Perceiving Surface Rainfall through Underwater Noise: Toward Environment-Aware Underwater Acoustic Communication

Sea-surface evolution generates intense ambient noise and induces rapid temporal variations in underwater acoustic (UWA) channels, severely degrading the reliability of underwater acoustic communication links. To address this challenge, this paper proposes a rainfall perception model based on the mixture of experts (MoE) architecture with the ResNet backbone for environment-aware underwater acoustic communication, intended to support algorithm selection under varying sea-surface conditions, with the potential to improve communication performance. By exploiting the spectral characteristics of rainfall-induced underwater noise, effective time-frequency representations and frame-length configurations for the proposed model are investigated. Experimental results based on real-world rainfall measurements demonstrate that the proposed model achieves accurate rainfall perception, confirming its suitability as an efficient front-end for environment-aware underwater acoustic communication.