Research On Indoor Environment Blind Zone Recognition Of Heterogeneous Wireless Network Based On Gaussian Process Regression

The increasing demand for system capacity has facilitated the development of communication technology.By introducing high-frequency communication signals to the system,gains in the system bandwidth can be achieved,which however leads to the increasing problem of blind zone phenomena.At the same time,the tre nd of heterogeneous network being the future network frame has become obvious,which brings more challenge to the blind zone recognition.There is no efficient technical method for blind zone recognition so far.Traditional methods of detecting blind zone area mainly focus on outdoor environment,and are with the characteristics of inflexibility,low efficiency,time-costing and low accuracy.Blind zone recognition in indoor environment mainly suffers from the intense influence of multipath effect,which leads to the densely distributed deep fading points.Under the circumstances,if one wants to get the accurate blind zone distribution,two main problems must be solved,namely,the influence of multipath effect and the problems of computation and storage.All those problems in blind zone recognition can be concluded into blind zone detection problem and blind zone reconstruction problem respectively.First,the usage of Gaussian process regression(GPR)on the task of blind zone reconstruction using limited sampling points is discussed.Then,we introduce some basic concepts of GPR and the well-known machine learning kernel trick.We also have a discussion on different kernels' property in the regression task,and the impact of different hyperparameter value on the model is analyzed.In the end,we make an attempt to utilize the traditional covariance function GPR to reconstruct the blind zone,and we give some inspirations to solve the drawbacks of this method.Then,we focus our research on the filtering method of sensor network nodes to eliminate multipath effect's influence.Through the analysis on the cause of multipath effect,we utilize fractional Fourier transform's(FRFT)concentration property to linear frequency modulated(LFM)signals and propose a multipath effect eliminating algorithm based on FRFT.And then,simulation is conducted combining the channel model.The comparison between traditional filtering method and the proposed method has shown that our method has characteristics of heterogeneous network adaptability and high accuracy.In the end,we combine the specific task of blind zone reconstruction with the design of covariance function.We fisrt begin with the inroduction of model and parameter selection.Then we apply the traditional covariance functions to the indoor environment blind zone recognition,and the performances of them are analyzed.We also discussed the problem of local optima.According to the performances of traditional covariance functions,we design a new covariance functi on specifically used in blind zone reconstruction.Simulation is then conducted to show the performance of the designed kernel and those of traditional ones.Finally,we conclude this paper and some problems to be solved as well as future research directio ns are mentioned.