Adaptive Resource Management Frameworks for Wireless Virtual Reality Networks: Applications and Optimization

Abstract

Wireless Virtual Reality (VR) networks face significant challenges in resource management and real-time performance optimization due to dynamic user mobility, viewport changes, and stringent latency constraints critical for quality of experience (QoE). This thesis investigates adaptive resource management frameworks for wireless VR networks through progressive evolution from optimization principles to machine learning approaches.

The research presents four major contributions categorized by their decision-making mechanisms:

Optimization-Based Contributions employ mathematical optimization for resource allocation:

- A framework exploiting temporal correlation in user viewing patterns to reduce transmission overhead while maintaining visual quality, achieving significant delay reduction through utility-based Small Base Station (SBS) selection and dynamic resource allocation.

- A framework combining competitive auction mechanisms with federated learning-based viewport prediction, incorporating a five-cluster preference model for quality, latency, computation, communication, and rendering requirements while preserving privacy.

Machine Learning-Driven Contributions employ reinforcement learning agents for decision-making:

- A distributed coordination system based on Proximal Policy Optimization enabling autonomous SBS decision-making with actor-critic networks, demonstrating substantial social welfare improvement.

- Mobile Crowdsensing integration with Deep Reinforcement Learning for context-aware resource allocation, incorporating reward shaping techniques to achieve significant QoE improvements.

This progressive evolution demonstrates systematic methodology in addressing wireless VR challenges, with each contribution building upon previous insights to advance adaptive resource management for immersive communication systems.