Artificial intelligence-driven resource allocation in 6G networks for ultra-low latency applications

The emergence of sixth-generation (6G) networks necessitates innovative solutions to meet the stringent requirements of ultra-low latency applications, such as holographic communications and autonomous systems. Traditional centralized scheduling techniques often compromise both scalability and user privacy. This paper presents the federated deep Q-network (FedDQN) to overcome these obstacles. This innovative architecture enables the decentralized and privacy-preserving allocation of resources by combining deep reinforcement learning and federated learning. Furthermore, FedDQN enhances user privacy by allowing base stations to collaborate and establish an international framework without exchanging raw data. This paper verifies the efficacy of FedDQN through comprehensive simulations by comparing its results with those of five baseline techniques, including heuristic and advanced reinforcement learning-based methods. Our findings demonstrate that FedDQN consistently outperforms the baselines in three key metrics: a 30% reduction in communication overhead, sub-5 ms latency, and fairness, as indicated by Jain's fairness index of 0.75. These results highlight FedDQN's ability to meet the rigorous specifications of 6G ultra-low latency applications, providing a scalable and private option for next-generation wireless networks.