Research on Semantic Communication and AI-Driven Cloud-Edge Resource Collaborative Transmission Strategies under the 6G Vision

To address the extreme requirements of low latency, high energy efficiency, and resource coordination for massive heterogeneous services under the 6G vision, existing cloud-edge transmission mechanisms lack deep coupling between semantic information representation and dynamic resource scheduling, making it difficult to simultaneously meet the semantic fidelity and real-time requirements of multimodal services. Therefore, this paper proposes a cloud-edge resource collaborative transmission strategy driven by semantic communication and AI. First, a hierarchical abstract model of multimodal semantic information is constructed, dividing semantics into signal, object, and target layers to provide differentiated semantic fidelity descriptions for different services. Second, a two-layer cloud-edge collaborative decisionmaking architecture based on deep reinforcement learning is designed: the upper-layer cloud center agent predeploys semantic models and plans baseline resource quotas on a minute-level cycle; the lower-layer edge node agent completes task offloading, joint scheduling of computing and communication resources within milliseconds. Through an adaptive optimization mechanism of immediate feedback in the inner loop and periodic aggregation in the outer loop, the strategy achieves continuous closed-loop evolution. Simulation results show that under balanced mixed load, the proposed strategy achieves an effective throughput of 215.7 tasks/second, an average processing latency of 36.5 milliseconds, an average edge computing resource utilization of 88.7%, and a total system energy efficiency of 24.1 tasks/kJ, verifying its comprehensive superiority in terms of throughput, latency, resource utilization, and energy efficiency. As the work targets semantic communication and cloud-edge transmission, it directly concerns electromagnetic wave propagation and resource scheduling in 6G wireless networks.