GPT Agent-Supported Edge Intelligence for Optimizing D2D Communications

The sixth-generation (6G) network integrates communication, sensing, and computation into a synergetic system. Device-to-device (D2D) communication has also received widespread attention, and to improve the performance of D2D communications with limited network edge resources, we propose the use of the generative pretrained transformer (GPT) agent. Specifically, we use GPT to achieve resource-optimized quality of service (QoS) and energy consumption, forming the QoS-CNNGPT method. The simulation results show that the proposed system can support multiple edge users with a 28.7% improvement in spectral efficiency compared with the weighted minimum mean square error (WMMSE) method and a 79.8% reduction in computation time compared with the overhead of reinforcement learning (RL)-based techniques. The proposed system can also meet the deployment and individualization requirements of different users in resource-limited D2D systems with strong robustness, which will help improve 6G networks.