Dictionary Self-Tuning Correlated Sparse Bayesian Learning for STAP in EPC-MIMO Radar

Range–Doppler dependence and range ambiguity violate the independent and identically distributed sample assumption in space–time adaptive processing (STAP), reducing the available sample size and folding far-range clutter into near-range clutter. To address these issues, a dictionary self-tuning correlated sparse Bayesian learning STAP (DST-CSBL-STAP) method for colocated airborne element–pulse-coded multiple-input multiple-output (EPC-MIMO) radar is proposed. First, a tailored initial dictionary that exploits structured clutter ridges is designed to reduce the original three-dimensional sparse recovery problem to a limited set of two-dimensional planes. Then, a plug-in dictionary self-tuning (DST) method based on marginal likelihood maximization is developed to refine dominant dictionary atoms under extremely limited sample conditions and thus alleviate off-grid errors. Finally, a spatial channel scanning strategy is presented to form the complete DST-CSBL-STAP method. We further provide a convergence analysis, demonstrating that the DST plug-in serves as a corrective mechanism to prevent the algorithm from trapped in local optima. The execution schedule of DST is adjustable, allowing for a flexible trade-off between computational complexity and convergence rate. Numerical simulations demonstrate the robustness and efficacy of the proposed method in both mild and severely heterogeneous clutter environments.