City Research Online

Abstract

Modern multi-access Beyond 5G (B5G) networks provide mobile terminals with aggregated capacity across heterogeneous paths, improving overall network performance. However, in high-mobility environments such as vehicular networks, supporting effective multiaccess connectivity remains challenging. Rapid wireless link quality fluctuations frequently outpace the responsiveness of existing multipath schedulers, causing out-of-order packet delivery, head-of-line blocking, and degraded quality of experience.

This work addresses these challenges through DRL-based Adaptive Rate Allocation (DARA), a framework integrating Transformer-based path state forecasting with deep reinforcement learning for multipath traffic splitting. The Transformer module predicts congestion window and round-trip time evolution over a 500 ms horizon, whilst a Deep Q-Network dynamically computes optimal path utilisation fractions based on current observations and predicted states. A six-component reward function balances throughput maximisation, delay minimisation, quality sustainability, preemptive adaptation, stability, and idleness prevention.

Performance evaluation employs an MP-DCCP testbed with traces captured from real mobile users across five network conditions of increasing volatility. Controlled burst experiments validate the predictive mechanism, demonstrating substantial delay reduction through preemptive allocation adjustment. Real-world trace evaluation across file transfer, adaptive streaming, live video, and concurrent workload scenarios shows consistent improvements over both rule-based and state-of-the-art learning-based schedulers. Crossprotocol comparison on encrypted QUIC traffic demonstrates that DARA maintains performance gains where existing intelligent schedulers degrade, whilst concurrent workload tests confirm balanced flow allocation without the starvation effects observed in competing approaches. DARA’s relative advantage increases with network volatility up to moderate-high mobility before graceful degradation under extreme conditions, validating the architecture’s targeting of challenging high-mobility scenarios where reactive schedulers fundamentally cannot adapt.

Publication Type:	Thesis (Doctoral)
Subjects:	Q Science > Q Science (General) T Technology > T Technology (General) T Technology > TK Electrical engineering. Electronics Nuclear engineering
Departments:	School of Science & Technology > Department of Engineering School of Science & Technology > School of Science & Technology Doctoral Theses Doctoral Theses

Export

Downloads