Citation

Farouk, Mohamed Mohsen and Chung, Gwo Chin and Roslee, Mardeni and Lee, It Ee and Pang, Wai Leong (2026) Hybrid Reinforcement Learning-Enabled Scheduler for 5G Burst Traffic. International Journal of Technology, 17 (3). p. 866. ISSN 2086-9614

Text IJTech_EECE-8208_Hybrid-Reinforcement-LearningEnabled-Scheduler-for.pdf - Published Version Restricted to Repository staff only Download (2MB)

Abstract

Resource allocation in wireless networks is inherently complex, a problem intensified in 5G by heterogeneous traffic classes and stringent quality of service (QoS) requirements. This challenge poses significant difficulties for traditional scheduling methods. In this study, we address these limitations using novel hybrid reinforcement learning (RL) architectures evaluated in a dynamic and realistic network environment. We designed and implemented three hybrid RL algorithms: Asynchronous Advantage Actor-Critic integrated with Proximal Policy Optimization (A3C-PPO), A3C with Proximal Policy Optimization and Session Persistence (A3CPPO-Persistent), and A3C with Twin Delayed Deep Deterministic Policy Gradients (A3C-TD3). These were compared against baseline A3C and Advantage Actor-Critic (A2C) approaches, as well as traditional proportional fair (PF), maximum rate (MR), and Round Robin (RR) schedulers. Simulations were performed in a challenging multicell environment with mobile user equipment and bursty traffic flows across four network traffic types: ultra-low latency (ULL), voice over IP (VoIP), vehicle-to-everything (V2X), and video streaming. Our hybrid RL schedulers showed promising performance in this highly dynamic setting, with A3C-PPO achieving the most balanced overall results, exhibiting 25%–40% lower average jitter and over four times higher packet delivery ratio (PDR) than traditional schedulers under heavy loads. Our results indicate that hybrid RL methods, particularly A3C+PPO, can provide resilient adaptive scheduling that can outperform both conventional techniques and standard RL algorithm models in realistic 5G networks.

Item Type:	Article
Uncontrolled Keywords:	5G
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 Rosnani Abd Wahab
Date Deposited:	30 Jun 2026 08:42
Last Modified:	30 Jun 2026 08:42
URII:	http://shdl.mmu.edu.my/id/eprint/16163

Downloads

Edit (login required)