Research On Indoor WiFi Localization Technology Based On CSI

With the great development of mobile Internet technology, people have an increasing demand for precise location information and various of localization services have brought great convenience for daily life. The Global Positioning System (GPS), which has achieved high accuracy in outdoor environment, is faced with numerous challenges in indoor environment due to the blocking of satellite signal by buildings, thus the research on localization approaches suitable for indoor environment is of great significance. The WiFi-based approach stands out from all kinds of indoor localization approaches and has become a focus recently due to its convenience and low cost, which can make use of existing WiFi infrastructure everywhere nowadays. The successful extraction of fine-grained channel state information (CSI) in commodity WiFi devices makes it possible to overcome the disadvantages of the RSSI-based indoor WiFi localization technologies and further improves the accuracy of indoor localization. Therefore, the research of CSI-based indoor localization technology is of great practical significance.The thesis focuses on the research of CSI-based WiFi indoor localization technology, firstly the advantages and disadvantages of the existing indoor localization approaches are summarized and analyzed comprehensively, then three main WiFi indoor localization approaches are introduced emphatically, and the comparisons of advantages and disadvantages between CSI and RSSI have been made. Due to the high complexity and difficult deployment of fingerprint-based approach, the CSI ranging-based localization scheme is proposed. The main contributions and work of the thesis are as follows.Firstly, the CSI preprocessing approaches have been summarized and the CSI preprocessing schemes specifical for indoor localization including Hampel filter, phase continuation, linear transformation and multipath suppression are proposed, which lay a good foundation for localization and Non-Line-of-Sight (NLOS) identification and mitigation.Since the accuracy of indoor localization techniques, especially ranging-based are faced with numerous challenges due to NLOS, a new NLOS identification and mitigation algorithm based on Support Vector Machine (SVM) is proposed. According to the theoretical analyses and observation of CSI from numerous experiments in different environments, the existing features have been summarized and a new feature and refined phase fator are proposed, which will be leveraged as candidate features to combine with LS-SVM to solve the nonlinearly separable problem of NLOS identification. What's more, a new nonlinear indoor propagation model based on NLOS identification is proposed to mitigate the NLOS error. The experimental results show that the proposed NLOS identification algorithm, NISVM has higher NLOS recognition rate and lower missed detection probability and false alarm probability compared with the existing algorithms, and the NLOS mitigation algorithm, NLRSVM has a smaller distance estimation error, compared with the existing CIR-based ranging estimation schemes.In order to further mitigate the NLOS error and optimize the geometric layout of base station, an improved localization algorithm based on virtual base stations (VBS) is proposed. The proposed algorithm constructs virtual base stations based on the estimating distances and obtains the angle of VBS by minimizing Geometric Dilution of Precision (GDOP) and then replaces the physical base station to realize localization leveraging two-step least squares. The simulation results demonstrate that the proposed algorithm has the advantages of fast convergence rate and high localization accuracy.