Lyapunov-Optimized Traffic Splitting in TN-NTN using QMIX for Enhanced Network Stability

Next-generation networks aim to deliver ubiquitous services such as enhanced mobile broadband (eMBB) and ultra-reliable low-latency communications (URLLC), where multi-connectivity (MC) leveraging multiple paths is essential. However, terrestrial networks (TNs) alone cannot provide global coverage, motivating the integration of low Earth orbit (LEO)-based non-terrestrial networks (NTNs). This heterogeneous environment introduces complex traffic orchestration challenges, with intelligent control still in its early research phase. To address this, this paper proposes a dynamic traffic splitting framework based on multi-agent reinforcement learning (MARL), where each user equipment (UE) cooperatively learns the optimal traffic ratio using the QMIX algorithm. A Lyapunov-based reward function ensures both QoS satisfaction and long-term stability. Simulation results confirm the framework’s effectiveness in enhancing user experience and network efficiency in integrated TN-NTN systems.