Radar signal processing with antenna arrays via maximum likelihood

This paper considers two different models for coherent radar data in which a single target is observed by an array of antennas in the presence of noise, clutter, and jamming. One model is parameterized in terms of the target signal's amplitude and direction of arrival (DOA), while the other uses art unstructured "spatial signature" instead. The maximum likelihood (ML) estimator and generalized likelihood ratio detection test for the unstructured model are derived, and compared with the standard solutions for the structured DOA-based model. Using the extended invariance principle, it is further shown how asymptotically optimal estimates of the target signal's amplitude, DOA, and doppler can be obtained from the unstructured ML parameter estimates. Several advantages of the two-step unstructured approach are noted including reduced computational cost and greater robustness to calibration errors.