In space‐time adaptive processing (STAP), the required interference covariance matrix is commonly estimated from homogeneous samples. However, the real interference environments are often heterogeneous, which is induced by the fact that the training samples do not share the same interference property with the cell under test (CUT), and thus the interference covariance matrix estimated from the training samples is mismatched with the real one. To improve the performance of STAP in heterogeneous interference environments, this paper proposes a robust training samples selection algorithm which is based on the spectral similarity of the interference. The spectrums of samples are estimated firstly, which are then utilised to calculate the similarity between the interference of the CUT and those of the training samples. Then, the samples whose interference spectrums are similar to that of the CUT are selected as the final training samples, such that the performance of the interference covariance matrix estimated from these samples can be improved significantly. Besides, to avoid the target self‐nulling effect, a simple technique is adopted to exclude the samples which are contaminated by target signal. The proposed method is applied to real radar data, and experimental results demonstrate the effectiveness of the proposed method.

Positioning and synchronization are fundamental problems before multiplatform signal coherence on a potential target, for the purpose of detection, in airborne distributed coherent aperture radars (ADCARs). Hence, in this paper, the identifiability analysis of positioning and synchronization errors, including platform position errors (PPEs), time alignment errors (TAEs), and initial phases (IPs), is investigated using calibrating scatterers separated in disjoint Doppler resolution bins. Firstly, the hybrid Cramér-Rao lower bounds (HCRLBs) of these three perturbations are derived in an assumed observation geometry determined by the platform formation and scatterer placement. Then based on application scenarios, we conduct identifiability analysis using the HCRLB tool in the following three different cases: In the PPE-absent case, with one scatterer, relative errors are identifiable, whereas their absolute ones are not. In the PPE-only case, at least three distributed scatterers exist to identify PPEs except for ill-posed observation geometries. In the ADCAR case, both the two previous conditions are required for the whole identifiability. We assert that relative errors are identifiable because of the extra degree of freedoms (DoFs) brought by different transmit/receive pairs. Moreover, the ill-posed geometries leading to PPEs unidentifiability, from hybrid Fisher information matrix (HFIM) and time difference of arrival (TDOA) perspectives, are also elaborated. Finally, numerical experiments validate our conclusions, and the scatterer-based calibration achieves signal coherence on transmitting and receiving.

In this paper, we consider the problems of off-grid effects elimination and fast implementations for sparse recovery based space-time adaptive processing (SR-STAP) methods. To improve the computational efficiency of recently proposed atomic norm minimization based space-time adaptive processing (ANM-STAP) method, we derive a fast iterative scheme by exploiting the framework of the alternating direction method of multipliers (ADMM), where the unknown parameters are iteratively updated with closed-form expressions. Furthermore, to bypass the selection of regularization parameter in ANM-STAP, we also develop two novel gridless STAP methods by utilizing the covariance fitting criterion (CFC) and the properties of the clutter plus noise matrix (CNCM). Likewise, the corresponding ADMM-based fast implementations are also derived for both CFC-based methods to reduce their computational complexities. Simulation results with both simulated and Mountain-Top data demonstrate that high computational efficiency and good performance of proposed algorithms are achieved.

The traditional space-time adaptive processing (STAP) method based on sparse Bayesian learning (SBL) has the problems of low computational efficiency and slow convergence speed. In this paper, a novel SBL approach based on statistical threshold is proposed to address these issues. To discriminate between the active and inactive atoms of the dictionary, we first develop an adaptive decision test. Next, the adaptive decision test is integrated into the SBL method, which increases its accuracy and convergence rate. Additionally, the detection threshold has an important property that is independent of signal and noise power. Therefore, the adaptive decision test is robust to external changes. Numerous simulations demonstrate that the proposed algorithm has excellent clutter suppression performance and fast convergence rate.