Heterogeneous User-Centric Cluster Migration Improves the Connectivity-Handover Trade-Off in Vehicular Networks

• Published in: IEEE transactions on vehicular technology Vol. 69; no. 12; pp. 16027 - 16043
• Main Authors: Lin, Yan, Zhang, Zhengming, Huang, Yongming, Li, Jun, Shu, Feng, Hanzo, Lajos
• Format: Journal Article
• Language: English
• Published: New York IEEE 01.12.2020; The Institute of Electrical and Electronics Engineers, Inc. (IEEE)
• Subjects: Clustering; Decision making; deep reinforcement learning; Handover; heterogeneous; max-bipartite matching; Network topology; Networks; Next generation networking; Roads; Switches; Throughput; Tradeoffs; User-centric clustering; Vehicle dynamics; Vehicle-to-infrastructure; Vehicular ad hoc networks; vehicular networks
• ISSN: 0018-9545; 1939-9359
• DOI: 10.1109/TVT.2020.3041521

User-centric (UC) clustering has recently emerged as a promising paradigm for enhancing the connectivity of mobile users by grouping an appropriate number of access points (APs), thus paving the way for seamlessly connected vehicular networks. However, for a high-velocity vehicular user, UC clustering may lead to overly frequent handovers (HOs), which increases the risk of throughput-reduction, call dropping and energy wastage. To mitigate this problem, we aim for reducing the HO overhead imposed on the heterogeneous UC (HUC) cluster migration process of vehicular networks. Specifically, we first conceive a novel hybrid HUC cluster migration strategy that adaptively switches between horizontal and vertical HOs for supporting both vehicle-to-vehicle (V2V) and vehicle-to-infrastructure (V2I) communication. Then, a dynamic decision-making problem is formulated for balancing the benefits of HUC cluster migration and the total HO overhead, subject to realistic HUC clustering constraints. In the face of unknown vehicular mobility, we propose a sequential HUC cluster migration solution based on max-bipartite matching theory imposing a low complexity. As a design alternative, we also propose a holistic solution relying on model-free deep reinforcement learning (DRL). Finally, our numerical results reveal the superiority of the proposed cluster migration design in terms of striking a compelling trade-off between the per-user average data rate (PAR) and the number of HOs in different scenarios.