Abstract
This letter highlights the combined advantages of Open Radio Access Network (O-RAN) and distributed Artificial Intelligence (AI) in network slicing. O-RAN's virtualization and disaggregation techniques enable efficient resource allocation, while AI-driven networks optimize performance and decision-making. We propose a federated Deep Reinforcement Learning (DRL) approach to offload dynamic RAN disaggregation to edge sites to enable local data processing and faster decision-making. Our objective is to optimize dynamic RAN disaggregation by maximizing resource utilization and minimizing reconfiguration overhead. Through performance evaluation, our proposed approach surpasses the distributed DRL approach in the training phase. By modifying the learning rate, we can influence the variance of rewards and enhance the convergence of training. Moreover, fine-tuning the reward function's weighting factor enables us to attain the targeted network Key Performance Indicators (KPIs).