Published in Engineering, a research paper proposes an explicit semantic base (Seb)-empowered semantic communication (SemCom) system designed to address key limitations of existing semantic communication technologies for sixth-generation (6G) mobile networks. Developed by a research team from Beijing University of Posts and Telecommunications and the University of Houston, the framework prioritizes interpretability and compatibility with current digital communication systems, while enhancing transmission efficiency and robustness for semantic information delivery.

Traditional SemCom approaches rely on end-to-end trained neural networks (NNs) to build implicit knowledge bases (KBs), which lack interpretability and require full retraining when channel conditions or message knowledge deviate from training data. Additionally, most NN-based SemCom systems assume joint training of application and physical layers, which is incompatible with existing communication protocol stacks. To solve these issues, the research introduces Sebs as the fundamental units for representing semantic connotations, establishing a mathematical model for Sebs to construct an explicit KB with a hierarchical, partially ordered structure. Seb granularity can be adjusted based on communication intent and transmission resources, with fine-grained Sebs for detailed semantic representation and coarse-grained Sebs for abstract, resource-efficient transmission, and Sebs also support multimodal data alignment by building a shared feature space.

The proposed Seb-based SemCom architecture operates in two modes: communication and KB update. The KB update mode is triggered by user equipment (UE) requests, enabling the system to add new Sebs, adjust Seb importance based on age of information, and remove obsolete Sebs—all without retraining NNs, thus realizing intelligent evolution of the communication system. The framework integrates a semantic encoder/decoder (Sem-codec) and channel codec, with strategically implemented unequal error protection (UEP) at two levels: Seb-wise and message-wise. UEP allocates different channel coding strategies based on the importance of Sebs and messages, minimizing errors for critical semantic information while allowing a degree of error tolerance for less important content.

To ensure practical deployment, the research designs a Seb-based SemCom protocol stack compatible with the 3rd Generation Partnership Project (3GPP) 5G protocol stack, introducing a semantic intelligence (SI) plane that integrates the explicit KB to connect the application and physical layers. This plane provides Seb and importance information to guide quality-of-service mapping, scheduling, and channel coding in existing protocol layers. A case study focused on image transmission validates the system’s performance, showing that the Seb-based SemCom outperforms state-of-the-art methods in learned perceptual image patch similarity (LPIPS) by over 20% under varying communication intents. The system also exhibits strong robustness under fluctuating channel conditions, and KB updates enable adaptive performance maintenance when communication intents change. Further experiments reveal that Seb granularity can be adaptively adjusted to balance communication efficiency and task performance, with fine-grained Sebs improving perceptual quality and classification accuracy at the cost of higher bandwidth consumption, while coarse-grained Sebs optimize resource usage for abstract semantic tasks.

This research establishes a flexible, interpretable SemCom framework for 6G, bridging the gap between semantic communication theory and practical implementation by aligning with current communication infrastructure, and provides a scalable solution for efficient semantic transmission in emerging 6G applications such as extended reality and digital twins.

The paper “Explicit Semantic-Base-Empowered Communications for 6G Mobile Networks,” is authored by Fengyu Wang, Yuan Zheng, Wenjun Xu, Junxiao Liang, Ping Zhang, Zhu Han. Full text of the open access paper: https://doi.org/10.1016/j.eng.2025.08.039. For more information about Engineering, visit the website at https://www.sciencedirect.com/journal/engineering.