Research On Non-line-of-sight Identification Based On CSI

The research on the randomness,time-varying and non-linearity of signals in wireless communication environment drives the development of wireless communication system towards higher transmission rate,more advanced coding algorithm and more efficient power efficiency.WiFi wireless technology,widely used in local environment,not only solves the last kilometer access problem of high-speed access to the Internet,but also breeds and promotes a series of wireless sensing applications such as indoor positioning,gesture and activity recognition,device-independent positioning and so on.The problem of non-line-of-sight identification of WiFi signal propagation is the guarantee of basic technology to improve the positioning accuracy of wireless positioning system,provide quality information of wireless link and optimize communication modulation mode.The physical layer of the new generation WiFi wireless network adopts multi-carrier technology,which can effectively suppress the effect of multipath interference by acquiring the CSI(channel state information)data of each subcarrier.Therefore,based on the channel state information(CSI)of wireless links,this paper proposes a systematic method of non-line-of-sight recognition based on machine learning.The main research contents are as follows:(1)In order to solve the problem that CSI data in WiFi wireless network are vulnerable to burst interference and phase offset,a method combining local outlier factor detection LOF,Hampel filter and linear phase transformation is proposed to detect,eliminate and correct the abnormal factors in the pretreatment of initial data.(2)Non-line-of-sight recognition in indoor environment.On the basis of extracting statistical features such as mean,variance,standard deviation,coefficient of variation,skewness,kurtosis,phase difference factor and Rician-K factor from CSI data,a grey wolf optimization algorithm is proposed to select penalty factor and kernel function parameters of support vector machine based on support vector machine method for classification model training and detection and recognition(3)Platform,scene layout,measured data and integration algorithm are built.The influence of parameter selection,feature clustering and performance analysis of classification algorithm are carried out respectively.The results show that the recognition rate of the proposed method is about 97%in indoor environment,and the performance of the proposed method is better than that of other methods.