Machine Learning-Driven Analysis of User Bandwidth Allocation and Performance in 5G Network

Abstract

Mobile devices have been an integral part of society today, with increasing reliance on wireless communication and mobile networks. Researchers have developed various methods to increase network users while maintaining the Quality of Experience (QoE) under the 5G Network. Various resource allocation methods, such as Quality of Service (QoS) priorities and network slicing, were proposed to improve the users’ QoE. The network-centric resource allocation algorithms allocate resources to specific services to maintain QoE. However, these network-centric algorithms failed to deliver QoE when the network was overloaded. Therefore, a novel user-priority resource allocation (UPRA) algorithm is proposed in this work. UPRA is compared to the current Network Slicing 5G architecture. The UPRA algorithm divides users into three separate priorities depending on the network congestion. Resources are split among the priorities, as each priority will have slices for services. Furthermore, the Enhanced-UPRA (EUPRA) algorithm is proposed for the dynamic allocation of users in each priority. Finally, Machine Learning (ML) (Q-Learning (QL) and Deep Q-Network (DQN)) models are applied on the EUPRA algorithm to dynamically configure resource allocation at various network states. The extensive simulation results showed a 12.73% improvement in QoE from the network slicing 5G algorithm when applying the proposed EUPRA-DQN algorithm when the 5G network is overloaded.