Theses and Dissertations
Deep reinforcement learning for advanced wireless networks enabling service and spectrum coexistence
ORCID
https://orcid.org/0000-0001-9888-7890
Advisor
Marojevic, Vuk
Committee Member
Young, Maxwell
Committee Member
Gurbuz, Ali
Committee Member
Ball, John E.
Committee Member
Kurum, Mehmet
Date of Degree
5-10-2024
Original embargo terms
Embargo 1 year
Document Type
Dissertation - Open Access
Major
Electrical and Computer Engineering
Degree Name
Doctor of Philosophy (Ph.D)
College
James Worth Bagley College of Engineering
Department
Department of Electrical and Computer Engineering
Abstract
The evolution from the fifth generation (5G) networks to 6G promises to revolution- ize connectivity, supporting a vast array of applications from high-definition video streaming and immersive augmented reality experiences to critical machine-type communications. However, this progression brings along the challenge of efficiently managing the limited radio spectrum (RF) resources to accommodate the diverse quality of service (QoS) requirements of a variety of service and user types. Another problem gaining traction with the advances of wireless communications technology is the coexistence between active communication systems and passive RF sensing operating in the same or adjacent bands. Central to addressing these challenges is the proposed application of deep reinforcement learning (DRL), which emerges as a tool for adaptive and intel- ligent radio resource management (RRM) in the face of the increasingly complex and dynamic RF system requirements. This dissertation investigates the application of DRL for service, user, and network management of advanced wireless networks operating in dedicated and shared spectrum. Through a series of innovative DRL-based frameworks and solutions to a variety of emerging RRM problems, this work contributes to the optimization of spectrum, transmission power, and band- width allocation, as well as network configuration. We contribute to the integration of cutting-edge technologies such as unmanned aerial vehicles as aerial base stations, reconfigurable intelligent sur- faces, and multi-user multiple input, multiple output systems for a seamless user experience. The core of the dissertation explores how DRL can adaptively manage spectrum resources that satisfy the QoS requirements of different 5G service classes, specifically enhanced mobile broadband and ultra-reliable low-latency communications, while also facilitating the integration of terrestrial and aerial network nodes to enhance coverage and capacity. This dissertation further extends into the domain of coexistence between active wireless communication systems and passive remote sensing technologies. We collect radiometric measurement data in a custom-built software-defined radio testbed for which we design different 5G downlink transmission patters and data sets. Based on the collected and processed data from the testbed’s radiometer, we propose a DRL-based strategy to manage 5G communications while reducing the RF interference impact on co-channel radiometric measurements. Through simulations, the proposed solution demonstrates the tradeoffs between communications and sensing operations in terms of common wireless network performance met- rics, such as sum data rate and user fairness, and brightness temperature readings obtained by the radiometer.
Recommended Citation
Alqwider, Walaa, "Deep reinforcement learning for advanced wireless networks enabling service and spectrum coexistence" (2024). Theses and Dissertations. 6072.
https://scholarsjunction.msstate.edu/td/6072