A Multi-Scale STFT Diffusion Model for Denoising UWA-OFDM Signals

Underwater acoustic (UWA) communications are constrained by inherently low signal-to-noise ratios (SNRs), which severely degrade the quality of the received signal. The diffusion model (DM) has a natural fit with such scenarios due to its inherent advantages in denoising. In this letter, a multi-scale short-time Fourier transform (STFT)-based diffusion model (MSDM) is proposed for recovering UWA orthogonal frequency division multiplexing (UWA-OFDM) signals. The model utilizes multi-scale STFT of the signal as network inputs, overcoming the limited representation ability of only time domain or frequency domain, and improving the feature representation. In addition, to mitigate the phase distortions introduced by the DM, a loss function adapted to UWA communications is designed, enhancing the robustness of signal reconstruction. Simulation results demonstrate that the proposed MSDM effectively improves the performance of subsequent communication modules and significantly raises the performance limit of traditional algorithms under the same conditions.