Research On WiFi Indoor Location Algorithm Based On Deep Neural Network

With the growing demand for location-based services in indoor environments,WiFi indoor positioning technology has attracted more and more attention due to its wide coverage,low cost and strong penetration.The algorithm based on Received Signal Strength Indication(RSSI)has a large granularity and low positioning accuracy,and is not suitable for the application scenarios where there is a common need for high positioning accuracy.In recent years,because the measurement of Channel State Information(CSI)has become more and more convenient,the indoor location of WiFi based on CSI has attracted the attention of researchers.Among them,measuring angle information and distance information based on CSI information is always a research hotspot.Due to the influence of multipath,traditional algorithms such as Multiple Signaling Classification Algorithm(MUSIC Algorithm)are difficult to achieve the positioning accuracy required for practical applications.Therefore,this paper designs and realizes an indoor location method based on channel state information combined with depth neural network.In this paper,a unique CSI processing method is presented,and then the processed data is input into neural network for parameter estimation.Specifically,for raw CSI information,this method first removes the CSI phase shift caused by each two parameters of Angle of Arrival(Ao A),Angle of Departure(Ao D)and Time of Flight(To F)in CSI,and then uses this clean data to estimate the other parameter.Then,according to some mapping relationship between CSI signal and Ao A,Ao D and To F,this paper proposes a new algorithm for data adaptive convolution neural network(DA-CNN)to estimate Ao A,Ao D and To F parameter information.The algorithm uses Super Net structure to search neural network structure,which can search neural network structure suitable for CSI data and improve the accuracy of parameter estimation.Experimental results show that the accuracy of the proposed method is much higher than that of the conventional neural network.Then,the CSI matrix is reconstructed by using the parameters that have been estimated by the DA-CNN algorithm,Set the time window T to superpose the CSI matrix of the previous T moment.And the CSI matrix are fed into Convolution Long Short-Term Memory network(Conv-LSTM),which could full use of the time and space information in the CSI sequence to track the user’s location in the indoor environment.The experimental results show that the positioning accuracy can reach decimeter level,the median error can reach 0.135 m,and the estimation error of 90% is less than 0.5m.