In this paper, the performance of eigenanalysis-based Doppler radar is investigated in a nonhomogeneous clutter environment, and is compared to the sample matrix inversion (SMI) and the generalized likelihood ratio (GLR) methods. Two clutter models are considered: edge transition and Weibull range-dependent. It is shown that in the asymptotic case (when the number of samples used to estimate the covariance matrix is large) the location of the clutter edge affects SMI and GLR, but not the eigencanceler. It is further shown that the effect of the nonhomogeneity in the clutter-edge model is to reduce the adaptive processor's training sample support, making the eigencanceler the preferred processor choice. In the Weibull range-dependent model, all processing methods perform poorly when the noise power variance is high. As the variance is reduced, the eigencanceler outperforms SMI and GLR for small training sets, similar to the homogeneous case.