OSD-JSCC: Neural Joint Source-Channel Coding with One-Step Diffusion
Abstract
Neural joint source-channel coding (JSCC) enables efficient image transmission over noisy channels, yet distortion-optimized approaches often produce blurry reconstructions. While diffusion priors can enhance visual fidelity, their iterative sampling incurs substantial decoding latency. To address this limitation, we propose OSD-JSCC, an SNR-adaptive JSCC framework with a single diffusion step. Specifically, OSD-JSCC transmits pretrained variational autoencoder (VAE) latents through an SNR-conditioned JSCC codec and incorporates two key designs. First, we introduce a stochastic diffusion-state interface that enables the JSCC decoder to predict a Gaussian diffusion state rather than a clean latent, while extracting image-specific semantic tokens. The sampled diffusion state preserves spatial structure and provides a good initialization for one-step refinement, whereas the semantic tokens, together with an SNR token, guide the reconstruction process. Second, we develop SNR-conditioned feature modulation and adaptive timestep selection to jointly adapt latent transmission and diffusion refinement to varying channel conditions. Furthermore, we adopt a three-stage training strategy with low-rank adaptation (LoRA) to efficiently adapt the pretrained diffusion prior to the JSCC task. Extensive experiments on Kodak and CLIC2020 demonstrate that OSD-JSCC achieves superior perceptual quality and decoding efficiency. Compared with existing diffusion-based JSCC baselines, OSD-JSCC reduces LPIPS by approximately 43% on Kodak and 53% on CLIC2020, while achieving up to a 50.6× reduction in decoding latency.
Related
- Joint Source-Channel Coding of Gaussian Sources over Block Erasure Channels: Nonasymptotic Bounds and Channel-Uniform Normal Approximations
- Secure Joint Source-Channel Coding of Multimodal Semantic Sources
- Domain-Adaptive Deep Joint Source-Channel Coding for Image Classification
- Polar Complexity: A New Descriptive Complexity with Applications to Source and Joint Source-Channel Coding
- Lossy Joint Source-Channel Coding over Unknown Channels
BibTeX
@misc{tang2026osdjscc,
title = {{OSD-JSCC: Neural Joint Source-Channel Coding with One-Step Diffusion}},
author = {Shunpu Tang and Qianqian Yang},
howpublished = {arXiv},
year = {2026},
month = nov,
doi = {10.48550/arxiv.2611.06437},
eprint = {2611.06437},
archivePrefix = {arXiv},
url = {https://arxiv.org/abs/2611.06437},
}
RIS
TY - UNPB
TI - OSD-JSCC: Neural Joint Source-Channel Coding with One-Step Diffusion
AU - Shunpu Tang
AU - Qianqian Yang
PY - 2026
DA - 2026/11/04
PB - arXiv
DO - 10.48550/arxiv.2611.06437
UR - https://arxiv.org/abs/2611.06437
AB - Neural joint source-channel coding (JSCC) enables efficient image transmission over noisy channels, yet distortion-optimized approaches often produce blurry reconstructions. While diffusion priors can enhance visual fidelity, their iterative sampling incurs substantial decoding latency. To address this limitation, we propose OSD-JSCC, an SNR-adaptive JSCC framework with a single diffusion step. Specifically, OSD-JSCC transmits pretrained variational autoencoder (VAE) latents through an SNR-conditioned JSCC codec and incorporates two key designs. First, we introduce a stochastic diffusion-state interface that enables the JSCC decoder to predict a Gaussian diffusion state rather than a clean latent, while extracting image-specific semantic tokens. The sampled diffusion state preserves spatial structure and provides a good initialization for one-step refinement, whereas the semantic tokens, together with an SNR token, guide the reconstruction process. Second, we develop SNR-conditioned feature modulation and adaptive timestep selection to jointly adapt latent transmission and diffusion refinement to varying channel conditions. Furthermore, we adopt a three-stage training strategy with low-rank adaptation (LoRA) to efficiently adapt the pretrained diffusion prior to the JSCC task. Extensive experiments on Kodak and CLIC2020 demonstrate that OSD-JSCC achieves superior perceptual quality and decoding efficiency. Compared with existing diffusion-based JSCC baselines, OSD-JSCC reduces LPIPS by approximately 43% on Kodak and 53% on CLIC2020, while achieving up to a 50.6× reduction in decoding latency.
ER -