Semantic Communication for Machines: A Joint Transmission and Understanding Design

Le résumé fourni par la source

Semantic communication excels in multimedia transmission and joint transmission-execution tasks, yet lacks universal architectures for semantic-level device control. This paper introduces an innovative framework enabling direct human/machine-to-device interaction via natural language representation of universal semantics. By eliminating application-layer protocols and predefined controls, our approach facilitates encounter-based control in dynamic environments. To overcome comprehension limitations in lightweight receivers, we develop a joint semantic transmission-understanding framework employing a generative low-complexity architecture that synthesizes slot values directly from slot names. This framework is enhanced by hybrid attention mechanisms and mixed-loss training, with SNR-aware channel adaptation for time-varying conditions. Furthermore, we contribute the first multi-domain machine-to-machine instruction dataset for drone inspections (6k+ annotations), enabling protocol-free interoperability evaluation. Extensive experiments demonstrate the efficacy of the proposed models and skills in one-to-one and one-to-many scenarios, with an emphasis on robustness against channel noise and a sharp reduction in receiver complexity.

Ce résumé expose les affirmations des auteurs. BNTIC ne l’interprète pas comme une validation indépendante des résultats.