Multi-Agent Dynamic Resource Allocation in 6G in-X Subnetworks with Limited Sensing Information

Area of Science:

• Wireless communication networks
• Mobile computing
• Network optimization

Background:

• Dense deployments of 6G mobile in-X subnetworks (inXSs) present challenges for dynamic resource selection.
• Autonomous inXSs require efficient resource management based on local information to minimize signaling overhead.

Purpose of the Study:

• To investigate dynamic resource selection in dense 6G inXS deployments.
• To address the multi-objective optimization problem of maximizing minimum capacity per inXS while minimizing intra-subnetwork signaling.
• To develop autonomous resource selection methods for inXSs.

Main Methods:

• Formulated resource selection as a multi-objective optimization problem.
• Developed a multi-agent Q-learning (MAQL) method using limited sensing information (SI).
• Proposed a rule-based algorithm, Q-Heuristics, for resource selection.
• Focused simulations on joint channel and transmit power selection.

Main Results:

• MAQL demonstrated fast convergence with appropriate Q-learning parameters, even with quantized SI.
• The MAQL approach significantly outperformed baseline heuristics in performance and robustness to delays.
• Q-Heuristics showed comparable performance to greedy methods and high robustness to sensing intervals, quantization, and switching delays.

Conclusions:

• MAQL offers an efficient and robust solution for dynamic resource selection in 6G inXSs.
• Q-Heuristics provides a robust alternative for resource selection with similar performance to baseline methods.
• Both proposed methods effectively manage resources in dense, autonomous subnetwork environments.