A Scalable Distributed Link Management Method for Massive IoT With Synchronous Message Passing Neural Network

作     者：Gou, Haosong Du, Pengfei Wang, Xidian Zhang, Gaoyi Zhai, Daosen 

作者机构：China Mobile Grp Sichuan Co Ltd Chengdu 610041 Peoples R China Xihua Univ Engn Res Ctr Intelligent Airground Integrated Vehi Minist Educ Chengdu 610039 Peoples R China Yulin Internet Things Collaborat Innovat Res Inst Yulin 719099 Peoples R China 

出 版 物：《IEEE TRANSACTIONS ON NETWORK SCIENCE AND ENGINEERING》 (IEEE Trans. Netw. Sci. Eng.)

年 卷 期：2025年第12卷第2期

页      面：750-762页

核心收录：

学科分类：0808[工学-电气工程] 08[工学] 0701[理学-数学] 

基　　金：Natural Science Foundation of Sichuan Province [2023NSFSC1377] National Natural Science Foundation of China China Mobile Communications Group Sichuan Company, Ltd Project of new technology research [R2411BMP] Innovation Capability Support Plan of Shaanxi Province [2024ZC-KJXX-077] 

主　　题：Training Internet of Things Message passing Graph neural networks Interference Signal to noise ratio Transmitters Scalability Power control Costs Link management graph neural network power control admission control distributed algorithm 

摘      要：The development of the next generation ubiquitous network has put forward higher requirements for the connection density of communication devices, which has led to a lot of research on link management. However, with the expansion of network scale, the weaknesses of the existing algorithms in computing efficiency, performance, and realizability have become prominent. The emerging graph neural network (GNN) provides a new way to solve this problem. In order to make full use of the broadcast feature of wireless communication, we design a cross-domain distributed GNN structure (named as synchronous message passing neural network (SynMPNN)) combining the measurable index of the actual scene with message passing mechanism. This new GNN structure and the additional input feature dimension (i.e., SINR) work together to provide more comprehensive information for network training. After the initial deployment of the power decision from SynMPNN, we select some links to shut down and others to reduce their transmit power to further improve the system performance and save energy. Simulation results show that our proposed method under distributed execution conditions reaches 83.1% performance of the centralized method. In addition, the discussion on scalability suggests that in order to save training cost, small-scale scenes with the same density can be selected for training in the application of large-scale scenes.