Research On Passive Indoor Positioning Based On CSI Fingerprint

With the rise of the Internet of Things,indoor location-based services have received widespread attention.The use of off-the-shelf indoor WiFi equipment for indoor positioning avoids the shortcomings of traditional indoor positioning technology,such as high deployment cost,monitoring blind spots and invasion of privacy.The indoor positioning technology based on received signal strength indicator(RSSI)has low positioning accuracy due to the influence of multipath effects in the environment.The channel state information(CSI)of the physical layer can represent the signal characteristics of the spatial domain in a fine-grained manner.The proposed CSI indoor positioning scheme is continuously innovative and improved in terms of applicability and positioning accuracy.The main work and contributions of this paper are as follows:1?In this paper,a indoor positioning scheme using CSI amplitude and phase information as a fingerprint is proposed,and the object to be positioned does not need to carry any device with signal transmission function.This scheme builds a one-transmitter multi-receiver experimental platform to collect CSI data.In the signal preprocessing stage,singular value removal and low pass filtering are performed on the CSI amplitude,and the CSI phase is corrected by a linear fitting method.The processed CSI amplitude and phase information are used as fingerprints,and the fingerprint samples are trained by the deep learning based fully connected classification neural network,and matched with the collected real-time data.Experiments show that the CSI amplitude and phase positioning method can improve the positioning recognition rate and accuracy.The indoor positioning accuracy reaches 0.6m and the position recognition rate reaches 98%.2?This paper attempts to perform CSI data acquisition in a three-dimensional experimental scenario.The three-dimensional experimental scene utilizes the receiving end of different heights,which increases the spatial dimension information and improves the spatial coverage.By comparing the influence of people on the receiving information of receiving ends with different height,the height information of the measured object can be preliminarily judged,which provides a basis for subsequent intrusion target feature recognition.3?In this paper,the deep learning based neural network classification algorithm is used to train the amplitude and phase of the pre-processed CSI samples.In order to reduce the computational complexity,This paper designs a simple,basic fully connected classification neural network model based on deep learning.After many experiments,the appropriate amount of training data is selected to achieve lower computational complexity and higher positioning accuracy.