Towards Practical Reinforcement Learning Designs for Wireless Network Optimization

Author: ORCID icon orcid.org/0000-0002-9714-4291
Yang, Kun, Electrical Engineering - School of Engineering and Applied Science, University of Virginia
Advisor:
Shen, Cong, EN-Elec & Comp Engr Dept, University of Virginia
Abstract:

In the evolving landscape of wireless communication, particularly with the proliferation of LTE and 5G
technologies, network expansion has surged exponentially, presenting significant challenges in optimizing resource allocation for network performance. This dissertation highlights the powerful sequential decision-making capabilities of reinforcement learning (RL) and investigates its unique applications and adaptations within wireless communication systems.

While most research has focused on centralized online RL with discounted rewards, this work explores a broader range of RL methodologies to enhance wireless system development. First, we introduce a decentralized multi-agent RL strategy for user scheduling, transitioning from a centralized to a distributed system to address scalability issues. Next, we delve into offline RL, utilizing existing datasets to improve efficiency. This includes exploring both user scheduling and Radio Access Network (RAN) slicing use cases, analyzing how datasets of varying quality impact offline RL performance.

We further enhance the RL framework by introducing a teaching scheme for individual RL agents,
addressing principal-client problems. Finally, we investigate average reward RL, which prioritizes long-
term objectives, aligning more closely with real-world wireless system requirements.
By exploring these diverse RL approaches, this dissertation aims to broaden the scope of RL applications in wireless communication, providing novel insights and solutions for optimizing network performance in complex and dynamic environments.

Degree:
PHD (Doctor of Philosophy)
Keywords:
Reinforcement Learning, Wireless Communication, Wireless Optimization, RAN slicing, Resource Allocation, Deep Learning
Sponsoring Agency:
National Science FoundationIntel Corporation
Language:
English
Issued Date:
2024/07/16