Admission Control Optimisation for QoS and QoE Enhancement in Future Networks

Perveen, Abida (2022) Admission Control Optimisation for QoS and QoE Enhancement in Future Networks. Doctoral thesis, Birmingham City University.

Abida Perveen PhD Thesis published_Final version_Submitted Feb 2022_Final Award Oct 2022 .pdf - Accepted Version

Download (5MB)


Recent exponential growth in demand for traffic heterogeneity support and the number of associated devices has considerably increased demand for network resources and induced numerous challenges for the networks, such as bottleneck congestion, and inefficient admission control and resource allocation. Challenges such as these degrade network Quality of Service (QoS) and user-perceived Quality of Experience (QoE). This work studies admission control from various perspectives. For example, two novel single-objective optimisation-based admission control models, Dynamica Slice Allocation and Admission Control (DSAAC) and Signalling and Admission Control (SAC), are presented to enhance future limited-capacity network Grade of Service (GoS), and for control signalling optimisation, respectively. DSAAC is an integrated model whereby a cost-estimation function based on user demand and network capacity quantifies resource allocation among users. Moreover, to maximise resource utility, adjustable minimum and maximum slice resource bounds have also been derived. In the case of user blocking from the primary slice due to congestion or resource scarcity, a set of optimisation algorithms on inter-slice admission control and resource allocation and adaptability of slice elasticity have been proposed.

A novel SAC model uses an unsupervised learning technique (i.e. Ranking-based clustering) for optimal clustering based on users’ homogeneous demand characteristics to minimise signalling redundancy in the access network. The redundant signalling reduction reduces the additional burden on the network in terms of unnecessary resource utilisation and computational time. Moreover, dynamically reconfigurable QoE-based slice performance bounds are also derived in the SAC model from multiple demand characteristics for clustered user admission to the optimal network. A set of optimisation algorithms are also proposed to attain efficient slice allocation and users’ QoE enhancement via assessing the capability of slice QoE elasticity. An enhancement of the SAC model is proposed through a novel multi-objective optimisation model named Edge Redundancy Minimisation and Admission Control (E-RMAC). A novel E-RMAC model for the first time considers the issue of redundant signalling between the edge and core networks. This model minimises redundant signalling using two classical unsupervised learning algorithms, K-mean and Ranking-based clustering, and maximises the efficiency of the link (bandwidth resources) between the edge and core networks.

For multi-operator environments such as Open-RAN, a novel Forecasting and Admission Control (FAC) model for tenant-aware network selection and configuration is proposed. The model features a dynamic demand-estimation scheme embedded with fuzzy-logic-based optimisation for optimal network selection and admission control. FAC for the first time considers the coexistence of the various heterogeneous cellular technologies (2G, 3G,4G, and 5G) and their integration to enhance overall network throughput by efficient resource allocation and utilisation within a multi-operator environment. A QoS/QoE-based service monitoring feature is also presented to update the demand estimates with the support of a forecasting modifier. he provided service monitoring feature helps resource allocation to tenants, approximately closer to the actual demand of the tenants, to improve tenant-acquired QoE and overall network performance. Foremost, a novel and dynamic admission control model named Slice Congestion and Admission Control (SCAC) is also presented in this thesis. SCAC employs machine learning (i.e. unsupervised, reinforcement, and transfer learning) and multi-objective optimisation techniques (i.e. Non-dominated Sorting Genetic Algorithm II ) to minimise bottleneck and intra-slice congestion. Knowledge transfer among requests in form of coefficients has been employed for the first time for optimal slice requests queuing. A unified cost estimation function is also derived in this model for slice selection to ensure fairness among slice request admission. In view of instantaneous network circumstances and load, a reinforcement learning-based admission control policy is established for taking appropriate action on guaranteed soft and best-effort slice requests admissions. Intra-slice, as well as inter-slice resource allocation, along with the adaptability of slice elasticity, are also proposed for maximising slice acceptance ratio and resource utilisation.

Extensive simulation results are obtained and compared with similar models found in the literature. The proposed E-RMAC model is 35% superior at reducing redundant signalling between the edge and core networks compared to recent work. The E-RMAC model reduces the complexity from O(U) to O(R) for service signalling and O(N) for resource signalling. This represents a significant saving in the uplink control plane signalling and link capacity compared to the results found in the existing literature. Similarly, the SCAC model reduces bottleneck congestion by approximately 56% over the entire load compared to ground truth and increases the slice acceptance ratio. Inter-slice admission and resource allocation offer admission gain of 25% and 51% over cooperative slice- and intra-slice-based admission control and resource allocation, respectively. Detailed analysis of the results obtained suggests that the proposed models can efficiently manage future heterogeneous traffic flow in terms of enhanced throughput, maximum network resources utilisation, better admission gain, and congestion control.

Item Type: Thesis (Doctoral)
2 February 2022Submitted
19 October 2022Accepted
Uncontrolled Keywords: Admission control optimisation, Network slicing, 5G-Open RAN, Resource management, Slice and Resource Allocation, Network Quality of Experience (QoE) and Quality of Service (QoS)
Subjects: CAH11 - computing > CAH11-01 - computing > CAH11-01-01 - computer science
Divisions: Doctoral Research College > Doctoral Theses Collection
Faculty of Computing, Engineering and the Built Environment > School of Computing and Digital Technology
Depositing User: Jaycie Carter
Date Deposited: 09 Mar 2023 16:19
Last Modified: 09 Mar 2023 16:19

Actions (login required)

View Item View Item


In this section...