A multi-task learning deep complex network for shallow-water source ranging in the complex environment.

Area of Science:

• Ocean Acoustics
• Signal Processing
• Machine Learning

Background:

• Internal solitary waves (ISWs) in shallow water cause environmental mismatches, degrading underwater acoustic ranging performance.
• Sound speed profile (SSP) mismatches are a primary factor in performance degradation for single-task learning deep complex networks (STL-DCN) and conventional matched-field processing (CMFP).

Purpose of the Study:

• To develop an advanced ranging algorithm capable of mitigating environmental mismatches in shallow water acoustics.
• To enhance the accuracy of underwater acoustic source localization in dynamic environments influenced by ISWs.

Main Methods:

• A multi-task learning deep complex network (MTL-DCN) was proposed, integrating an adaptive weighted multi-task learning mechanism.
• The MTL-DCN simultaneously estimates underwater acoustic source range and the sound speed profile (SSP) along the propagation path.
• The algorithm was validated using simulation analysis and experimental data from the South China Sea.

Main Results:

• Simulation analysis confirmed SSP mismatch as the key factor degrading ranging performance.
• The MTL-DCN demonstrated superior representation and processing of complex acoustic data with phase relationships compared to STL-DCN and CMFP.
• Experimental results showed significantly improved range estimation accuracy using the MTL-DCN.

Conclusions:

• The proposed MTL-DCN algorithm reliably estimates underwater acoustic source ranges in complex, fluctuating environments influenced by ISWs.
• The adaptive weighted multi-task learning approach effectively addresses environmental mismatches, offering a robust solution for shallow water acoustics.