Frequency Analysis of Acoustic Data Using Multiple-Measurement Sparse Bayesian Learning.

Area of Science:

• Acoustics
• Signal Processing
• Machine Learning

Background:

• Passive sonar systems detect marine objects via acoustic signals, requiring precise frequency analysis for target identification.
• Existing methods like FFT, ESPRIT, and RMUSIC have limitations in resolution, parameter estimation, and performance under noisy conditions.

Purpose of the Study:

• To introduce and evaluate Sparse Bayesian Learning (SBL) for high-resolution frequency analysis in passive sonar.
• To enhance SBL's robustness by utilizing multi-domain (time and space) measurements for improved target detection.

Main Methods:

• Formulating frequency analysis as a linear system and applying Sparse Bayesian Learning (SBL) for sparse solution reconstruction.
• Utilizing multiple measurements across time and space domains to improve SBL's robustness and noise reduction capabilities.
• Comparing SBL performance against FFT, ESPRIT, and RMUSIC using synthetic and in-situ data.

Main Results:

• SBL demonstrates superior performance with lower estimation errors and higher recovery ratios compared to FFT, ESPRIT, and RMUSIC on synthetic data.
• SBL-based frequency components proved most effective in in-situ data analysis.
• Multi-domain measurements significantly enhance SBL estimation by preserving consistent signal frequencies while attenuating random noise.

Conclusions:

• Sparse Bayesian Learning (SBL) provides a high-resolution and robust method for frequency analysis in passive sonar.
• The integration of multi-domain measurements substantially improves SBL's effectiveness in marine target detection.
• SBL represents a significant advancement over conventional algorithms for acoustic signal processing in challenging environments.