Acoustic-Intensity-Guided Local Grid-Refinement Sparse Bayesian Learning for Broadband Direction-of-Arrival Estimation Using a Single Acoustic Vector Sensor

Broadband direction-of-arrival (DOA) estimation using a single acoustic vector sensor (AVS) is an important problem in passive underwater source localization and underwater acoustic signal processing, especially for compact underwater platforms and passive acoustic monitoring applications. However, conventional grid-based sparse Bayesian learning (SBL) may suffer from grid mismatch when the true bearing lies between adjacent predefined grid points. Although a dense grid can reduce this mismatch, it increases computational cost and dictionary coherence. To address this problem, this paper proposes an acoustic-intensity-guided local grid-refinement SBL method, termed AI-LGR-SBL. The pressure and particle-velocity channels are first used to construct acoustic intensity information and detect candidate source regions. The coarse bearing results then guide target-related spectral peak selection during SBL iterations, and local grid refinement is performed only around the selected directions. Simulations involving single-source and two-source scenarios show that AI-LGR-SBL yields sharper spatial spectra and lower estimation errors than conventional grid-based SBL. Compared with basic SBL, AI-LGR-SBL reduces the RMSE by approximately 10% in the low-SNR region and by 4–7% at relatively high SNRs. Compared with globally dense-grid SBL, it reduces the average runtime by approximately 47.3% and 43.6% in the single-source and equal-power two-source scenarios, respectively. Lake-trial data further demonstrate clear bearing–time trajectories and effective sub-grid peak refinement, supporting the feasibility of the proposed method for broadband underwater DOA estimation and passive source localization using a single AVS.