Correction physics-informed neural network-aided matched field processing technique for underwater passive source range estimation

Matched field processing (MFP), a technique that is extensively used for underwater passive source ranging, primarily faces a challenge of replica environmental mismatch. In this paper, a correction physics-informed neural network (CrPINN)-aided MFP (CrPIMFP), a data-efficient and physics-conforming ranging method, is proposed. The CrPINN uses very few measured data to correct replicas generated by acoustic propagation model to mitigate mismatches. During this process, a normalized correction loss is applied to align the replica with the measured field. Additionally, CrPINN integrates the Helmholtz equation and the boundary condition into loss function, thereby enabling interpolation to unsampled points of the replica and decreasing the dependence on labeled data. The discrepancy between the corrected replica generated by the trained CrPINN and the measured field is reduced, which is beneficial for the ranging performance of MFP. Simulation results demonstrate that, compared with MFP, CrPIMFP is more resistant to bottom sediment depth/sound speed mismatches and mildly range-dependent environment in shallow water. The experimental results from the SWellEx-96 environment show that CrPIMFP outperforms some conventional ranging algorithms in four scenarios-few training samples with varying sampling conditions, generalization to unseen environments, few array elements, and great distances-verifying its robustness.