Perceptual Graph-Theoretic Semantic Communication for Highly Constrained Bandwidth Multimedia Transmission: A Hybrid Generative Approach
IEEE Access · 2026 · 14 · 110901-110918
DOI: 10.1109/access.2026.3714499 ↗Kurz gefasst
Fachartikel in der Zeitschrift IEEE Access, 2026. Verfasst von Disha Daniel und 3 weiteren Autor:innen. Die Arbeit nennt 21 Literaturangaben. Der Volltext ist frei zugänglich (Open Access, CC BY 4.0).
Abstract (Originalsprache)
The upcoming Sixth-Generation (6G) age will create a completely connected world where the amount of High-Definition (HD) multimedia traffic that various applications like Metaverse, Extended Reality (XR), and automated surveillance create will vastly exceed the limits of the Shannon-Nyquist framework. Standard communication systems focus on perfect bit reconstruction and generally waste valuable spectrum space to transmit semantically meaningless information, such as noise or redundancy caused by the background. Instead, Semantic Communication (SemCom) allows for a radically new way to communicate by focusing on transmitting the meaning or utility of data, as opposed to transmitting the actual physical copy of that same information. Traditional architectures for SemCom rely heavily on exhaustive end-to-end deep neural networks, which consume significant computational resources, are opaque closed boxes because they are theoretically unexplainable, and will not generalize well when they are deployed outside of the training datasets. This paper presents an innovative, mathematically explainable, and lightweight framework that combines Spectral Graph Theory with Generative Artificial Intelligence (AI). A weighted undirected graph formed from the Perceptual-Saliency Hierarchical Graph (PSHG) algorithm, which utilizes the Commission Internationale de l’Éclairage $L^{*}a^{*}b^{*}$ (CIELAB) color space, is used to create a topology governed by the saliency of the scene depicted. This allows for transmitting these maps independently over both an ideal Additive White Gaussian Noise (AWGN) channel and a harsh Rayleigh fading channel by means of 5G Polar Codes to ensure the highest levels of reliability even at very low signal-to-noise ratios (SNRs). Furthermore, a hybrid reconstruction module has been developed that receives the reconstructed semantic maps from the AWGN channel transmission and uses them as control inputs to an existing neural ControlNet, based on Latent Diffusion Models (LDMs), to create photo-realistic images according to text prompts provided by the transmitter. Through evaluations of the proposed framework, it was determined that a Compression Ratio (CR) up to 10 times greater than both H.264 and H.265 standards can be attained, while achieving robust semantic recovery and providing visual quality at an $E_{b}/N_{0}$ level of 2.5 dB. Moreover, this study lays the groundwork for next-generation semantic transmissions that utilize combinations of classical spectral clustering techniques, improved forward error correction (FEC) codes, and AI-enhanced generative techniques.
Zitieren
APA
Daniel, D., Ali, A. M., Kumar, A. H., & Tamilarasan, I. (2026). Perceptual Graph-Theoretic Semantic Communication for Highly Constrained Bandwidth Multimedia Transmission: A Hybrid Generative Approach. IEEE Access, 14, 110901-110918. https://doi.org/10.1109/access.2026.3714499
Vancouver
Daniel D, Ali AM, Kumar AH, Tamilarasan I. Perceptual Graph-Theoretic Semantic Communication for Highly Constrained Bandwidth Multimedia Transmission: A Hybrid Generative Approach. IEEE Access. 2026;14:110901-110918. doi:10.1109/access.2026.3714499
BibTeX
@article{Daniel2026,
author = {Daniel, Disha and Ali, Abdulla M. and Kumar, A. Harsha and Tamilarasan, Ilavarasan},
title = {Perceptual Graph-Theoretic Semantic Communication for Highly Constrained Bandwidth Multimedia Transmission: A Hybrid Generative Approach},
journal = {IEEE Access},
year = {2026},
volume = {14},
pages = {110901--110918},
doi = {10.1109/access.2026.3714499}
}
EUVIMED