Channel State Information Compression Method Based On K-Singular Value Decomposition

With the growing demand for wireless Internet of things and Internet access,Wi-Fi technology has been everywhere.In addition to wireless communication,Wi-Fi system can also use channel state information(CSI)for passive sensing.Compared with other intelligent sensing solutions,it has advantages of low cost,non-invasiveness,privacy protection and illumination insensitivity.Considering users’ requirements for low cost and low power consumption and the limited computing resources of Wi-Fi terminals,the usual solution is to transmit CSI data to the cloud server,which will calculate these data in a centralized manner and then return the results to the user.However,due to the high dimension and sampling rate of CSI data,real-time CSI data flow may lead to serious communication burden,which may directly hinder the basic function of Wi-Fi,that is,Internet access.Based on the above background,this thesis carries out the research on CSI compression and reconstruction algorithm under Wi-Fi platform.Its main contents include the following aspects:Firstly,this thesis studies the CSI compression reconstruction model based on compressed sensing(CS).On this basis,a dimension reduction algorithm of compressed model observation matrix based on semi tensor product is proposed.Firstly,the algorithm constructs a low-order observation matrix satisfying the restricted isometry property(RIP),and optimizes it by singular value decomposition.Then,the original CSI data is compressed by semi tensor product matrix multiplication.Finally,the compressed CSI data is reconstructed by orthogonal matching pursuit(OMP)algorithm.The algorithm solves the problem of limited computing resources and storage space of Wi-Fi terminal to a certain extent.Secondly,in order to improve the compression and reconstruction performance of CSI compression and reconstruction algorithm,this thesis proposes a compression ratio optimization algorithm based on k-singular value decomposition(KSVD).The algorithm first randomly selects the sampled original CSI data to initialize an over complete dictionary,then uses the KSVD algorithm to alternately solve the sparse dictionary matrix and sparse coefficient matrix until convergence,and finally replaces the original universal sparse dictionary with the obtained sparse dictionary,so as to increase the CSI data sparse representation ability of the sparse dictionary,and then improve the compression and reconstruction performance of the algorithm.Finally,this thesis uses commercial Wi-Fi equipment to build an experimental platform,and carries out behavior recognition experiments based on CSI compression reconstruction in open and dense indoor areas.The experimental results show that when the amount of CSI data is reduced by 90%,the average accuracy of indoor regional behavior recognition of the system is 90.5%,which is higher than 17.9% and 23.1% of CSI compression reconstruction algorithms based on singular value decomposition and curve fitting,respectively.