基于复数向量余弦相似度KNN和深度度量学习的高精度无源室内定位  

High-Precision Passive Indoor Localization Based on Complex Vector Cosine Similarity KNN and Deep Metric Learning

作　　者：何之源 张志本[2] 沈琼霞[2] 巩江超 王德胜[1] HE Zhiyuan;ZHANG Ziben;SHENG Qiongxia;GONG Jiangchao;WANG Desheng(Huazhong University of Science and Technology,Wuhan 430074,China;Fiberhome Telecommunication Technologies Co.,Ltd.,Wuhan 430205,China)

机构地区：[1]华中科技大学,湖北武汉430074 [2]烽火通信科技股份有限公司,湖北武汉430205

出　　处：《移动通信》2024年第8期77-84,共8页Mobile Communications

基　　金：国家自然科学基金“异构无线密集网络高维资源智能部署研究”(62071192);湖北省重点研发计划项目“6G高维资源智能部署与高效大规模接入关键技术”(2022BAA006)。

摘　　要：AI及深度学习的发展为未来6G实现高精度室内定位系统提供了新思路。通过分析CSI数据的物理特性,提出了一种基于复数向量余弦相似度的改进KNN算法,显著提升了无源定位的性能。进一步地,采用度量学习方法,设计了Structured Embedding Loss损失函数,并引入Softmax Structure Loss优化神经网络模型,实现了端到端的高效训练和推理。实验结果表明,这些创新方法显著提高了无源定位的精度和鲁棒性,定位准确率和Macro-F1评分分别达到99.15%和99.1%,为无源定位、无线信号处理等领域提供了新的研究视角和技术路径。The development of AI and deep learning has opened up new avenues for achieving high-precision indoor localization systems in future 6G networks.By analyzing the physical characteristics of CSI data,this paper proposes an improved KNN algorithm based on complex vector cosine similarity,significantly enhancing the passive localization performance.Furthermore,a metric learning approach is adopted by designing a structured embedding loss function,and a softmax structure loss is introduced to optimize the neural network model,enabling efficient end-to-end training and inference.Experimental results demonstrate that these innovative methods significantly improve the accuracy and robustness of passive localization,achieving a localization accuracy and Macro-F1 score of 99.15%and 99.1%,respectively,offering new research perspectives and technical paths for fields of passive localization and wireless signal processing.

关 键 词：深度学习 度量学习 无源定位 信道状态信息 机器学习 

分 类 号：TN929.5[电子电信—通信与信息系统]