Low Complexity Neural Network Guided Robust Space-Time Adaptive Wideband Beamforming

Wideband beamforming serves as a critical signal processing technology in applications such as wireless communications, radar, and sonar systems. However, conventional tapped delay-line space-time adaptive wideband beamformers (SAWB) exhibit performance degradation due to limitations in snapshots, contamination of the data covariance matrix (DCM) by the signal of interest (SOI), and mismatched steering vectors (SVs). To address these challenges, a DCM reconstruction (DCMR) method for robust SAWB (DRSAWB) is proposed, along with an efficient interference direction prediction network (EJDPN)-guided DRSAWB (ENDRSAWB). The contributions of this work are threefold: First, traditional narrowband covariance-based methods are extended to wideband scenarios, with two distinct reconstruction strategies introduced: one requiring prior knowledge of interference angles and the other operating without such information. Second, EJDPN is utilized to reduce the computational complexity of DCMR. By formulating a quadratically constrained quadratic programming (QCQP) framework, the pointing mismatch of reduced-rank vectors is compensated to optimize SVs. Simulation results validate the robustness and effectiveness of the proposed algorithms in adaptive wideband beamforming applications.