Loading...
5G new radio and fog computing scalability and QoS management
Author(s)
Pana, Vuyo
Date Issued
2024
Type
Thesis
Publisher
Cape Peninsula University of Technology
Abstract
This thesis investigates the advancements in fifth-generation (5G) mobile technology, with a
particular focus on the 5G Fog Radio Access Network (5G-FRAN) architecture and resource
optimization. The proliferation of Internet of Things (IoT) deployments has heightened the
demand for latency-efficient applications positioned closer to end-users, necessitating a critical
evaluation of resource management in 5G networks.
The research begins with a comprehensive literature review of Cloud Radio Access Network
(CRAN) architectures and their variations. CRAN is pivotal for reducing operational and
maintenance costs for mobile network operators. However, the exponential growth of IoT
devices, surpassing 900 million by 2023, exposes the limitations of CRAN’s centralized
network structure, particularly its vulnerability to fronthaul congestion. To address these
challenges, the 5G-FRAN architecture offers significant enhancements, including improved
mobility, optimized radio resource allocation, superior service quality, and low latency. Despite
these advancements, challenges such as resource allocation, end-to-end latency, and energy
efficiency persist. Consequently, this thesis proposes machine learning-based resource
management schemes to mitigate these challenges and meet industrial requirements
effectively.
Thus, a novel cell association technique is developed to manage bandwidth in 5G-FRANs by
accounting for subscriber mobility. The proposed approach employs probability distributions of
time intervals and cell identifications to construct a Hidden Semi-Markov Model (HsMM) that
predicts the optimal next cell for meeting downstream rate requirements. The model addresses
critical issues, including ping-pong effects, missing data, and restrictive wireless environments,
thereby enhancing user traffic prediction. Results demonstrate that the HsMM-based method
achieves high accuracy with respect to observation time intervals and significantly improves
user satisfaction compared to existing systems.
Furthermore, Eigen decomposition (ED)-based feature extraction techniques, specifically
Principal Component Analysis (PCA) and Linear Discriminant Analysis (LDA), which are
applied to attributes derived from the dataset. The extracted features are then utilized in
machine learning models to detect and classify service segments. Hence, in the thesis a
detailed description of the dataset and the implementation of the ED-based feature extraction
techniques is provided. The workflow of the machine learning models is illustrated using block
diagrams, followed by an in-depth explanation of the experimental design. Experiments were
conducted using a range of machine learning algorithms, including Decision Tree (DT), Support Vector Machines (SVM), K-Nearest Neighbours (K-NN), Random Forest (RF),
Artificial Neural Networks (ANN), and their combinations with PCA and LDA. These models
were tested on fog computing application datasets featuring multi-class classification scenarios
and noise-injected environments.
Thereafter, a thorough analysis of the experimental results, focusing on the impact of
dimensionality reduction on model accuracy, efficiency, and resilience to noise in various
machine learning configurations.
This thesis makes several notable contributions. It provides an in-depth analysis of the latest
5G network architectures and demonstrates the efficacy of HsMM in optimizing limited
spectrum resources using machine learning techniques. Additionally, the study introduces and
evaluates two ED-based feature extraction techniques (PCA and LDA) across multiple
machine learning models, significantly reducing computational complexity while improving
model performance. These techniques enable efficient processing of large datasets,
enhancing the accuracy, reliability, and scalability of machine learning models in 5G-FRAN
applications.
particular focus on the 5G Fog Radio Access Network (5G-FRAN) architecture and resource
optimization. The proliferation of Internet of Things (IoT) deployments has heightened the
demand for latency-efficient applications positioned closer to end-users, necessitating a critical
evaluation of resource management in 5G networks.
The research begins with a comprehensive literature review of Cloud Radio Access Network
(CRAN) architectures and their variations. CRAN is pivotal for reducing operational and
maintenance costs for mobile network operators. However, the exponential growth of IoT
devices, surpassing 900 million by 2023, exposes the limitations of CRAN’s centralized
network structure, particularly its vulnerability to fronthaul congestion. To address these
challenges, the 5G-FRAN architecture offers significant enhancements, including improved
mobility, optimized radio resource allocation, superior service quality, and low latency. Despite
these advancements, challenges such as resource allocation, end-to-end latency, and energy
efficiency persist. Consequently, this thesis proposes machine learning-based resource
management schemes to mitigate these challenges and meet industrial requirements
effectively.
Thus, a novel cell association technique is developed to manage bandwidth in 5G-FRANs by
accounting for subscriber mobility. The proposed approach employs probability distributions of
time intervals and cell identifications to construct a Hidden Semi-Markov Model (HsMM) that
predicts the optimal next cell for meeting downstream rate requirements. The model addresses
critical issues, including ping-pong effects, missing data, and restrictive wireless environments,
thereby enhancing user traffic prediction. Results demonstrate that the HsMM-based method
achieves high accuracy with respect to observation time intervals and significantly improves
user satisfaction compared to existing systems.
Furthermore, Eigen decomposition (ED)-based feature extraction techniques, specifically
Principal Component Analysis (PCA) and Linear Discriminant Analysis (LDA), which are
applied to attributes derived from the dataset. The extracted features are then utilized in
machine learning models to detect and classify service segments. Hence, in the thesis a
detailed description of the dataset and the implementation of the ED-based feature extraction
techniques is provided. The workflow of the machine learning models is illustrated using block
diagrams, followed by an in-depth explanation of the experimental design. Experiments were
conducted using a range of machine learning algorithms, including Decision Tree (DT), Support Vector Machines (SVM), K-Nearest Neighbours (K-NN), Random Forest (RF),
Artificial Neural Networks (ANN), and their combinations with PCA and LDA. These models
were tested on fog computing application datasets featuring multi-class classification scenarios
and noise-injected environments.
Thereafter, a thorough analysis of the experimental results, focusing on the impact of
dimensionality reduction on model accuracy, efficiency, and resilience to noise in various
machine learning configurations.
This thesis makes several notable contributions. It provides an in-depth analysis of the latest
5G network architectures and demonstrates the efficacy of HsMM in optimizing limited
spectrum resources using machine learning techniques. Additionally, the study introduces and
evaluates two ED-based feature extraction techniques (PCA and LDA) across multiple
machine learning models, significantly reducing computational complexity while improving
model performance. These techniques enable efficient processing of large datasets,
enhancing the accuracy, reliability, and scalability of machine learning models in 5G-FRAN
applications.
Additional information
Thesis (DEng (Electrical Engineering))--Cape Peninsula University of Technology, 2024
File(s)![Thumbnail Image]()
Loading...
Name
Pana, V_207191859.pdf
Size
3.71 MB
Format
Adobe PDF
Checksum
(MD5):79a9f89cac27afc720b24068bb850ac4