GDE: A Generative Diffusion Equalizer for ICSI-Free Semantic Communications in SAGINs
IEEE Transactions on Cognitive Communications and Networking · 2026 · 12 · 9902-9916
DOI: 10.1109/tccn.2026.3709876 ↗Kurz gefasst
Fachartikel in der Zeitschrift IEEE Transactions on Cognitive Communications and Networking, 2026. Verfasst von Huaifeng Shi und 3 weiteren Autor:innen. Die Arbeit nennt 36 Literaturangaben. Gefördert von: National Natural Science Foundation of China, Foundation of Key Laboratory of Intelligent Support Technology for Complex Environments, Ministry of Education.
Abstract (Originalsprache)
In Space-Air-Ground Integrated Networks (SAGINs), high nodal mobility and long propagation delays make instantaneous channel state information (ICSI) prone to aging and estimation mismatch, which poses a major challenge to robust semantic recovery. Existing DeepJSCC-based semantic communication receivers commonly rely on pilot-aided coherent equalization before semantic decoding and are therefore sensitive to ICSI mismatch, especially in dynamic non-terrestrial links. To address this issue, this paper proposes the Generative Diffusion Equalizer (GDE), an ICSI-free receiver framework for semantic communications in SAGINs. Instead of inverting the instantaneous channel, GDE formulates receiver-side semantic recovery as a statistically guided agentic inference process. Specifically, the reverse diffusion process is conditioned on slowly varying statistical channel state information (SCSI), with the Rician factor serving as a representative statistical prompt, rather than on volatile instantaneous channel estimates. An adaptive reinforcement learning policy is further introduced to select the reverse-diffusion starting timestep, thereby controlling the fidelity–latency trade-off during inference. Experimental results show that GDE improves reconstruction robustness under ICSI mismatch, varying Rician fading statistics, and moderate SCSI estimation errors. The results also indicate that GDE leverages Contextual Inpainting in noise-limited regimes and Manifold Projection in mismatch-limited high-SNR regimes to mitigate error floors caused by imperfect channel estimation. Compared with DeepJSCC baselines under imperfect ICSI, GDE achieves an approximately 50% bandwidth reduction under moderate ICSI mismatch while maintaining comparable reconstruction quality.
Förderung
- National Natural Science Foundation of China
- Foundation of Key Laboratory of Intelligent Support Technology for Complex Environments, Ministry of Education
Zitieren
APA
Shi, H., Zhou, L., Tian, W., & Liu, C. (2026). GDE: A Generative Diffusion Equalizer for ICSI-Free Semantic Communications in SAGINs. IEEE Transactions on Cognitive Communications and Networking, 12, 9902-9916. https://doi.org/10.1109/tccn.2026.3709876
Vancouver
Shi H, Zhou L, Tian W, Liu C. GDE: A Generative Diffusion Equalizer for ICSI-Free Semantic Communications in SAGINs. IEEE Transactions on Cognitive Communications and Networking. 2026;12:9902-9916. doi:10.1109/tccn.2026.3709876
BibTeX
@article{Shi2026,
author = {Shi, Huaifeng and Zhou, Long and Tian, Wen and Liu, Chaofan},
title = {GDE: A Generative Diffusion Equalizer for ICSI-Free Semantic Communications in SAGINs},
journal = {IEEE Transactions on Cognitive Communications and Networking},
year = {2026},
volume = {12},
pages = {9902--9916},
doi = {10.1109/tccn.2026.3709876}
}
EUVIMED