Publication: Reinforcement Learning Model Selection for Resource Allocation and Subcarrier Spacing Optimization in 5G Sliced Spectrum Networks
No Thumbnail Available
Date
2024
Authors
Samidi F.S.
Radzi N.A.M.
Aripin N.M.
Journal Title
Journal ISSN
Volume Title
Publisher
Institute of Electrical and Electronics Engineers Inc.
Abstract
The evolution of 5G technology presents unique challenges and opportunities for optimizing network performance through advanced techniques such as Reinforcement Learning (RL). Effective 5G network management requires precise optimization of subcarrier spacing and uplink-downlink (UL-DL) allocation, yet selecting the optimal RL model for this task remains a significant challenge. This paper evaluates the performance of RL models within the RaSSo framework for optimizing 5G networks, aiming to identify the most suitable RL approach. The primary challenge addressed is determining which RL model, in our case, Proximal Policy Optimization (PPO) or Advantage Actor-Critic (A2C), performs better in managing the dynamic and complex 5G environment. The study compares these models in terms of episode reward, entropy loss, value loss, and frames per second (FPS). The results indicate that PPO's stable performance, efficient learning process, and robust adaptation to dynamic conditions make it a more suitable choice for real-time 5G network management. The results indicate that PPO's stable performance, efficient learning process, and robust adaptation to dynamic conditions make it a more suitable choice for real-time 5G network management. Overall, this paper contributes to the field by offering a thorough comparison of RL models tailored for the nuanced demands of 5G network optimization, providing valuable insights for future advancements in this area. ? 2024 IEEE.
Description
Keywords
Network performance , Queueing networks , Reinforcement learning , Resource allocation , Dynamic condition , Efficient learning , Learning process , Networks management , Optimisations , Performance , Real- time , Reinforcement learning models , Stable performance , Subcarrier spacing , 5G mobile communication systems