EUVIMEDEUVIMED entdecken
Open Access · CC BY 4.0

D-SIC: Energy-Efficient Digital Semantic Image Communication via Large Generative Models

Bilal Khalid, Jaroslaw E. Prilepsky, Sergei K. Turitsyn, Pedro Freire

IEEE Transactions on Cognitive Communications and Networking · 2026 · 12 · 9114-9129

DOI: 10.1109/tccn.2026.3706089 ↗

Kurz gefasst

Fachartikel in der Zeitschrift IEEE Transactions on Cognitive Communications and Networking, 2026. Verfasst von Bilal Khalid und 3 weiteren Autor:innen. Die Arbeit nennt 64 Literaturangaben. Der Volltext ist frei zugänglich (Open Access, CC BY 4.0). Gefördert von: European Union’s Horizon Europe research and innovation programme MSCA-DN NESTOR, UK Research and Innovation, Horizon Europe Guarantee scheme.

AbstractBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

Semantic Communication (SemCom) systems powered by generative AI models have shown great promise in image communication tasks. While early SemCom research focused on analog transmission, recent efforts have shifted towards digital schemes, largely dominated by Joint Source-Channel Coding (JSCC) paradigm. In contrast, we advocate for the traditional Separate Source-Channel Coding (SSCC) scheme in SemCom that leverages advanced deep learning and generative AI models and offers superior modularity, simplified training, and direct compatibility with existing digital hardware. In this paper, we introduce D-SIC, a robust digital Semantic Image Communication (SIC) framework built upon an extensively modified architecture of Stable Cascade. The main Channel-Aware Diffusion Model (CADM) is conditioned on both Channel State Information (CSI) and corrupted image embeddings, allowing it to effectively mitigate channel-induced distortions. To further improve reconstruction fidelity, we introduce a Lightweight Denoising Neural Network (LDNN) at the receiver that provides the CADM with a refined latent embedding as conditioning. The proposed framework allows for the use of adaptive coding and transmission strategies guided by the channel conditions and is capable of reliable image reconstruction without any channel coding in high-SNR regimes in fading wireless channels. Performance evaluations highlight that this approach not only outperforms conventional standards like JPEG2000 and BPG but also surpasses the end-to-end trained analog SemCom systems like DeepJSCC.

Förderung

  • European Union’s Horizon Europe research and innovation programme MSCA-DN NESTOR
  • UK Research and Innovation
  • Horizon Europe Guarantee scheme
  • EPSRC Core Equipment Fund

Zitieren

APA

Khalid, B., Prilepsky, J. E., Turitsyn, S. K., & Freire, P. (2026). D-SIC: Energy-Efficient Digital Semantic Image Communication via Large Generative Models. IEEE Transactions on Cognitive Communications and Networking, 12, 9114-9129. https://doi.org/10.1109/tccn.2026.3706089

Vancouver

Khalid B, Prilepsky JE, Turitsyn SK, Freire P. D-SIC: Energy-Efficient Digital Semantic Image Communication via Large Generative Models. IEEE Transactions on Cognitive Communications and Networking. 2026;12:9114-9129. doi:10.1109/tccn.2026.3706089

BibTeX
@article{Khalid2026,
  author = {Khalid, Bilal and Prilepsky, Jaroslaw E. and Turitsyn, Sergei K. and Freire, Pedro},
  title = {D-SIC: Energy-Efficient Digital Semantic Image Communication via Large Generative Models},
  journal = {IEEE Transactions on Cognitive Communications and Networking},
  year = {2026},
  volume = {12},
  pages = {9114--9129},
  doi = {10.1109/tccn.2026.3706089}
}
Kostenlos starten →