EUVIMEDEUVIMED entdecken
Crossref

DGSemCom: Digital Generative Semantic Communications via Discrete Denoising Diffusion Model for Latent Error Correction

Junxiao Liang, Wenjun Xu, Long Liu, Fengyu Wang, Xiaodong Xu, Jiejie Guo, Ping Zhang

IEEE Transactions on Wireless Communications · 2026 · 25 · 19482-19497

DOI: 10.1109/twc.2026.3709956 ↗

Kurz gefasst

Fachartikel in der Zeitschrift IEEE Transactions on Wireless Communications, 2026. Verfasst von Junxiao Liang und 6 weiteren Autor:innen. Die Arbeit nennt 43 Literaturangaben. Gefördert von: National Natural Science Foundation of China, Fundamental Research Funds for the Central Universities, Fundamental Research Funds for the Pengcheng Laboratory.

AbstractBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

The hardware incompatibility of continuous-valued semantic representations and the vulnerability of discrete semantic transmissions to channel noise are key bottlenecks in semantic communications (SemComs). This paper proposes a digital generative SemCom (DGSemCom) framework, which incorporates product quantization (PQ) for efficient discrete feature representation and a discrete denoising diffusion model for latent error correction. First, to achieve efficient discrete semantic representation, PQ decomposes high-dimensional semantic space into independent low-dimensional subspaces, each discretized by a compact codebook. Second, to mitigate the vulnerability of discrete semantic transmission, DGSemCom reformulates error correction as a posterior decoding process by decoupling semantic priors and channel likelihood modeling. Specifically, a discrete denoising diffusion model is adopted to model semantic priors, and the posterior sampling process is guided by received corrupted discrete semantic representations as constraints. Meanwhile, channel likelihoods are dynamically adapted to real-time channel states, effectively avoiding the overfitting problems of existing error correction schemes and enhancing generalization across diverse channels. Extensive experiments demonstrate that DGSemCom achieves superior performance compared to baselines, with a 2 dB higher peak signal-to-noise ratio (PSNR) and a 36.1% reduction in Learned Perceptual Image Patch Similarity (LPIPS) in terms of representation capability, along with a gain of over 2 dB in noise robustness.

Förderung

  • National Natural Science Foundation of China
  • Fundamental Research Funds for the Central Universities
  • Fundamental Research Funds for the Pengcheng Laboratory

Zitieren

APA

Liang, J., Xu, W., Liu, L., Wang, F., Xu, X., Guo, J., & Zhang, P. (2026). DGSemCom: Digital Generative Semantic Communications via Discrete Denoising Diffusion Model for Latent Error Correction. IEEE Transactions on Wireless Communications, 25, 19482-19497. https://doi.org/10.1109/twc.2026.3709956

Vancouver

Liang J, Xu W, Liu L, Wang F, Xu X, Guo J, et al. DGSemCom: Digital Generative Semantic Communications via Discrete Denoising Diffusion Model for Latent Error Correction. IEEE Transactions on Wireless Communications. 2026;25:19482-19497. doi:10.1109/twc.2026.3709956

BibTeX
@article{Liang2026,
  author = {Liang, Junxiao and Xu, Wenjun and Liu, Long and Wang, Fengyu and Xu, Xiaodong and Guo, Jiejie and Zhang, Ping},
  title = {DGSemCom: Digital Generative Semantic Communications via Discrete Denoising Diffusion Model for Latent Error Correction},
  journal = {IEEE Transactions on Wireless Communications},
  year = {2026},
  volume = {25},
  pages = {19482--19497},
  doi = {10.1109/twc.2026.3709956}
}
Kostenlos starten →