Underwater Acoustic Propagation Models with Transformers

The underwater acoustic propagation model plays an important role in the field of underwater applications. The traditional underwater acoustic propagation models are mainly numerical models. These methods calculate the sound field by solving differential equations, typically requiring precise underwater environmental parameters. To overcome the shortcomings, the paper proposed an underwater ACoustic propagation model with Transformers (AcT). It uses the transformer structure to learn the potential transfer modes of the sound field propagation instead of solving differential equations. The proposed AcT converts the input field range slice into patches and learns the mapping method of the field data from one range to another. During sound field inference, the AcT takes the current field slice as input and recursively predicts the next range slice. The simulation results demonstrate the effectiveness of the proposed method, the sound fields predicted by AcT are consistent with the numerical model results while preserving the local characteristic structure of the sound field. The method provides the research foundation for low environmental parameter requirements data-driven methods. The application of the AcT shall be studied in the future.