metadata
license: mit
tags:
- underwater-acoustic
- channel-estimation
- denoising
- deep-learning
DSS-Net Checkpoints
Pre-trained model checkpoints for DSS-Net: Dynamic-Static Separation Networks for UWA Channel Denoising.
Available Models
| Model | File | Size | Description |
|---|---|---|---|
| DSS-Net (Full) | dss_net_full_best.pth |
499MB | Best performing model (NMSE: -25.27 dB) |
| Baseline U-Net | baseline_unet_best.pth |
355MB | Baseline for comparison (NMSE: -20.41 dB) |
Usage
import torch
from model import UNetDecomposer
# Load model
model = UNetDecomposer(
in_channels=2,
base_channels=64,
depth=4,
use_attention=True
)
# Load weights
checkpoint = torch.load('dss_net_full_best.pth', map_location='cpu')
model.load_state_dict(checkpoint['model_state_dict'])
model.eval()
Citation
@article{yang2025dssnet,
title={DSS-Net: Dynamic--Static Separation Networks for Physics-Inspired UWA Channel Denoising},
author={Yang, Xiaoyu and Chen, Yinda and Tong, Feng and Zhou, Yuehai},
journal={IEEE Transactions on Wireless Communications},
year={2025}
}
Links
- GitHub: https://github.com/ydchen0806/dss_net
- Paper: IEEE TWC 2025