Reduced-rank signal-dependent classification of training data for space-time adaptive processing

Nonhomogeneous and nonstationary training data are two of the key technical challenges facing space-time adaptive processing (STAP) for target detection using space-based radar. We describe a technique for training data selection that seeks to classify the available data in a reduced-rank subspace based on the multistage Wiener filter (MWF). The method builds on previous work by allowing each data sample to associate with multiple covariance classes within the separate subspaces defined by the signal and the class, thereby enabling larger sample support for each class. Performance of the new method is assessed using simulated data based on a monostatic space-based radar scenario. Comparison is made with conventional training data selection techniques