A Successive Deep Q-Learning Based Distributed Handover Scheme for Large-Scale LEO Satellite Networks

With the rapid increasing number of the deployed satellites, the handover strategy becomes more challenging for large-scale low-earth orbit (LEO) constellations. In this paper, a distributed satellite handover scheme for large-scale LEO constellations is proposed, which not only takes the handover delay, handover failure, quality-of-service (QoS) requirements of users, and inter-satellite traffic balancing into consideration, but also enables each user to dynamically perform the handover process only with the local information. Specifically, we adopt a shadowed Rice model to characterize the user-satellite channel, which is determined by the elevation angle between the user and satellite. Then, the user utility function is designed, where the user transmission rate requirement and the number of available channels of visible satellites are jointly considered. An overall long-term utility maximization problem is further formulated. By exploiting the independence feature of different satellites and the fact that each user only has finite number of visible satellites, a low-complexity successive deep Q-learning algorithm is developed, which can significantly reduce the dimensions of state spaces and efficiently solve the formulated problem in a distributed manner. Simulation results show that the proposed scheme can achieve better performance over existing methods.