Integrating Reinforcement Learning into M/M/1/K Retry Queueing Models for 6G Applications

Area of Science:

• Wireless Communication
• Artificial Intelligence
• Network Engineering

Background:

• Next-generation wireless networks, particularly 6G operating at Terahertz frequencies, face challenges in sustainable, efficient, and fair resource allocation.
• Existing queue management systems struggle to adapt to dynamic traffic conditions and high-speed demands.

Purpose of the Study:

• To develop and evaluate an intelligent queue management system for 6G networks using reinforcement learning.
• To minimize network delays, reduce energy consumption, and ensure equitable resource access.

Main Methods:

• Integration of Deep Q-Networks (DQN), a reinforcement learning algorithm, into a 6G Retrial Queueing System (RQS).
• Extensive simulations were conducted to analyze performance under varying arrival rates, queue sizes, and reward scaling factors.
• Singular Value Decomposition (SVD) was used to analyze the agent's learning process and adaptation.

Main Results:

• The proposed 6G-RQS model with DQN significantly enhances queue management effectiveness.
• The system demonstrated improved performance by increasing the number of served mobile terminals, even under high traffic demands.
• SVD analysis confirmed effective learning and adaptation by the reinforcement learning agent.

Conclusions:

• Reinforcement learning-based queue management is a viable and promising solution for addressing 6G network challenges.
• The intelligent 6G-RQS offers a dynamic and adaptive approach to optimize performance in high-speed communication networks.
• This research paves the way for more efficient and equitable wireless network resource allocation.