Multiple source localization using learning-based sparse estimation in deep ocean.

Area of Science:

• Acoustics
• Signal Processing
• Machine Learning

Background:

• Underwater acoustic sensing presents challenges for accurate source localization.
• Estimating multiple sources without prior knowledge of their number is complex.
• Existing methods like Bartlett and feedforward neural networks have limitations.

Purpose of the Study:

• To introduce and evaluate a gated feedback gated recurrent unit network (GFGRU) for multiple underwater source localization.
• To assess GFGRU's performance against established algorithms in deep ocean environments.
• To demonstrate GFGRU's robustness to array tilt mismatch and its applicability to real-world data.

Main Methods:

• Developed a learning-based sparse estimation algorithm, GFGRU, utilizing sound field replicas.
• Trained GFGRU on simulated acoustic data generated by a propagation model.
• Compared GFGRU performance with Bartlett processor, feedforward neural network (FNN), and sparse Bayesian Learning (SBL).

Main Results:

• GFGRU demonstrated performance comparable to the Sparse Bayesian Learning (SBL) algorithm.
• GFGRU offered modest localization improvements over Bartlett and FNN, especially with array tilt mismatch.
• Real-world data validated the robustness of GFGRU in deep ocean conditions.

Conclusions:

• GFGRU is a viable and robust learning-based algorithm for multiple source localization in deep ocean acoustics.
• The proposed method provides a valuable alternative for underwater acoustic sensing applications.
• GFGRU's ability to handle unknown source numbers and array imperfections is a key advantage.