Enhancing XR Application Performance in Multi-Connectivity Enabled mmWave Networks

• Published in: IEEE Open Journal of the Communications Society
• Main Authors: Muhammad Affan Javed, Pei Liu, Shivendra S. Panwar
• Format: Article
• Language: English
• Published: IEEE 2023-01-01
• Subjects: Blockages; deadline-driven scheduling; deep learning; DQN; handover; low latency
• Online Access: https://ieeexplore.ieee.org/document/10272712/

mmWave communications are paving the way for next-generation cellular networks due to their inherent ability to provide high data rates and mitigate interference. Coupled with this are the enormous potential and challenges posed by eXtended Reality (XR) applications which are becoming increasingly ubiquitous. In this paper, we leverage the unique characteristics of mmWave networks to re-think and re-design fundamental network architecture and functions in order to meet the strict requirements of deadline-driven XR applications. We propose a multi-tiered multi-connectivity architecture that allows users (UEs) to connect to multiple base stations (gNBs) simultaneously and switch rapidly between them in case of blockages. By replicating UE data at multiple gNBs close to the UE, we ensure that we satisfy strict Quality of Service (QoS) constraints even with unpredictable, dynamic blockages of the mmWave links. We show through extensive system-level simulations that our network architecture allows us to shield UEs from high handover delays and minimizes data plane interruptions in case of blockages. Moreover, we note that existing algorithms for network functions such as gNB selection and scheduling are not optimized for the multi-connectivity paradigm, nor do they specifically cater to strict deadline constraints or intermittent wireless links. We propose a Deep Reinforcement Learning framework that selects gNBs for data replication by explicitly optimizing to meet strict deadline constraints of XR traffic. Our Deep Learning agent analyzes global state information and predicts the best selection of gNBs to preemptively replicate data for future transmissions. Furthermore, we propose a scheduler based on maximal weight matching, dubbed <inline-formula> <tex-math notation="LaTeX">$\beta -$ </tex-math></inline-formula>MWM, which is specifically tailored to exploit multi-connectivity. We show that our Deep Learning based Data Replication Predictor and <inline-formula> <tex-math notation="LaTeX">$\beta -$ </tex-math></inline-formula>MWM scheduler perform better than existing, conventional algorithms and result in markedly better performance for XR applications with strict deadlines.