AI-DRIVEN MULTI-TIER LOAD BALANCING FOR ENHANCED QOS IN LEO/MEO/GEO 5G SATELLITE NETWORKS

Abstract

The multi-tier satellite constellations, that is, the Low, Medium, and Geostationary Earth Orbit (LEO/MEO/GEO) satellites, combined with 5G networks on the ground, are the key to attaining ubiquitous connectivity across the globe. This heterogeneous architecture is modeled as a dynamic three-tier network system, where the topological arrangement and orbital geometry define the network architecture-QoS relationships governing and load distribution. To address these challenges, this paper proposes a novel Hierarchical Multi-Agent Deep Reinforcement Learning framework for Multi-Tier Load Balancing (Hi MADRL-MTLB). It is an architecture with a two-level control hierarchy; a meta-controller manages global inter tier traffic distribution, and distributed local controllers manage intra-tier optimization in each orbital cluster and use a Centralized Training with Decentralized Execution (CTDE) paradigm. The framework specifically includes 5G classes of services (eMBB, URLLC, and mMTC) and orbital dynamics in its state representation and reward model to facilitate proactive and QoS-driven decision-making. Extensive simulation findings show that, under the condition of the largest load variance decrease, there is a 67 percent reduction in QoS violations and an 18.4 percent increase in aggregate throughput, demonstrating that Hi-MADRL-MTLB outperforms traditional shortest-path routing. It also has a practical 50-millisecond decision cycle so that it can scale to immense constellations. This study conclusively proves that hierarchical artificial intelligence control is necessary in the management of resources efficiently in next-generation integrated satellite-terrestrial networks, giving a base answer to the developing 5G and the following 6G ecosystems.