Study On Modeling Of 28GHz Millimeter Wave Channel Based On Bp Neural Network Method

As one of the key technologies for 5G,millimeter-wave wireless transmission is considered to be able to expand the use of new spectrum resources and deeply exploit spatial dimension wireless resources.It is one of the most potential research directions for future wireless communication systems.Based on BP neural network,the millimeter-wave channel modeling research was carried out in this paper,and the multi-path clustering characteristics of 28 GHz millimeter wave in indoor line-of-sight environment are studied and analyzed.The main research contents and results of this paper are as follows:(1)Firstly,the research status of millimeter wave channel measurement and modeling,millimeter wave propagation simulation technology and channel modeling usin artificial neural network are summarized in this paper.Secondly,the basic theories of millimeter-wave wireless channel,Shooting and Bouncing Ray Tracing/Image Method(SBR/IM),BP neural network method and its computer realization process are introduced in this paper(2)In this paper,the 28 GHz millimeter wave simulation is carried out in the conference room line-of-sight(LOS)environment.The path loss is modeled based on BP neural network method and compared with the measured data.It is concluded that the model has the highest accuracy when the number of neurons in the hidden layer is 6,and the correctness of the BP neural network channel modeling is verified.The experimental results show that the BP neural network path loss model is more accurate than that of SBR/IM method.Then the BP neural network method is used to analyze the root mean square(RMS)delay spread and Ricean K factor.The results show that BP neural network can predict the simulation data very well.The RMS delay spread can be describled as a normal distribution.The normalized Ricean K factor,which can be expressd as an exponential function with distance between receiver and transmitter,can be described by an exponential distribution.(3)According to the existing literature,the T-shaped corridor is simulated in this paper.Combining with the BP neural network method,this paper analyzed the path loss,shadow fading,RMS delay spread and Ricean K factor according to corridor LOS environment and corridor nonline-of-sight(NLOS)environment,respectively.Results show that the BP neural network can predict these parameter values well.The shadow fading,RMS delay spread and Ricean K factor can be described as normal distribution with a mean of zero,normal distribution and exponential function.Moreover,the relationship of Ricean K factor and distande between transmitter and receiver can be expressed by an exponential function.Finally,the path loss from corridor LOS to NLOS environment was modeled and analyzed.(4)In this paper,the multipath clustering characteristics of the 28 GHz millimeter-wave in an indoor LOS environment are simulated and analyzed based on BP neural network results.Firstly,the threshold clustering method is used to cluster from the power-delay domain and the power-angle domain respectively.It is found that there is a strong relationship between the multipath delay and the angle,relevance.Then the K-means algorithm was used to cluster the multipath from twodimensional(2D)and three-dimensional(3D).The optimal number of clusters is determined by the CH index.Through the K-means algorithm,the number of sub-paths of each cluster and the coordinates of the center of the cluster are determined.Then the intra-cluster and inter-cluster parameters of each cluster were studied.It was found that the parameters within the intra-cluster and inter-cluster can be express by a normal distribution.