A DRL-based handover strategy for ISL-enabled LEO mega-constellation networks under heterogeneous QoS demands

The rapid deployment of Low Earth Orbit (LEO) mega-constellations with Inter-Satellite Links (ISL) promises global coverage and enhanced connectivity for next-generation networks. However, the exponential expansion of constellation scale, ISL-induced long transmission paths, and diverse user demands will result in complex handover issues, leading to significant end-to-end latency and jitter. To address these issues, this paper proposes a Deep Reinforcement Learning (DRL)-based handover strategy for ISL-enabled mega-constellation networks, where the end-to-end propagation delay, heterogeneous user requests, and limited LEO satellite capacity are taken into consideration. Specifically, we first construct analytical models for Ground-Satellite Links (GSL) and ISL to rapidly estimate the user service latency. Three novel handover factors including access success rate, average service duration, and satellite load balancing are designed to quantify handover efficiency. Subsequently, we formulate a multi-objective handover optimization problem that maximizes long-term network utility by jointly considering the above handover factors under dynamic topology and varied Quality of Service (QoS) constraints. To reduce service interruptions and signaling overhead, a centralized handover procedure integrated with preparation and implementation stages is designed. Based on the proposed handover procedure, we further develop a multi-agent RL-based handover algorithm to efficiently solve the formulated problem, where users act as collaborative agents coordinated by a management center with reduced state and action space. Extensive simulation results have validated the superiority of the proposed handover strategy in ISL-enabled mega-constellation networks.