Open Collections

Mehrabian, Ali

University of British Columbia

Predictive analysis of traffic demands for the sixth-generation (6G) wireless systems plays an important role in network resource provisioning. As 6G networks aim to support various applications, forecasting traffic demands accurately is instrumental for efficient resource allocation and ensuring high-quality user experience in dynamic wireless environments. Moreover, high data rate requirements of content-rich applications necessitate the exploration of emerging technologies. This study focuses on two such technologies: terahertz (THz) band communication and reconfigurable intelligent surface (RIS). The utilization of the THz band enables wireless systems to achieve ultra-high data rates, while RIS can improve the coverage service of wireless networks. However, these technologies introduce previously uncharted challenges, which require novel resource allocation approaches to address them and fully leverage their potential. In this thesis, we propose graph neural network (GNN) learning algorithms for traffic demand prediction and network resource optimization in order to improve the performance of 6G wireless systems. First, we propose a dynamic Bernstein graph recurrent network (DBGRN). The proposed learning algorithm utilizes the information in the spatial, temporal, and spectral domains to predict traffic in wireless cellular networks. The experimental results using a real-world traffic dataset show that the proposed DBGRN outperforms four state-of-the-art baseline models, and provides a lower root mean squared error (RMSE) and mean absolute error (MAE). Second, we study the sum-rate maximization problem with quality-of-service (QoS) constraints in RIS-aided multiuser multiple-input multiple-output (MU-MIMO) THz systems. We propose a metapath-based heterogeneous graph-transformer network (MHGphormer) to jointly optimize the precoding, RIS phase shifts, and THz sub-bands bandwidth allocation. Simulation results show that our proposed MHGphormer achieves a higher system sum-rate with faster convergence when compared with two other learning-based algorithms.

Text

2023-10-18

Attribution-NonCommercial-NoDerivatives 4.0 International

http://hdl.handle.net/2429/86194

Electrical and Computer Engineering

University of British Columbia

UBCV

http://creativecommons.org/licenses/by-nc-nd/4.0/