Energy-Efficient Resource Allocation in LEO-assisted Open RAN architecture towards 6G

In recent years, the Low Earth Orbit (LEO) satellite has demonstrated the wireless access ability to terrestrial unconnected users that have no direct connection to the terrestrial 6 G network. Artificial Intelligence (AI) has shown superior performance in network optimization and energy efficiency, and AI is also the key enabler for future 6 G. In addition, 6 G will provide AI abilities to User Equipment (UE). Therefore, bringing AI elements into LEO is urgent. Especially, the Radio access network Intelligence Controller (RIC) in Open Radio Access Networks (Open RAN), including near-real-time RIC (near-RT RIC) and non-real-time RIC (non-RT RIC), provides intelligent elements to RAN. To this end, we propose a novel LEO-assisted Open RAN (LO RAN) architecture. In this architecture, the near-RT RIC can be deployed on the LEO satellites to provide intelligence computing service to terrestrial UE. On the other hand, the non-RT RIC is deployed in terrestrial computing centers to train and infer the AI models for intelligent computing services. We also proposed a Deep Deterministic Policy Gradient (DDPG) that considers task offloading decisions, computing resource allocation, and cache updated aiming at minimizing energy consumption. We conduct numerical simulations to demonstrate the performance of the proposed DDPG. Compared with Soft Actor-Critic (SAC) and Dueling Deep-Q-Network (DDQN) methods, the results show that DDPG reduces an average of 35% energy consumption to SAC and 38% to DDQN.