Sparse Bayesian learning based on spatio-temporal structure-aware for matched field processing.

Area of Science:

• Underwater acoustics
• Signal processing
• Bayesian inference

Background:

• Traditional matched field processing (MFP) for underwater acoustic source localization processes each time step independently.
• This approach overlooks the continuous spatial changes of a source over time, potentially limiting localization accuracy.

Purpose of the Study:

• To develop an improved algorithm for underwater acoustic source localization by incorporating spatio-temporal information.
• To enhance localization performance by leveraging the continuous spatial structure of source movement.

Main Methods:

• A sparse Bayesian learning (SBL) algorithm incorporating a spatio-temporal structure-aware prior is proposed.
• A structure prior for sparse coefficients captures spatial continuity between adjacent time steps.
• Variational Bayesian inference (VBI) is used to infer hidden variables within a hierarchical Bayesian framework.
• The method is extended to the multi-frequency case for further performance enhancement.

Main Results:

• The proposed SBL algorithm effectively utilizes prior information on continuous spatial source changes.
• The method automatically selects optimal update strategies based on neighbor information or independent updates.
• Localization accuracy is improved compared to conventional methods.
• The algorithm demonstrates efficacy on both simulated data and the SWellEx-96 experimental dataset.

Conclusions:

• The proposed spatio-temporal structure-aware SBL algorithm offers a robust approach to underwater acoustic source localization.
• It enhances accuracy by integrating temporal continuity without requiring a predefined motion model.
• The method shows significant potential for real-world underwater acoustic applications.