Date of Award

12-2025

Document Type

Dissertation

Degree Name

Doctor of Philosophy (PhD)

Department

Electrical and Computer Engineering

Committee Chair/Advisor

Dr. Fatemeh Afghah

Committee Member

Dr. Kuang-Ching Wang

Committee Member

Dr. Linke Guo

Committee Member

Dr. Ismail Guvenc

Committee Member

Dr. Xiaoyong Yuan

Abstract

Next-generation wireless networks must deliver highly adaptive, scalable, and intelligent connectivity to satisfy the heterogeneous demands of emerging services, including enhanced mobile broadband, massive machine-type communications, and ultra reliable low latency applications. The Open Radio Access Network (O-RAN) paradigm has emerged as a key enabler of this vision, introducing openness, virtualization, and artificial intelligence (AI)-driven control into the RAN ecosystem. O-RAN’s disaggregated architecture facilitates multi-vendor interoperability and empowers intelligent management through the RAN Intelligent Controller (RIC). However, achieving real-time, autonomous, and generalized optimization in such a dynamic environment remains a significant challenge due to its distributed nature, non-stationary traffic, and diverse Quality of Service (QoS) requirements.

This dissertation addresses these challenges by developing a unified framework for AI-driven resource management and network slicing in O-RAN, centered on reinforcement learning (RL) and its advanced extensions. The research makes significant progress through several novel contributions, each addressing a fundamental limitation of conventional optimization and control methods. First, an Evolutionary Deep Reinforcement Learning (EDRL) framework is proposed to enhance the generalization and stability of O-RAN slice management by leveraging population-based exploration. The results demonstrate a substantial improvement, up to 62.2% in maximum cumulative reward, compared to baseline DRL models, highlighting the benefit of evolutionary diversity in dynamic spectrum and resource allocation. Building on this foundation, a Multi-Agent Reinforcement Learning (MARL) paradigm is developed to enable distributed intelligence among O-RAN components. By deploying cooperative agents across Distributed Units (DUs) and coordinating through the RIC, this approach achieves scalable, fault tolerant control that maintains QoS under fluctuating traffic patterns. To address traffic non-stationarity, the framework is extended with a hybrid LSTM-predictive DRL design that forecasts traffic loads and proactively reallocates resources to prevent QoS degradation.

To further enhance adaptability, a Meta-Deep Reinforcement Learning (Meta-DRL) approach is introduced, inspired by Model Agnostic Meta Learning (MAML), which enables rapid policy adaptation to unseen network conditions. This method significantly accelerates convergence and enhances robustness across varying O-RAN configurations. Additionally, to overcome the instability and poor generalization associated with sharp minima in non-convex optimization, a Sharpness Aware Reinforcement Learning (SAM-RL) framework is developed. By integrating Sharpness Aware Minimization into the policy optimization loop, the controller achieves flatter optima and more reliable performance across diverse environments. Finally, the research culminates in the introduction of LLM-Augmented Reinforcement Learning for O-RAN control, where Large Language Models (LLMs) are incorporated as reasoning and contextualization modules to guide decision making. Through frameworks such as ORAN-GUIDE and ORANSight-GPT, LLMs enable domain aware prompt generation and semantic understanding of network states, transforming the traditional numeric optimization pipeline into a cognitively guided decision process. This integration bridges symbolic reasoning with data driven control, enabling interpretable, adaptive, and generalized policy formation.

This dissertation establishes a comprehensive trajectory from distributed reinforcement learning to optimization aware and cognitively guided O-RAN intelligence. The proposed frameworks collectively advance the frontiers of autonomous network management by achieving improved convergence, interpretability, and generalization under real world non-stationary conditions. These contributions lay the groundwork for the evolution of fully autonomous, self optimizing, and self healing wireless networks envisioned in 6G and beyond.

Author ORCID Identifier

https://orcid.org/0009-0009-1691-0029

Share

COinS
 
 

To view the content in your browser, please download Adobe Reader or, alternately,
you may Download the file to your hard drive.

NOTE: The latest versions of Adobe Reader do not support viewing PDF files within Firefox on Mac OS and if you are using a modern (Intel) Mac, there is no official plugin for viewing PDF files within the browser window.