Gridless super-resolution sparse recovery for non-sidelooking STAP using reweighted atomic norm minimization

A novel sparse recovery‐based space‐time adaptive processing algorithm based on gridless sparse Bayesian learning for non‐sidelooking airborne radar

“…, that is, the spatialtemporal frequencies are not restricted to fall on discretised grids. Thus, following the idea of sparse representation, the atomic norm of X c is defined as [29]…”

“…Motived by the burgeoning development of compressed sensing techniques, sparse recovery theory was first applied to estimate a high-resolution clutter spectrum by Maria in 2006 [16], demonstrating great potential for tackling the sample shortage problem. Since then, numerous sparse recovery-based STAP (SR-STAP) algorithms have been developed [17][18][19][20][21][22][23][24][25][26][27][28][29], exploiting the intrinsic sparsity of the clutter spectrum. Studies have shown that SR-STAP algorithms can outperform traditional statistical STAP methods with limited training samples (even with a single sample) in ideal cases [26,27].…”

“…Studies have shown that SR-STAP algorithms can outperform traditional statistical STAP methods with limited training samples (even with a single sample) in ideal cases [26,27]. Depending on the adopted sparse signal model, SR-STAP can be further classified into two categories, namely, grid-based methods [17][18][19][20][21][22][23][24][25][26][27] and gridless methods [28,29].…”

“…Then, the ANM method was extended to the 2-D domain for the SR-STAP problem by Feng et al [28]. An improved reweighted ANM (RAM)-based STAP method was proposed as a smooth surrogate for the atomic l 0 norm minimisation problem [29], which enhances sparsity and resolution. Li et al [31] proposed a gridless mixed-norm minimisation (MNM) STAP algorithm and its fast implementation utilising the alternating direction method of multipliers (ADMM) framework.…”

“…Most existing grid‐based SR‐STAP methods critically rely on the on‐grid assumption that the clutter space‐time support frequencies exactly reside on the predefined grids [20]. However, the distribution of actual clutter ridge is affected by the parameters of the radar system as well as its array geometrical configuration [29]. In particular, for non‐sidelooking arrays, the clutter ridge is a curve on the two‐dimensional plane, and thus the on‐grid assumption often fails to hold, resulting in basis mismatch or off‐grid effect.…”