Machine learning approach to slice admission control in 5G wireless network
| dc.contributor.advisor | Ramotsoela, Daniel | |
| dc.contributor.author | Ojijo, Mourice | |
| dc.date.accessioned | 2026-01-20T08:38:37Z | |
| dc.date.available | 2026-01-20T08:38:37Z | |
| dc.date.issued | 2025 | |
| dc.date.updated | 2026-01-15T07:58:06Z | |
| dc.description.abstract | The advent of fifth-generation (5G) wireless communication has introduced a paradigm shift in how cellular networks operate and how network resources are allocated. As data networks become increasingly dynamic and complex, resource allocation can not be uniformly applied to all network users; rather, it should be implemented through a user-centric construction of virtual functions , commonly known as network slices. However, network slicing can be complemented by a slice admission control (SAC) process that permits only those slice requests that satisfy quality-of-service (QoS) requirements. Furthermore, automated and intelligent networking approaches can be employed to enhance SAC and optimize key objectives such as revenue, fairness, scheduling efficiency, and network resilience. This thesis investigates SAC in a resource-constrained, inter-domain 5G network, with a focus on improving the utility, fairness, scheduling, and network resilience of infrastructure providers (InPs). Firstly, this thesis develops an inter-domain resource allocation framework that elucidates the inherent non-linearity of its formulation. The resulting formulation is presented as a mixed-integer non-linear programming (MINLP) problem and is proven to be NP-hard. In this context, the study conducts a comprehensive analytical comparison of various feasible solution approaches to address this problem, including branch and bound (BnB), successive convex approximation (SCA), the alternating direction method of multipliers (ADMM), heuristic methods, genetic algorithms (GA), and machine learning (ML) techniques. Secondly, this thesis introduces a multi-server, multi-queue resource scheduling framework aimed at accurately predicting costs and resource availability over a 24-hour period. By leveraging virtualized inter-domain resource blocks, the transient probabilities of queues are derived. The investigation finds that the predictions of the deep Q-learning (DQL) agent generally fall within an acceptable average variation of 6%. Thirdly, this study investigates the impact of integrating slice admission control (SAC) with an auction-based game, which is double-ended bidding mechanism that allow both network resource buyers and sellers to place preference so that resources can only be allocated to the most deserving bidder, this aim to maximize overall utility and enhance fairness. The thesis introduces inter-domain resource models for n-class zoned 5G slices. By increasing the probability of admission during periods of resource availability, the findings demonstrate that the reinforcement learning agent improves long-term utility and fairness by 68.2%. Finally, this thesis proposes a novel sequential twin-actor critic (STAC) method that optimizes a two-stage action process—namely, slice admission control (SAC) and resilience maintenance—by adjusting network resource blocks (RBs) to maximize overall utility. Additionally, the probability of slice acceptance is evaluated and compared with that of similar schemes. Given the anticipated density of up to one million devices per square kilometer, this study provides supplementary insights into SAC in an anomalous multi-node environment by analyzing data from admitted slice requests to detect any irregular patterns. The results demonstrate that the adopted reinforcement learning (RL) scheme outperforms the compared approaches. | |
| dc.identifier.apacitation | Ojijo, M. (2025). <i>Machine learning approach to slice admission control in 5G wireless network</i>. (). University of Cape Town ,Faculty of Engineering and the Built Environment ,Department of Electrical Engineering. Retrieved from http://hdl.handle.net/11427/42618 | en_ZA |
| dc.identifier.chicagocitation | Ojijo, Mourice. <i>"Machine learning approach to slice admission control in 5G wireless network."</i> ., University of Cape Town ,Faculty of Engineering and the Built Environment ,Department of Electrical Engineering, 2025. http://hdl.handle.net/11427/42618 | en_ZA |
| dc.identifier.citation | Ojijo, M. 2025. Machine learning approach to slice admission control in 5G wireless network. . University of Cape Town ,Faculty of Engineering and the Built Environment ,Department of Electrical Engineering. http://hdl.handle.net/11427/42618 | en_ZA |
