RC-DQN Based Intelligent Handover for Large-Scale LEO Satellite Networks

Large-scale low earth orbit (LEO) satellite networks are expected to act as a key enabler to provide ubiquitous and continuous connectivity in the era of the sixth generation of wireless systems (6G). However, the high mobility and limited coverage time of LEO satellites lead to a problem of frequent handovers. Conventional satellite handover schemes usually rely on a single specific criterion, such as signal quality or service time, and cannot bring balanced performance. Handover schemes based on deep Q-network (DQN) have exhibited the ability to make handover decisions more intelligent and balanced. To resolve the challenging satellite handover problem, we propose a DQN based handover method with the reward centering technique, which considers signal quality, satellite load and remaining service time in the reward design and subtract the reward average from the observed rewards. Simulation results show that the proposed satellite handover method can significantly reduce the number of handovers compared with the benchmark and provide robustness to different discount factors and reward offsets.