Robust Airborne Radar Clutter Suppression Algorithm via Deep Neural Networks Under Small-Sample Conditions

Abstract

Space-time adaptive processing (STAP) is a powerful technique for clutter suppression in airborne radar systems. However, in practical applications, STAP performance is often compromised by the limited availability of independent and identically distributed (i.i.d.) samples, which restricts its clutter suppression capabilities. Inspired by compressed sensing, researchers have extensively investigated sparse recovery-based STAP (SR-STAP) algorithms, which can achieve near-optimal performance under ideal conditions. Yet, these algorithms experience notable performance degradation in the presence of spatial and temporal errors. To address this issue, we propose a deep neural network framework for clutter suppression, cascading a gridless sparse recovery network (GLSRNet) with a generative adversarial network (GAN), ensuring accurate clutter covariance matrix estimation under small-sample conditions with errors. During GAN training, we observed that the traditional binary cross-entropy (BCE) loss function led to significant oscillations in the training loss, impeding effective convergence. To resolve this, we design a new loss function that achieves stable convergence. Extensive experiments validate the effectiveness of the proposed algorithm in clutter suppression.