Intelligent Semantic Communication System Based on Kolmogorov–Arnold Networks Driven by Dynamic Terminal-Side Computing Power Network

With the advent of the 6G era, the number of IoT devices has experienced explosive growth, leading to the generation of massive amounts of data at the network edge. Semantic communication, as an innovative solution to handling this data deluge, can significantly enhance communication efficiency. However, the limited storage and computational resources of terminal devices constrain the widespread application of semantic communication in 6G networks. To address this issue, we propose a terminal-side-computing-driven intelligent semantic communication solution. Specifically, we introduce a semantic communication model based on Kolmogorov–Arnold Networks (KANs), named K-DeepSC. Using image-reconstruction tasks as an example, the proposed K-DeepSC reduces the number of model parameters by 44% compared to semantic communication models based on Multi-Layer Perceptrons (MLPs), while maintaining similar performance. Furthermore, to fully leverage idle terminal computing power for semantic tasks, we explore computation offloading in dynamic Terminal-Side Computing Power Networks. By optimizing task delay minimization, a deep reinforcement learning algorithm is employed to determine the optimal offloading strategy. Simulation results demonstrate that our proposed solution effectively reduces semantic task processing delay.