Improved Linear Prediction-based DOA Estimation for Acoustic Vector Sensor Array

To mitigate the challenge of degraded direction of arrival (DOA) estimation accuracy in spatially constrained platforms, an improved linear prediction (IMLP) method for acoustic vector sensor array (AVSA) is proposed in this paper. To extend the aperture of AVSA beyond its physical constraints, the multi-component sensing capability of the AVSA is first used to predict virtual AVS data through linear least squares (LLS). Then, based on the actual data and predicted virtual data of AVSA, a prediction coefficient matrix is dynamically updated through an iterative refinement process to suppress the accumulation of prediction errors. Furthermore, eigenvalue threshold truncation is applied to regularize the covariance matrix, enhancing its condition number and numerical stability. Then, by removing signal components via eigenanalysis, a robust covariance matrix is reconstructed, which reduces estimation errors and enables accurate steering vector correction by isolating noise subspace characteristics. Finally, DOA estimation is performed using the minimum variance distortionless response (MVDR) method. Simulation results demonstrate that the proposed IMLP method under both ideal and non-ideal conditions achieves higher DOA estimation accuracy and stronger noise suppression capability than existing techniques for small-aperture AVSA configurations.