Date of Award
1-24-2024
Degree Type
Dissertation
Degree Name
Doctor of Philosophy (PhD)
Department
Electrical Engineering and Computer Science
Advisor(s)
Mustafa Cenk Gursoy
Abstract
In recent years, cellular networks and data servers have started experiencing high communication and computational loads. Mobile edge computing (MEC) is an architecture that can be utilized to alleviate communication and computation bottlenecks experienced due to increasing demand on high data traffic and growing number of applications with high computational requirements by processing the offloaded services/tasks from user equipments (UEs) at the edge nodes of networks rather than the remote cloud center. As one common scenario, MEC servers are usually deployed at the base stations (BSs) to process the users’ computational tasks so that the congestion in the network is reduced. Compared with conventional cloud computing, MEC servers, by being deployed in proximity to the end UEs, enable low-latency services while also supporting energy-efficient operation. Indeed, energy efficiency is a critical requirement especially for resource-limited UEs and servers. As a consequence, designing energy efficient strategies and policies is of vital importance in the considered MEC networks. This dissertation focuses on six key perspectives on improving the energy efficiency within the MEC structure: (1) unmanned aerial vehicle (UAV)-assisted communication; (2) optimization and learning based green mobile edge computing; (3) reliability-optimal designs in MEC networks with finite blocklength (FBL) codes; (4) reconfigurable intelligent surface (RIS)-aided MEC networks with FBL codes; (5) UE scheduling with NOMA and content caching; and (6) collaborative inference. In UAV-assisted networks, the mobility of UAVs are leveraged to have them move closer to the UEs and operate as MEC servers to support computationally intensive and latency-critical tasks. In MEC networks, the selection of the MEC server to offload, offloading data ratios of UEs and computational resource (i.e., CPU frequency) allocations at the MEC server are addressed as three important considerations and are optimized to improve energy efficiency. To support computation intensive and latency-critical applications, FBL codes are utilized in wireless data transmission/offloading to satisfy the latency constraints, and hence the reliability in the communication phase is further characterized in the FBL regime. Moreover, RISs are considered as effective means to improve both the spectral efficiency and coverage in wireless systems. By properly setting the phase shift matrix, the RIS can enhance the propagation environment substantially and assist MEC networks. Furthermore, compared with conventional orthogonal multiple access (OMA), non-orthogonal multiple access (NOMA) allows multiple UEs to perform simultaneous transmissions by sharing the same bandwidth and hence improve the spectral usage. Moreover, as both UEs and the MEC server possess the capability to execute learning and perception tasks (such as classification, recognition, reasoning, etc.) to varying extents, collaborative inference in MEC networks has been studied as a strategy to diminish energy consumption while adhering to constrained inference latency. In all aforementioned MEC network models, the overall objective in this dissertation is to maximize the energy efficiency while the system reliability and latency can still be guaranteed.
Access
Open Access
Recommended Citation
Yang, Yang, "Optimization and Learning Based Energy Efficient Designs in Edge Computing" (2024). Dissertations - ALL. 1864.
https://surface.syr.edu/etd/1864
