- Library Home /
- Search Collections /
- Open Collections /
- Browse Collections /
- UBC Theses and Dissertations /
- Machine learning-based algorithms design for network...
Open Collections
UBC Theses and Dissertations
UBC Theses and Dissertations
Machine learning-based algorithms design for network slicing, federated learning, and 360° video streaming in wireless systems Setayesh, Mehdi
Abstract
The next generation of wireless systems aims to provide services with higher data rates, greater reliability, and lower latency compared to their predecessors (e.g., the fifth generation (5G) wireless systems). To facilitate the support of such services, new emerging technologies, such as edge intelligence and terahertz (THz) band communication, will be incorporated into wireless systems. Additionally, network slicing technology is inherited from 5G to enable the support of services with diverse quality of service (QoS) requirements within a shared network infrastructure. To fully harness the potential of these technologies and improve the performance of wireless systems, it is necessary to employ intelligent resource allocation algorithms using machine learning (ML) techniques within such systems. ML-based algorithms can adapt well to the network dynamics and can provide desired solutions in a timely manner. In this thesis, we employ various ML techniques to optimize wireless system performance for three specific problems. First, we consider a radio resource slicing problem to support enhanced mobile broadband (eMBB) and ultra-reliable low-latency communication (URLLC) services in a shared radio access network (RAN). To solve this multi-timescale problem, we propose a hierarchical deep learning framework. Second, we study federated learning (FL) algorithm as an enabler for edge intelligence and aim to improve its performance under heterogeneous data and device settings. To tackle this problem, we propose a personalized FL algorithm with optimized masking vectors called PerFedMask. Third, we consider 360° video streaming in a multi-user THz wireless system with multiple multi-antenna access points (APs). We propose a content-based viewport prediction framework to determine which video tiles should be sent to the users. Additionally, we propose a hierarchical deep reinforcement learning (DRL) framework to optimize the bitrate selection of the video tiles and the beamforming vectors at the APs. Simulation results show that compared with the benchmarks, our proposed ML-based approaches can achieve a higher aggregate throughput in the RAN slicing problem, a higher test accuracy with a lower average number of trainable parameters for FL under heterogeneous settings, and a higher quality of experience (QoE) for the users watching 360° videos in a THz wireless system.
Item Metadata
Title |
Machine learning-based algorithms design for network slicing, federated learning, and 360° video streaming in wireless systems
|
Creator | |
Supervisor | |
Publisher |
University of British Columbia
|
Date Issued |
2024
|
Description |
The next generation of wireless systems aims to provide services with higher data rates, greater reliability, and lower latency compared to their predecessors (e.g., the fifth generation (5G) wireless systems). To facilitate the support of such services, new emerging technologies, such as edge intelligence and terahertz (THz) band communication, will be incorporated into wireless systems. Additionally, network slicing technology is inherited from 5G to enable the support of services with diverse quality of service (QoS) requirements within a shared network infrastructure. To fully harness the potential of these technologies and improve the performance of wireless systems, it is necessary to employ intelligent resource allocation algorithms using machine learning (ML) techniques within such systems. ML-based algorithms can adapt well to the network dynamics and can provide desired solutions in a timely manner. In this thesis, we employ various ML techniques to optimize wireless system performance for three specific problems. First, we consider a radio resource slicing problem to support enhanced mobile broadband (eMBB) and ultra-reliable low-latency communication (URLLC) services in a shared radio access network (RAN). To solve this multi-timescale problem, we propose a hierarchical deep learning framework. Second, we study federated learning (FL) algorithm as an enabler for edge intelligence and aim to improve its performance under heterogeneous data and device settings. To tackle this problem, we propose a personalized FL algorithm with optimized masking vectors called PerFedMask. Third, we consider 360° video streaming in a multi-user THz wireless system with multiple multi-antenna access points (APs). We propose a content-based viewport prediction framework to determine which video tiles should be sent to the users. Additionally, we propose a hierarchical deep reinforcement learning (DRL) framework to optimize the bitrate selection of the video tiles and the beamforming vectors at the APs. Simulation results show that compared with the benchmarks, our proposed ML-based approaches can achieve a higher aggregate throughput in the RAN slicing problem, a higher test accuracy with a lower average number of trainable parameters for FL under heterogeneous settings, and a higher quality of experience (QoE) for the users watching 360° videos in a THz wireless system.
|
Genre | |
Type | |
Language |
eng
|
Date Available |
2024-04-19
|
Provider |
Vancouver : University of British Columbia Library
|
Rights |
Attribution-NonCommercial-NoDerivatives 4.0 International
|
DOI |
10.14288/1.0441439
|
URI | |
Degree | |
Program | |
Affiliation | |
Degree Grantor |
University of British Columbia
|
Graduation Date |
2024-05
|
Campus | |
Scholarly Level |
Graduate
|
Rights URI | |
Aggregated Source Repository |
DSpace
|
Item Media
Item Citations and Data
Rights
Attribution-NonCommercial-NoDerivatives 4.0 International