OverGNN Assisted Power Allocation for Heterogeneous Ultra-Dense Networks

doi:10.1109/WCSP58612.2023.10405334

	OverGNN Assisted Power Allocation for Heterogeneous Ultra-Dense Networks
	Sisi Lin 1; Mengyuan Lee 2; Qimei Chen 1; Dingzhu Wen3 ; Wen Du 4; Zhidong He 4
	2023-11-04
会议录名称	2023 INTERNATIONAL CONFERENCE ON WIRELESS COMMUNICATIONS AND SIGNAL PROCESSING (WCSP)
页码	152-157
发表状态	已发表
DOI	10.1109/WCSP58612.2023.10405334
摘要	With the development of wireless communications, heterogeneous ultra-dense networks (HUDNs) have emerged timely to meet the requirements of massive connectivity, high data rate, and low latency in the 5G era. Nevertheless, HUDN indicates large-scale and high-density scenarios, which usually lead to a high-complexity and non-convex NP-hard resource allocation problem. Graph neural network (GNN) is regarded as a promising approach to deal with the above issue. However, naive GNNs fail to model the high-order interactions among the network nodes and are prone to over-smoothing after multi-layer convolution. Hence, we construct a heterogeneous GNN with a novel high-dimensional computation structure (namely OverGNN) for the power allocation problem under HUDNs in this work. Particularly, OverGNN enables nodes directly interact with high-order neighbors and extract abundant graph topological information, which can facilitate effective feature aggregation among nodes as well as alleviate the over-smoothing problem. Based on this fact, an efficient message passing scheme for user equipments (UEs) under the same base station (BS) is developed to approximate the optimal power allocation strategy for maximizing system throughput. In addition, we propose an unsupervised approach to train the GNN model that can reduce the computation complexity of training and enhance the scalability of our proposed method. Numerical results verify the effectiveness of the proposed OverGNN and demonstrate its advantages over the benchmarks. © 2023 IEEE.
关键词	Heterogeneous network ultra-dense network power allocation GNN high-order message passing
会议名称	15th IEEE International Conference on Wireless Communications and Signal Processing, WCSP 2023
会议地点	Hangzhou, China
会议日期	2-4 Nov. 2023
URL	查看原文
收录类别	EI
语种	英语
出版者	Institute of Electrical and Electronics Engineers Inc.
EI入藏号	20240915634359
原始文献类型	Conference article (CA)
来源库	IEEE
文献类型	会议论文
条目标识符	https://kms.shanghaitech.edu.cn/handle/2MSLDSTB/352516
专题	信息科学与技术学院_PI研究组_文鼎柱组
作者单位	1.Wuhan University 2.Zhejiang University 3.ShanghaiTech University 4.DS Information Technology Co., Ltd.
推荐引用方式 GB/T 7714	Sisi Lin,Mengyuan Lee,Qimei Chen,et al. OverGNN Assisted Power Allocation for Heterogeneous Ultra-Dense Networks[C]:Institute of Electrical and Electronics Engineers Inc.,2023:152-157.