Intelligent multicast routing method based on multi-agent deep reinforcement learning in SDWN.

Area of Science:

• Computer Science
• Networking
• Artificial Intelligence

Background:

• Traditional wireless networks struggle with dynamic state information for multicast routing, impacting Quality of Service (QoS).
• High device density in wireless environments exacerbates challenges for efficient multicast communication.

Purpose of the Study:

• To propose a new multicast routing method, MADRL-MR, for software-defined wireless networking (SDWN) environments.
• To improve throughput, reduce delay, and enhance multicast routing efficiency in dense wireless networks.

Main Methods:

• Utilizing Software-Defined Wireless Networking (SDWN) for flexible network configuration and global state information acquisition via traffic matrices.
• Employing multiagent deep reinforcement learning (MADRL) to divide multicast routing into cooperative subproblems.
• Designing agent state and action spaces based on traffic, multicast tree status, and AP nodes, with a novel single-hop action strategy and a four-state reward function.

Main Results:

• MADRL-MR demonstrates superior performance over existing algorithms in throughput, delay, and packet loss rate.
• The method successfully establishes more intelligent multicast routes in dynamic network conditions.
• Decentralized training with transfer learning accelerates convergence and improves agent adaptability.

Conclusions:

• MADRL-MR offers a robust and intelligent solution for multicast routing in dense, dynamic wireless networks.
• The integration of SDWN and MADRL provides significant advantages for network performance and QoS.
• The proposed approach effectively addresses the limitations of traditional multicast routing methods.