Research On CSI Fingerprint Localization Technology Based On Subcarrier Selection Strategy

Due to the low cost,wide signal transmission range and strong applicability of Wi-Fi signals,indoor localization technology using Wi-Fi signals has received much attention.Recently,the Wi-Fi indoor localization contains received signal strength(RSS)based and channel state information(CSI)based technique.Compared with RSS measurement,CSI measurement has better channel characteristics and status description between the transmitter and receiver.In order to cope with the problem of indoor localization in complex environments,in this thesis,the CSI based indoor localization algorithm using machine learning algorithms is studied.The main contributions of this work are described as follows:(1)The fundamental knowledge of CSI based indoor localization technology is studied and an experimental platform is built.At first,the relevant theory of CSI measurement parameters is introduced.And the machine learning algorithms used in the CSI based fingerprint localization model are described.Then,a hardware platform for CSI measurement collection is built using ESP32 chip.The software platform is debugged using the ESP-IDF software development environment.At last,the distance and similarity are used as metrics to analyze the CSI measurement collected by the platform.It provides the foundation for machine learning based localization research.(2)A coarse-refine localization algorithm using subcarrier selection strategy is proposed.In the offline phase,variance filtering is used to perform CSI amplitude measurement dimension reduction.The filtering preprocessing is proposed to eliminate outliers and measurement noise.Then,the Ftest classification and F-test regression are used to select the subcarriers for coarse localization and refine localization,respectively.At last,a one-dimensional convolutional neural network(CNN)is used to perform coarse localization region classification learning.And a region classification model is obtained.For another,the CSI images are constructed using the subcarriers for refine localization,and then a CNN is used for refine localization position regression learning,and a position estimation model is obtained.Experimental results show that the proposed algorithm achieves better position estimation performance.(3)A long short term memory(LSTM)localization algorithm using subcarrier selection strategy is proposed.In the offline phase,variance filtering is used to perform CSI amplitude measurement dimension reduction.The hampel filtering,robust principle component analysis(RPCA)and data smoothing techniques are used for data preprocessing.Then,the F-test regression method is used to select the amplitude measurements of subcarriers for localization.And the CSI amplitude images are constructed.The CNN with SGE attention mechanism is used to extract features from the CSI images.At last,the extracted features are utilized for regression learning with LSTM.The position regression model is obtained.The experimental results validate the efficiency of the proposed algorithm.