Multimodal Semantic Communication With Information-Theoretic Disentangled Representations
IEEE Transactions on Wireless Communications · 2026 · 25 · 19432-19447
DOI: 10.1109/twc.2026.3707164 ↗Kurz gefasst
Fachartikel in der Zeitschrift IEEE Transactions on Wireless Communications, 2026. Verfasst von Fan Gao und 3 weiteren Autor:innen. Die Arbeit nennt 50 Literaturangaben. Gefördert von: National Natural Science Foundation of China, National Key Research and Development Program of China, Program of Jiangsu Province.
Abstract (Originalsprache)
As a pivotal technology for intelligent 6G wireless applications, multimodal semantic communication enables efficient transmission by utilizing semantic understanding across heterogeneous data modalities. However, existing end-to-end frameworks fuse heterogeneous modal data into a monolithic representation, rendering task inference fragile under channel impairments and bandwidth allocation inefficient under stringent resource constraints. This paper proposes a Disentangled Multimodal Semantic Communication (DMSC) framework with switchable transmission, which decomposes multimodal semantics into four semantic components: shared, unique, synergistic, and reconstruction semantics, motivated by Partial Information Decomposition (PID) theory. To learn these semantics, we derive information-theoretic surrogate objectives leveraging mutual information lower and upper bounds, and introduce a staged training protocol with orthogonal regularization to encourage discriminative representation learning. Furthermore, a switchable semantic service strategy is developed to adapt dynamically to channel conditions and service requirements. This strategy features a shared-only mode for robust task inference under limited bandwidth or poor channel quality, a full-task mode integrating unique and synergistic semantics for enhanced task utility, and a reconstruction mode for accurate modality recovery. Extensive experiments on multimodal benchmarks demonstrate that the proposed framework achieves consistent improvements over conventional and JSCC baselines under various channel conditions.
Förderung
- National Natural Science Foundation of China
- National Key Research and Development Program of China
- Program of Jiangsu Province
Zitieren
APA
Gao, F., Wang, Y., Qin, Z., & Gao, F. (2026). Multimodal Semantic Communication With Information-Theoretic Disentangled Representations. IEEE Transactions on Wireless Communications, 25, 19432-19447. https://doi.org/10.1109/twc.2026.3707164
Vancouver
Gao F, Wang Y, Qin Z, Gao F. Multimodal Semantic Communication With Information-Theoretic Disentangled Representations. IEEE Transactions on Wireless Communications. 2026;25:19432-19447. doi:10.1109/twc.2026.3707164
BibTeX
@article{Gao2026,
author = {Gao, Fan and Wang, Youzheng and Qin, Zhijin and Gao, Feifei},
title = {Multimodal Semantic Communication With Information-Theoretic Disentangled Representations},
journal = {IEEE Transactions on Wireless Communications},
year = {2026},
volume = {25},
pages = {19432--19447},
doi = {10.1109/twc.2026.3707164}
}
EUVIMED