Three Variants of an Outlier Removal Algorithm for Radar STAP

For space-time adaptive processing (STAP), we present three variants of an outlier removal algorithm, the self-censoring reiterative fast maximum likelihood (SCRFML). Specifically, these three SCRFML variants are implemented for three STAP methods, i.e., normalized adaptive matched filter (NAMF), normalized parametric adaptive matched filter (NPAMF), and low-rank normalized adaptive matched filter (LRNAMF). We demonstrated that the SCRFML algorithm can systematically reduce the number of dominant eigenvalues after successive regularization have been applied to the sample/parametric covariance matrix estimate. Then, we show that the model order for NPAMF can be determined from the simulation of cutoff signal-to-noise ratio versus sample support. After the application of outlier removal variants, we compare the detection performance and computational costs of these STAP methods