Evaluation of Transport Layer QUIC over LEO Constellations

Sammanfattning

Low Earth Orbit (LEO) satellite networks are emerging as a key technology for global Internet connectivity. However, their highly dynamic nature, driven by satellite mobility, frequent handovers, and time-varying propagation delays, introduces significant challenges for transport layer protocols. In such environments, conventional congestion control mechanisms often struggle to distinguish between congestion-induced and mobility-induced variations in network conditions. This thesis evaluates the performance of the QUIC transport protocol in LEO satellite networks using a realistic simulation framework based on OMNeT++, the INET framework, and mobility models derived from satellite orbital data. The study first analyzes the behavior of QUIC under different application workloads, including transactional and continuous streaming traffic, highlighting its stability under varying traffic patterns. To address the limitations of traditional congestion control, a reinforcement learning (RL)-based adaptive pacing mechanism is proposed. The approach operates as a control layer on top of the BBR congestion control algorithm, refining the pacing rate through small adjustments based on observed network conditions such as round-trip time, throughput, and packet loss. The RL agent is trained using Proximal Policy Optimization (PPO) to ensure stable learning under non-stationary conditions. Experimental results demonstrate that the proposed RL-based approach improves delay stability, reduces packet loss, and significantly enhances recovery performance during satellite handovers, while maintaining comparable throughput to baseline QUIC-BBR. The findings indicate that reinforcement learning can effectively enhance transport layer adaptability in highly dynamic LEO satellite environments. Overall, this work provides insights into QUIC performance in satellite networks and demonstrates the potential of learning-based techniques for improving congestion control in next-generation communication systems.