Attention - Enhanced Denosing diffusion probabilistic modles for underwater acoustic signals

The underwater acoustic signal processing gather difficulties due to the maze-like noise environments to
marked by the ambient to interference from ocean currents, sea traffics,and biological source, coupled
with pronounced signal degradation resulting from the multipath propagation and frequency-dependent
reduction. The denoising techniques to conventional on oversimplified noise models and necessitate
considerable human parameter adjustment, their diminishes in dynamic marine settings . to study
introduces an innovative and frame work that integrates denoising diffusion probabilistic
models(DDPM) with self-attention processes .the methods used for a (1-D U-net )architecture that works
on a row time-domin waveform sampled at 16 kHz and 200 step diffusion process used with linear noise
scheduling.this type attention model uses six transformer-style blocks with 4-head scaled dot-product in
attention placed included between the convolutional encoder and decoder stages. The advanced hybrid
model to places attention at deeper U-Net stages and to capture the bottleneck long-range of
dependencies alongside in diffusion-based noise removal.this type of Experiments on synthetic
underwater signals (multi-tone sinusoids at 50–800 Hz with coloured noise at 5 dB SNR) the each model
is showed has distinct strength. The models of attention only can achieved the best in SNR of 8.90 dB
through a deterministic inference.the DDPM with attention model to achieved the best spectral
reconstruction, with matching the clean reference exactly a spectral flatness of 0.000, the both individual
models on frequency-domain fidelity.the only model of DDPM provided the effective noise reduction
but the broadband artefact introduced from its stochastic sampling. These diffusion and attention offer
complementary strengths are show in result , and the combination of them produces the most faithful
spectral recovery, to the framework a practical candidate for making the application in autonomous
underwater vehicles, sonar systems, and marine monitoring.