基于多域射频层析成像的被动无线定位与状态识别方法研究

Device-free wireless localization and activity recognition is a new technique which could localize a target and recognize its activity by analyzing the shadowing effect of the target on wireless signals, while removing the requirement of equipping the target with any device. This technique could enable traditional wireless networks to have the new ability of sensing the location and activity of the target within its deployment area, and could provide basic support for emerging new applications, e.g., smart city. However, the lacking of an effective and robust model to characterize the relationship between the link measurement and the target location and activity is a major problem which limits the performance of the device-free localization and activity recognition system. To solve this problem, the proposal investigates the relationship between multi-domain features of wireless links and target location and its activity, and realizes high performance device-free localization and activity recognition based on multi-domain radio tomography from the view of model, algorithm, and implementation. Firstly, we explore the method of estimating time-domain parameters, frequency-domain parameters, and wavelet-domain parameters based on the link measurement from multi-antenna and multi-subcarrier hardware, extracting discriminative robust multi-domain features with deep learning and transfer learning method, and establishing the model to characterize the relationship between multi-domain features and target location and its activity. Secondly, we propose an iterative sparse representation classification algorithm which is immune to noise, and a two phase algorithm which makes full use of the intrinsic relationship between the location and activity. Finally, we also study the problem of how to realize the radio tomography system rapidly and energy efficiently. The above work would provide novel thought and new methods for realizing device-free localization and activity recognition.

本课题研究一种在人员不携带任何设备的条件下，仅通过分析人体对无线链路的影响估计其位置与状态的新方法。它可使无线网络演进为具有位置与状态感知能力的智能网络，可为智慧城市等新兴应用提供基础支撑。然而，缺少健壮有效的描述人体位置及状态与无线链路测量间关系的方法一直是制约其性能的重要因素。针对此，本课题探索链路多域特征与人体位置、状态的关系，从模型、算法、实施三个层面研究基于多域射频层析成像实现高性能被动无线定位与状态识别的新思路。首先，基于多天线、多载波测量无线链路信号并估计其时域、频域、小波域参数，探索利用深度学习与迁移学习提取具有显著区分力的健壮多域特征的方法，揭示多域特征与人体位置、状态间的关系。其次，提出具有较强噪声抑制能力的迭代稀疏表示分类算法，以及考虑位置与状态内在关联性的两阶段算法。最后，探索系统实施中的快速布设与节能问题。研究成果将为实现被动无线定位与状态识别提供新思路与新方法。

基于多域射频层析成像的被动无线定位与状态识别方法是一种通过分析人体对无线信号的影响模式实现人体位置及状态进行估计的新方法，其在智慧城市、智慧空间等领域有着广阔的应用前景。本课题针对被动无线定位与状态识别中的模型、算法、实施方法展开研究。在模型方面，系统的探索了基于机器学习的链路特征提取与处理方法，给出了链路观测信息的高纬度表征方法，提出了差分相位模型用以阐明被动定位识别的机理；在算法方面，提出了基于信号本征特征、结构化特征、以及空间统计特征的被动无线定位与状态识别新方法；在系统实施方面，设计了低复杂度深度学习算法，设计了适用于被动无线定位的频谱资源感知与动态利用策略以及网络内数据高效传输方案。上述研究工作发表IEEE期刊长文论文20篇。论文主要发表在IEEE Journal on Selected Areas in Communications、IEEE Wireless Communications、IEEE Transactions on Wireless Communications、IEEE Transactions on Vehicular Technology、IEEE Transactions on Industrial Informatics、IEEE Network、IEEE Internet of Things Journal等IEEE知名期刊。有1篇论文获得辽宁省自然科学优秀论文一等奖。同时，课题组获批发明专利3项。基于上述成果，项目负责人为第一完成人获得2020年辽宁省自然科学奖二等奖（已公示）。本课题从模型、算法、实施三个层面探索了被动无线定位与状态识别领域面临的科学难题，有效地促进了该领域的技术发展，为其赋能智慧应用提供了方法支撑。