Citation

Farouk, Mohamed Mohsen (2025) QOS provisioning using hybrid reinforcement learning for 5G mobile networks. Masters thesis, Multimedia University.

Full text not available from this repository.

Abstract

The 5G wireless network standard put forth through the 3rd Generation Partnership Project (3GPP) promises major improvements over older standards, such as higher speeds, lower latency, better reliability, and more. These advancements enable numerous new applications, a variety of usage scenarios, and device capabilities. All these factors contribute to the increased complexity of resource management in 5G. This added complexity pushes researchers to explore new methods for resource management, including machine learning (ML)-enabled network schedulers. In this case, reinforcement learning (RL) enables a more flexible and autonomous approach, as it relies on a reward-driven approach built on exploration through trial and error. This study proposed an RL-enabled network scheduler for multi-cell 5G networks. It was designed to support mobile user equipment (UEs) and dynamic, bursty traffic flows without compromising QoS. The proposed design implemented a scheduler that processed a large set of observation parameters and a comprehensive multi-part reward function to impact resource allocation at multiple levels. Furthermore, the flexibility of the proposed approach allowed for different RL algorithms to be tested in the same proposed framework, which included the proposed hybrid RL algorithms: Asynchronous Advantage Actor-Critic (A3C) integrated with proximal policy optimisation (A3C-PPO), A3C-PPO with session persistence (A3C-PPO-Persistent), and A3C with Twin Delayed Deep Deterministic Policy Gradients (A3C-TD3). The proposed algorithms’ performance was also compared with three traditional schedulers: proportionally fair (PF), maximum rate (MR), and round robin (RR). The simulations revealed a complex performance landscape where the proposed A3C-PPO scheduler achieved the most consistent performance improvements. It also maintained more than 99% packet delivery ratio (PDR) where traditional schedulers catastrophically failed, including over four times higher PDR than traditional schedulers at 50 UEs. However, this performance came with significant trade-offs in jitter and throughput efficiency.

Item Type:	Thesis (Masters)
Additional Information:	Call No.: TK5103.25 .M64 2025
Uncontrolled Keywords:	5G mobile communication systems
Subjects:	T Technology > TK Electrical engineering. Electronics Nuclear engineering > TK5101-6720 Telecommunication. Including telegraphy, telephone, radio, radar, television
Divisions:	Faculty of Artificial Intelligence & Engineering (FAIE)
Depositing User:	Ms Nurul Iqtiani Ahmad
Date Deposited:	10 Jun 2026 04:40
Last Modified:	10 Jun 2026 04:40
URII:	http://shdl.mmu.edu.my/id/eprint/16101

Downloads

Edit (login required)