Generating Underwater Acoustic Communication Channel Impulse Responses Using a Diffusion Model

Underwater acoustic communication data is expensive to collect, often yielding experimental data sets that are sufficient for proof-of-concept analysis, but are insufficient for applications that demand orders of magnitude more channel realizations, such as data-driven methods for outage-capacity analysis, network simulation, or statistical regression of new communication algorithms. For such applications, a realistic synthetic dataset is a necessity for the successful generalization of data-driven underwater acoustic (UWA) communication systems. This paper investigates the application of diffusion models for UWA data augmentation for such systems. Diffusion models, as opposed to physics-based models such as BELLHOP or other acoustic propagation tools, extract the essential characteristics of the data without explicit knowledge of environmental parameters. We demonstrate the capability of such models by generating data whose multipath structure and spatiotemporal correlation match those of the Kauai ACOMMS MURI 2011 (KAM11) experiment.