This Is AuburnElectronic Theses and Dissertations

Mobility and Energy Management in 5G Ultra-Dense Networks




Sun, Li

Type of Degree

PhD Dissertation


Computer Science and Software Engineering


Triggered by the development of 5G technologies, the demand for mobile data has grown tremendously in recent years. It leads to an urgent need to upgrade the current macro-cell-based architecture of network infrastructure from which no substantial amount of future system performance gains could be obtained. Ultra-dense network (UDN) is a promising technique to meet the requirement of explosive data traffic in the 5G era, because of its ability to provide better spectrum efficiency. However, a large number of small base stations (SBSs) or access points (APs), associated with massive MIMO and millimeter wave (mmWave) technologies, make the UDN suffer from severe interference, signaling overhead, and power consumption issues. Therefore, effective mobility and energy management strategies that take account of the architecture of UDN are required to take advantage of 5G technologies fully. In this dissertation, I propose several novel strategies and methods to improve mobility and energy management in 5G UDNs. Specifically, I address the issue of frequent handover in mmWave UDN with the goal of enhancing time-frequency resource efficiency. By considering the spatial and temporal features of handover, I propose two multi-armed-bandit (MAB) based handover strategies to reduce the handover frequency by exploiting the empirical knowledge distribution of the user's geographical location and the line-of-sight (LOS) blockage. Secondly, to address the signaling overhead issue of centralized downlink precoding in cell-free massive MIMO systems, I propose a novel bandwidth-efficient global zero-forcing precoding strategy associated with a model-based CSI compression method leveraging the physical structure of Rician fading channels. Thirdly, to address the power consumption and scalability issues in cell-free massive MIMO systems, I propose a multi-agent deep reinforcement learning-based AP activation strategy as a scalable solution to improve energy efficiency in UDNs.