Robust and Stable Multipath Transmission in Large-Scale LEO Satellite Networks: An Agentic AI-Driven Approach
Abstract
Recently, the large-scale Low Earth Orbit (LEO) satellite networks are emerging as a cornerstone of future 6G systems, promising global coverage and massive throughput. However, the complex space environments, such as solar outage and ionospheric scintillation, can lead to regional link impairments that severely undermine connectivity. These adverse conditions can significantly compromise end-to-end paths, which drastically diminishes the reliability of the whole connection and eventually degrade the overall transmission performance. To address these gaps, we propose a novel Agentic AI-driven multipath transmission approach to ensure robust and stable data delivery in LEO satellite networks. It features two key innovations: 1) Intent-based multipath routing scheme: Leveraging a hybrid domain-based architecture, distributed agents perceive regional network states to autonomously establish robust multipath routes aligned with specific intent objectives. 2) Fine-grained Multipath QUIC (MPQUIC) congestion control algorithm: Derived from a multipath fluid model, this algorithm performs fine-grained congestion balancing across all sub-paths, and ensures throughput and TCP-friendliness simultaneously in unstable LEO satellite environments. We evaluate the proposed approach through extensive experiments in the Kuiper K3 shell network simulated via UltraStar. Experimental results demonstrate that this approach significantly outperforms other benchmarks in large-scale LEO satellite networks.