SAGEComm: Semantic-Aware Generation and Encoding for Next-Gen Communication

Publication Date

3-23-2026

Document Type

Article

Publication Title

IEEE Transactions on Vehicular Technology

DOI

10.1109/TVT.2026.3676671

Abstract

Vehicular technology encompasses the electronic systems, communication protocols, and control mechanisms that enable modern transportation systems. In this context, we propose SAGEComm-Semantic-Aware Generation and Encoding for Communication-a next-generation semantic communication framework that integrates Large Language Models (LLMs), Retrieval-Augmented Generation (RAG), and channel-aware vector compression. To overcome limitations of prior approaches, including inflexible compression, static retrieval, and the inability to control semantic-cost trade-offs, SAGEComm adopts a principled, modular design. A hybrid optimization algorithm jointly adjusts the semantic quantizer, channel-aware retrieval module, and generative decoder, enabling robust adaptation to dynamic channel conditions (e.g., signal-to-noise ratio, delay, fading) while preserving semantic fidelity. Furthermore, a Semantic Transmission Cost (STC) metric is introduced to evaluate the tradeoffs among semantic fidelity, compression efficiency, and total processing latency. Simulation results indicate that SAGEComm consistently surpasses existing semantic communication systems in semantic fidelity, compression performance, and end-to-end latency. This work underscores the critical role of integrating deep language understanding with physical-layer adaptation to achieve reliable, efficient, and intelligent communications for 6G vehicular networks and beyond.

Funding Number

ECCS-2302469

Funding Sponsor

Japan Science and Technology Agency

Keywords

Large Language Models (LLMs), Retrieval-Augmented Generation (RAG), Semantic communication, Semantic compression, Semantic Transmission Cost (STC)

Department

Applied Data Science

Link to Full Text

Share

COinS