Yi Zhao: Exploring Resource Allocation for Evolving Wireless and Mobile Computing Systems

Abstract

Wireless communication and computing systems, spanning wireless mobile networks, edge computing infrastructures, and satellite networks, are becoming increasingly heterogeneous, dynamic, and data-intensive. Despite differences in technologies and application scenarios, these systems share a fundamental challenge: The need to accommodate a variety of services using limited system resources. This dissertation addresses this challenge by developing resource allocation strategies to achieve enhanced services and support the emerging ones: 1) Allocation of radio resource for wireless mobile networks, 2) caching, recommendation, and computation of contents for edge computing, and 3) allocation of models, computing resource, and inter-satellite link capacity for federated learning (FL) in satellite networks.

The second part of this dissertation comprises six papers. For research topic 1), Paper I investigates an array of integer linear programming models for radio channel allocation under coupled rate constraints, the relationships between the models’ linear programming approximation. Paper II presents a method based on the derived closed-form solutions and matching theory, for the allocation of frequency-time resource blocks, towards throughput maximization in multi-cell non-orthogonal multiple access (NOMA) scenarios with interference coupling. For research topic 2), Paper III derives approximation algorithms based on problem decomposition and submodularity for cache-hit-ratio maximization, via jointly optimizing content caching and recommendation at network edge. Paper IV considers in particular artificial intelligence (AI)-generated contents and utilizes convex optimization for the allocation of content delivery mode, computing resource, and communication resource, for total utility maximization. For research topic 3), both Paper V and Paper VI present solutions of client selection and inter-satellite routing for fast convergence of FL, based on derived upper bounds of the empirical risk, convergence analysis, and network flow optimization. Paper V focuses on the classical FL framework, while Paper VI designs a novel FL architecture with knowledge distillation, accommodating both the teacher and student models.

Taken together, the dissertation demonstrates that resource allocation, guided by optimization techniques, is a unifying thread connecting wireless, edge, and satellite systems, to deliver enhanced and emerging services.