| dc.identifier.ris | TY - Thesis / Dissertation AU - Ojijo, Mourice AB - The advent of fifth-generation (5G) wireless communication has introduced a paradigm shift in how cellular networks operate and how network resources are allocated. As data networks become increasingly dynamic and complex, resource allocation can not be uniformly applied to all network users; rather, it should be implemented through a user-centric construction of virtual functions , commonly known as network slices. However, network slicing can be complemented by a slice admission control (SAC) process that permits only those slice requests that satisfy quality-of-service (QoS) requirements. Furthermore, automated and intelligent networking approaches can be employed to enhance SAC and optimize key objectives such as revenue, fairness, scheduling efficiency, and network resilience. This thesis investigates SAC in a resource-constrained, inter-domain 5G network, with a focus on improving the utility, fairness, scheduling, and network resilience of infrastructure providers (InPs). Firstly, this thesis develops an inter-domain resource allocation framework that elucidates the inherent non-linearity of its formulation. The resulting formulation is presented as a mixed-integer non-linear programming (MINLP) problem and is proven to be NP-hard. In this context, the study conducts a comprehensive analytical comparison of various feasible solution approaches to address this problem, including branch and bound (BnB), successive convex approximation (SCA), the alternating direction method of multipliers (ADMM), heuristic methods, genetic algorithms (GA), and machine learning (ML) techniques. Secondly, this thesis introduces a multi-server, multi-queue resource scheduling framework aimed at accurately predicting costs and resource availability over a 24-hour period. By leveraging virtualized inter-domain resource blocks, the transient probabilities of queues are derived. The investigation finds that the predictions of the deep Q-learning (DQL) agent generally fall within an acceptable average variation of 6%. Thirdly, this study investigates the impact of integrating slice admission control (SAC) with an auction-based game, which is double-ended bidding mechanism that allow both network resource buyers and sellers to place preference so that resources can only be allocated to the most deserving bidder, this aim to maximize overall utility and enhance fairness. The thesis introduces inter-domain resource models for n-class zoned 5G slices. By increasing the probability of admission during periods of resource availability, the findings demonstrate that the reinforcement learning agent improves long-term utility and fairness by 68.2%. Finally, this thesis proposes a novel sequential twin-actor critic (STAC) method that optimizes a two-stage action process—namely, slice admission control (SAC) and resilience maintenance—by adjusting network resource blocks (RBs) to maximize overall utility. Additionally, the probability of slice acceptance is evaluated and compared with that of similar schemes. Given the anticipated density of up to one million devices per square kilometer, this study provides supplementary insights into SAC in an anomalous multi-node environment by analyzing data from admitted slice requests to detect any irregular patterns. The results demonstrate that the adopted reinforcement learning (RL) scheme outperforms the compared approaches. DA - 2025 DB - OpenUCT DP - University of Cape Town KW - 5G KW - Wireless network LK - https://open.uct.ac.za PB - University of Cape Town PY - 2025 T1 - Machine learning approach to slice admission control in 5G wireless network TI - Machine learning approach to slice admission control in 5G wireless network UR - http://hdl.handle.net/11427/42618 ER - | en_ZA |
| dc.identifier.uri | http://hdl.handle.net/11427/42618 | |
| dc.identifier.vancouvercitation | Ojijo M. Machine learning approach to slice admission control in 5G wireless network. []. University of Cape Town ,Faculty of Engineering and the Built Environment ,Department of Electrical Engineering, 2025 [cited yyyy month dd]. Available from: http://hdl.handle.net/11427/42618 | en_ZA |
| dc.language.iso | en | |
| dc.language.rfc3066 | eng | |
| dc.publisher.department | Department of Electrical Engineering | |
| dc.publisher.faculty | Faculty of Engineering and the Built Environment | |
| dc.publisher.institution | University of Cape Town | |
| dc.subject | 5G | |
| dc.subject | Wireless network | |
| dc.title | Machine learning approach to slice admission control in 5G wireless network | |
| dc.type | Thesis / Dissertation | |
| dc.type.qualificationlevel | Doctoral | |
| dc.type.qualificationlevel | PhD |