Model Division-Based SeComm: Energy Consumption Minimization by Exploring Semantic Information Similarities

In semantic communication (SeComm), semantic information (SI) extraction plays a vital part. However, existing research on SeComm did not delve into the inherent similarities of SI, resulting in more energy consumption when transmitting SI from different users. To address this issue, we propose a model division-based SeComm framework. In the framework, the BS divides the users into groups by the similarity of their SI after SI extraction. Then the SI of the users in the same user group is divided into shared and personalized information, which is the process of model division performed by the BS. To minimize energy consumption, we formulated an optimization problem where transmission power, bandwidth, SI extraction ratio, and shared information partition are jointly considered. Since the objective function and the constraints consist of separate terms, we propose an alternative optimization algorithm to solve the problem efficiently. The simulation results show that the proposed algorithm can achieve a minimum energy consumption reduction of 15% under bandwidth-limited conditions, and a minimum energy consumption reduction of 24% under total transmit power-limited conditions compared with reference themes. Moreover, the additional simulation results indicate that the proposed framework exhibits lower latency, making it promising in delay-sensitive communication scenarios.