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

This study proposes a ranging algorithm based on a multi-task learning deep complex network (MTL-DCN) to address the environmental mismatch caused by internal solitary waves (ISWs) in shallow water. Simulation analysis demonstrates that the mismatch of sound speed profiles (SSPs) constitutes the primary factor leading to the degradation of the ranging performance of both the single-task learning deep complex network (STL-DCN) and conventional matched-field processing (CMFP). Therefore, an adaptive weighted multi-task learning mechanism is introduced to simultaneously estimate the source range and the SSP along the sound propagation path. Experimental data from the South China Sea indicate that the MTL-DCN more effectively represents and processes acoustic data with complex phase relationships. In comparison to the STL-DCN and the CMFP, the MTL-DCN achieves superior range estimation performance. Within the spatiotemporal fluctuation environment influenced by ISWs, the proposed method reliably estimates the ranges of the underwater acoustic source.