A novel KA-STAP method based on Mahalanobis distance metric learning

Abstract

The estimation accuracy of the clutter covariance matrix will be degraded when using the heterogeneous and contaminated training samples. To improve the performance of clutter suppression in heterogeneous environments, a novel knowledge aided space-time adaptive processing method is proposed in this paper, which is based on Mahalanobis distance metric learning (MML-KA STAP). Exploiting the difference of the Mahalanobis distances between the actual echo data and the synthesized data, the data is divided into the pure part and non-pure part. Taking the pure data as training samples, the interference covariance matrix of the cell under test (CUT) is reconstructed. Finally, by learning the Mahalanobis distance function between the CUT and training samples, the proper training samples can be chosen and reweighted to construct the interference covariance matrix. When the termination condition of iteration is satisfied, a relatively accurate interference covariance matrix can be estimated. The MML-KA method avoids the effect of target self-nulling and has superior performance in heterogeneous environments. Using the measured data, numerical simulations validate the effectiveness of the proposed method.