Reinforcement Learning-based User-centric Handover Decision-making in 5G Vehicular Networks
MetadataShow full item record
AbstractThe advancement of 5G technologies and Vehicular Networks open a new paradigm for Intelligent Transportation Systems (ITS) in safety and infotainment services in urban and highway scenarios. Connected vehicles are vital for enabling massive data sharing and supporting such services. Consequently, a stable connection is compulsory to transmit data across the network successfully. The new 5G technology introduces more bandwidth, stability, and reliability, but it faces a low communication range, suffering from more frequent handovers and connection drops. The shift from the base station-centric view to the user-centric view helps to cope with the smaller communication range and ultra-density of 5G networks. In this thesis, we propose a series of strategies to improve connection stability through efficient handover decision-making. First, a modified probabilistic approach, M-FiVH, aimed at reducing 5G handovers and enhancing network stability. Later, an adaptive learning approach employed Connectivity-oriented SARSA Reinforcement Learning (CO-SRL) for user-centric Virtual Cell (VC) management to enable efficient handover (HO) decisions. Following that, a user-centric Factor-distinct SARSA Reinforcement Learning (FD-SRL) approach combines time series data-oriented LSTM and adaptive SRL for VC and HO management by considering both historical and real-time data. The random direction of vehicular movement, high mobility, network load, uncertain road traffic situation, and signal strength from cellular transmission towers vary from time to time and cannot always be predicted. Our proposed approaches maintain stable connections by reducing the number of HOs by selecting the appropriate size of VCs and HO management. A series of improvements demonstrated through realistic simulations showed that M-FiVH, CO-SRL, and FD-SRL were successful in reducing the number of HOs and the average cumulative HO time. We provide an analysis and comparison of several approaches and demonstrate our proposed approaches perform better in terms of network connectivity.
The following license files are associated with this item:
- Creative Commons