An Effective Handover Strategy for Fast-moving Devices in LEO Satellite Network Based on Deep Reinforcement Learning

With the development of aerospace technology, the cost of deploying low Earth orbit (LEO) satellites is decreasing. LEO satellites can provide low latency and high throughput networks on a global scale. Recently, LEO satellite networks are discussed to provide network services for user equipments (UEs) in airplanes. However, both airplanes and LEO satellites are moving rapidly, resulting in dynamic changes to the network topology. This work proposes an efficient handover strategy based on deep reinforcement learning to achieve seamless network service in this scenario. In this approach, fast-moving devices (e.g., airplanes) are treated as agents that learn to adapt to changes in the network environment. The designed Deep QNetwork (DQN) model determines suitable handover trigger timings and target satellites for these fast-moving devices, guided by comprehensive reward functions that considers multiple performance metrics. Simulation results indicate that, the proposed scheme can significantly enhance network throughput and reduce latency.