Federated Reinforcement Learning-Based Radio Resource Scheduling for Short Video Applications

Short videos, as one of the most popular forms of multimedia, impose stringent transmission requirements on wireless networks. A Cell-Free network with numerous access points and a cloud server based on artificial intelligence is a promising approach for improving users' quality of experience. This research presents a federated reinforcement learning based radio resource scheduling method for access points to construct a global model at the cloud server to explore the potential of the Cell-Free networks while avoiding privacy leakage. First of all, local access points train their local models and upload their weights to cloud server. Then, the server aggregates the local models to obtain the global model, which is subsequently sent back to the access points. Finally, the access points update their local models with the latest global model. The simulation results show that the advantages of the proposed algorithm are the effective resource scheduling policy, the convergence speed and the scalability. Meanwhile, the proposed algorithm outperforms baselines with significant performance improvement.