Research On Indoor Localization Algorithm Based On WiFi Channel State Information

With the development of wireless network technology,the application of channel state information technology research to various indoor scenarios has been a problem to be solved in the field of indoor positioning.The active localization technique based on channel state information achieves indoor localization by signal phase.In the signal processing stage,we analyze the sampling clock offset and sampling frequency offset in the signal acquisition process,and propose to eliminate the phase error by linear fitting and multiple input and multiple output phase processing methods.In the estimation of signal arrival angle,this thesis estimates more accurate antenna arrival angle by rotating the same antenna for several experiments and combining the multi-signal classification algorithm to combine the mathematical equation.In order to select the apparent propagation paths that can be used for localization in complex indoor environments,this thesis sets different weights by Gaussian clustering and selects five apparent propagation paths with the highest probability,and at the same time establishes an indoor coordinate system to estimate the target position by combining the existing coordinate positions,and finally uses the least squares method to obtain the target position with minimized error.The average accuracy of the active indoor positioning system designed in this thesis is 0.32 m.The passive positioning technique based on channel state information achieves person detection in terms of signal amplitude.The signals collected by the passive technique based on channel state information come from non-visual propagation paths,and the signals collected are not only reflected signals from the human body but also reflected signals from other objects.The signal is smoother than the traditional filtering method.In order to extract the human activity features from the signal,this thesis uses the short-time Fourier transform to convert the extracted signal principal components into a spectrogram,and identifies the human activity features through the spectrogram and draws the moving speed curve of each part of the human body,finally the spectrogram and the drawn speed curve are used as features to train the classifier by machine learning methods,and the results show that the trained classifier model has a high recognition rate for the four experimenters.