Year: 2023
Type: dissertation
Abstract: Reconfigurable intelligent surface (RIS) can intelligently manipulate the impinging electromagnetic wave to improve the wireless propagation environment between the transceivers. New communication algorithms need to be explored as conventional algorithms are difficult to deploy in RIS-assisted networks due to the passive structure of the hardware. This thesis focuses on finding deep learning-based approach to address two challenging problems in RIS-assisted networks, namely 1) user scheduling in densely populated networks and 2) uplink localization via active sensing. To address the former problem, this thesis shows that graph neural networks can be used to schedule users and design RIS configurations to achieve high overall throughput while ensuring user fairness. For the latter problem, this thesis shows that long short-term memory networks can be used to design the RIS to gradually narrow down the search area to locate the user. Numerical results show that the proposed learning-based approaches outperform existing conventional approaches.
Source: TSpace (University of Toronto)
Author: Zhang, Zhongze
Institution: University of Toronto
Language: English
Subfield: Electrical and Electronic Engineering
Field: Engineering
Domain: Physical Sciences
Open Access status: green
OpenAlex ID: W7132921584
