Semantic-Based Resource Management Based on D2D Multicast Content Delivery: A Game-Theoretic Approach

Zhi Ji, Xinrong Guan, Jie Liu, Xinxin Shen, Meng Wang

Abstract: Device-to-Device (D2D) multicast communication is an important component of wireless network design, bringing greater flexibility and efficiency to 5G and future network architectures. Utilizing semantic communication for D2D multicast is considered a promising approach. However, two challenges need to be addressed: user clustering and resource management based on semantic communication. For the first issue of user clustering, we introduce a semantic triple-based structure to extract semantic features of images and define user similarity for image delivery tasks based on semantic features and other indicators. Subsequently, improvements are made to the K-medoids algorithm to achieve efficient user clustering. For the other issue of resource management, a scaling compression factor is first designed to adjust the fidelity of image delivery for the semantic image communication process of semantic encoding and decoding. Then, we propose a mini-batch model training method by randomly selecting batches to train sub-semantic models, balancing model performance and training complexity. Secondly, we establish an optimization problem to maximize user quality of experience (QoE) through optimizing semantic compression ratios and D2D multicast channel selection. We analyze it from a game theory perspective, modelling the maximization of average QoE for all cell users as a potential game. Finally, we design a joint semantic compression ratio selection and channel allocation strategy based on the spatial adaptive (JSCSA) algorithm to achieve Nash Equilibrium (NE) and demonstrate the convergence of the algorithm. Simulation results confirm the superiority of the proposed algorithm.

External IDs:dblp:journals/tvt/JiGLSW25