Joint Optimization of Access Control and Resource Allocation for LEO Satellite IoT Using Deep Reinforcement Learning

Integrated satellite-terrestrial communication is one of the key features of 6G, and the application of low-Earth orbit (LEO) satellites in global Internet of Things (IoT) services has attracted significant attention. In LEO satellite IoT systems, terrestrial devices connect to the network using random access. However, the rapid growth of IoT devices and significant satellite-terrestrial link delays pose serious challenges to existing access mechanisms. To address these issues, this paper proposes a joint optimization algorithm for efficient access control and resource allocation based on a cross-layer design. In the access control phase, an optimization problem is formulated to maximize the average random access efficiency, and matching theory is applied to reduce signaling overhead. In the resource allocation phase, a Lyapunov optimization-based long-term online algorithm was proposed, which achieves efficient time-frequency resource scheduling by transforming the long-term optimization problem into a series of low-complexity short-term subproblems. To further enhance optimization and reduce computational complexity, a deep reinforcement learning (DRL) model is employed to jointly optimize access control and resource allocation. The proposed algorithm is analyzed theoretically and validated through simulations. Simulation results show that the proposed DRL-based joint optimization algorithm performs well under various system configurations and significantly improves access efficiency and resource utilization.