Stochastic system identification approach to radar data processing

Space-time adaptive processing (STAP) algorithms typically consist of a data transformation step to reduce the number of degrees of freedom and a sampling step wherein radar returns from adjacent range bins are used to estimate interference statistics. The reduction in degrees of freedom, inadequate sample support, presence of target in sampled data, and range dependence of interference are some of the main reasons for STAP performance loss. In this paper, we present another approach to target detection and localization that mitigates these performance losses. The approach is based on the well-known stochastic realization algorithm from system identification theory. In this approach, we use the radar return data for a given range bin to identify a minimal stochastic state space model for the return data. The angle and Doppler for all targets in the range bin are computed from the state space matrices. All the computations involved use standard linear algebra. As interference statistics are not directly computed and since there is no sampling from adjacent range bins, the proposed approach is more robust to sample support, target in training and range dependence of clutter. A numerical comparison of the proposed approach with beam-space post-Doppler STAP using simulated data is given.