A Bandwidth Utilization-Oriented Traffic Shaping Scheme for LEO-IoT

Low Earth Orbit (LEO) satellites and the Internet of Things (IoT) have rapidly developed, giving rise to LEO satellite IoT (LEO-IoT) systems that feature wide coverage and low latency. However, large-scale IoT traffic is prone to network congestion, and the network struggles to meet the QoS requirements for diverse services. This paper proposes a dynamic optimization scheme based on reinforcement learning (RL) for address the traffic shaping and scheduling challenges at LEO-IoT access nodes. First, the upper bound of delay is analyzed using stochastic network calculus (SNC). Second, an optimization problem is formulated to maximize long-term average bandwidth utilization through joint resource allocation and traffic shaping, framed within a service and network matching approach, with delay as the core parameter. To address the complex decision-making challenges in dynamic environments, RL is employed to enable intelligent decision-making. Allocation policies are dynamically adapted based on real-time sensing of network state and the traffic with differentiated delay requirements. Simulations demonstrate that this study provides theoretical support and practical algorithms for efficient resource management and QoS assurance in LEO-IoT.