Paper #5

Heterogeneous IoT User Association and Channel Resource Joint Scheduling Method Based on MADDQN and DMADDPG

Yankai Xie

Abstract: The large-scale access of multiple types of terminals in the heterogeneous IoT makes it difficult to balance system performance and scalability due to the strong coupling relationship between user association and channel resource scheduling. Existing deep reinforcement learning methods still have shortcomings in multi-agent collaborative decision-making, hybrid discrete-continuous action modeling, and dynamic environment adaptability. Therefore, this study proposes a joint scheduling method of user association and channel resources based on MADDQN and DMADDPG. This method decouples discrete user scheduling and continuous resource allocation, and achieves multi-agent collaborative optimization under a centralized training and distributed execution framework. The results showed that the total throughput of the research method reached 468.52 Mbps, which was 116.42 Mbps higher than the weighted minimum mean square error of 352.10 Mbps. When the user scale was expanded to 200, the non-compliance rate of the research method was only 16.82%, which was 32.30% lower than the DMADDPG algorithm, and still maintained an average rate of 2.48 Mbps. In summary, the proposed method has good robustness and scalability while improving system performance, and provides an effective solution for large-scale heterogeneous IoT resource scheduling.

Keywords: Channel resource allocation; Double deep Q-network; Heterogeneous Internet of Things; Multi-agent reinforcement learning; User association.