Multi-frequency sparse Bayesian learning for robust matched field processing.

Area of Science:

• Signal Processing
• Acoustics
• Machine Learning

Background:

• Matched field processing (MFP) is crucial for source localization in complex environments.
• Traditional processors like Bartlett and MVDR face challenges with environmental mismatch and multiple sources.
• Need for robust and computationally efficient adaptive processors in underwater acoustics.

Purpose of the Study:

• To derive and evaluate a multi-snapshot, multi-frequency sparse Bayesian learning (SBL) processor for MFP.
• To compare SBL performance against established processors (Bartlett, MVDR, WNC) under realistic conditions.
• To assess SBL's robustness to environmental mismatch and its ability to handle complex source scenarios.

Main Methods:

• Development of a multi-snapshot, multi-frequency sparse Bayesian learning (SBL) processor.
• Simulation of a two-source scenario with realistic mismatches (array tilt, data/replica vector misalignment).
• Performance comparison using metrics like localization accuracy and robustness against Bartlett, MVDR, and WNC processors.

Main Results:

• SBL demonstrated adaptive behavior, effectively localizing a weaker source amidst a stronger one.
• SBL exhibited significant robustness to environmental mismatch.
• SBL achieved improved localization performance compared to Bartlett, MVDR, and WNC processors.
• SBL automatically determines sparsity, offering an interpretable ambiguity surface.

Conclusions:

• Sparse Bayesian learning (SBL) provides a computationally efficient and high-performance solution for adaptive and robust matched field processing.
• SBL is a practical alternative for applications demanding reliable source localization in challenging underwater environments.
• The SBL processor's ability to handle mismatch and complex source configurations makes it a valuable advancement in signal processing.