Load Balancing User Association and Handover in Millimeter-Wave Enabled Wireless Networks
Alizadeh, Alireza.
2021
-
Thesis
(Ph.D.)--Tufts University, 2021.
Submitted to the Dept. of Electrical Engineering.
Advisor: Mai Vu.
Committee: Brian Aull, Liping Liu, and Edmund Yeh.
Keyword: Electrical engineering.
In a wireless network, user association is the problem of optimally connecting base stations (BS) and user equipment (UEs) in order to maximize a network ... read moreutility while balancing the BSs' loads (number of UEs each BS can serve simultaneously). In a millimeter wave (mmWave)-enabled network, user association changes the highly directional beamforming connections and hence significantly affects the network interference. In this thesis, we introduce a new user association formulation by considering these association-interference-data rate dependencies and propose several novel approaches to address user association and handover in beyond 5G (B5G) networks. We first formulate the user association problem as integer nonlinear programming and design a polynomial-time centralized algorithm, called Worst Connection Swapping (WCS), to find a near-optimal solution. The proposed algorithm outperforms other generic algorithms for combinatorial programming and reaches a near-optimal solution. Then, we employ matching theory and propose novel and efficient distributed matching games, called \textit(early acceptance) (EA), which allow UEs to apply for association in a distributed fashion and get accepted with minimal delay while the game is running. Next, we propose online reinforcement learning-based algorithms for load balancing user association and handover. The proposed approach allows UEs to engage in best-rate data transmission while effectively participating in a background learning process indefinitely. We also apply a realistic measurement model to capture user mobility and channel variations and address handover in dynamic networks. Our proposed centralized WCS algorithm outperforms existing user association schemes, which ignore the effect of user association on interference, confirming that the interference is highly dependent on user association.The proposed distributed EA matching game outperforms the well-known deferred-acceptance matching game in achieving both a higher throughput, reaching closely that of the centralized WCS algorithm, and a significantly shorter association delay. Our novel reinforcement learning-based algorithms exhibit fast convergence and require no offline training, a feature that allows efficient online implementation for practical networks. Together, the formulations and algorithms in this thesis provide an integrated set of approaches and solutions to load balancing user association and handover in dynamic B5G networks, offering both theoretical performance benchmarking and novel solutions applicable in practice.read less - ID:
- jq0860934
- To Cite:
- TARC Citation Guide EndNote
- Usage:
- Detailed Rights
