.png){kind=icon}
Return
PODD: Path-Optimized Denoising Diffusion Models for Wireless Semantic Communications
.png){kind=icon}
PRE
AI
DOI：10.1109/tmc.2026.3734492.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
Diffusion models have recently attracted increasing attention in wireless communication systems due to their strong denoising capability and generative modeling ability. However, existing diffusion-based semantic communication schemes usually rely on iterative reverse sampling, which leads to high inference latency and limits their applicability to delay-sensitive wireless scenarios. To address this issue, this paper proposes a path-optimized denoising diffusion model (PODD) with joint source-channel coding (JSCC) for efficient and high-fidelity wireless semantic communications. Specifically, PODD enhances the channel-aware denoising capability by incorporating multi-scale channel state information into the diffusion network, enabling more effective recovery of channel-corrupted semantic features. Furthermore, a graph-based path optimization mechanism is introduced to refine the reverse sampling path, allowing the denoising process to be compressed into only two sampling steps while reducing cumulative propagation errors. Experimental results on multiple datasets demonstrate that the proposed joint PODD and JSCC system achieves superior reconstruction performance to other benchmark systems across multiple evaluation metrics. For example, on the DIV2K dataset under the Rayleigh fading channel with a signal-to-noise ratio (SNR) of 10 dB, PODD improves the peak signal-to-noise ratio (PSNR) by 0.47 dB and 1.06 dB over CDDM and JSCC, respectively. Moreover, PODD significantly reduces the inference time by compressing the sampling process into only two steps, achieving a $\bf {22.14\times }$ speedup compared with CDDM.
Keywords:
Denoising diffusion models
joint source-channel coding
path optimization
wireless semantic communications
Journal
IF:
9.2
Papers:
5.9K
Citations:
1.8W
Organization
View more organizations
Cited Papers
No cited papers available