Adaptive Semantic Rate and Modulation for Digital Semantic Communication Using Mixture of Experts
IEEE Communications Letters · 2026 · 30 · 2785-2789
DOI: 10.1109/lcomm.2026.3718318 ↗Kurz gefasst
Fachartikel in der Zeitschrift IEEE Communications Letters, 2026. Verfasst von Duc Trong Minh Hoang und 1 weiteren Autor:innen. Die Arbeit nennt 13 Literaturangaben.
Abstract (Originalsprache)
In digital semantic communication (DSC), where both semantic representation and wireless transmission share a finite digital interface, the transmitter must jointly decide how much semantic information to convey and how to efficiently deliver it over the wireless channel. In this letter, we propose SARM-DSC, a novel DSC transceiver that jointly performs semantic rate control and adaptive modulation within a unified end-to-end trainable framework. The proposed design first constructs a progressive finite-alphabet semantic payload to regulate representation quality and then adaptively selects the modulation mode based on payload size and channel conditions. Experimental results on text reconstruction demonstrate that the proposed adaptive framework outperforms state-of-the-art baselines across the tested SNR range. Specifically, the proposed design effectively balance the token accuracy and number of transmitted symbols in both low and high SNR regimes.
Zitieren
APA
Hoang, D. T. M., & Le, L. B. (2026). Adaptive Semantic Rate and Modulation for Digital Semantic Communication Using Mixture of Experts. IEEE Communications Letters, 30, 2785-2789. https://doi.org/10.1109/lcomm.2026.3718318
Vancouver
Hoang DTM, Le LB. Adaptive Semantic Rate and Modulation for Digital Semantic Communication Using Mixture of Experts. IEEE Communications Letters. 2026;30:2785-2789. doi:10.1109/lcomm.2026.3718318
BibTeX
@article{Hoang2026,
author = {Hoang, Duc Trong Minh and Le, Long Bao},
title = {Adaptive Semantic Rate and Modulation for Digital Semantic Communication Using Mixture of Experts},
journal = {IEEE Communications Letters},
year = {2026},
volume = {30},
pages = {2785--2789},
doi = {10.1109/lcomm.2026.3718318}
}
EUVIMED