Study On Propagation Characteristics Of Indoor Millimeter Wave Based On Improved Convolution Neural Network Algorithm

With the development of 5G technology,millimeter wave has become a research hotspot in mobile communications due to its broadband,low latency,and high speed characteristics.Currently,most research on the propagation characteristics of millimeter wave channel is based on traditional channel modeling methods.However,with the increasing complexity of communication scenarios,higher communication requirements and the rapid development of antenna integration technology,it is difficult for traditional methods to handle the diverse and massive channel data,thus,a method with self-learning and adaptive learning mechanisms is required to establish millimeter wave channel propagation models.Convolutional Neural Networks(CNNs)have the unique advantages of local connectivity,weight sharing and translation invariance,which can explore the mapping relationship between input parameters and channel propagation characteristics parameters more efficiently and deeply,thereby improving the predictive performance of the model.In this thesis,the propagation characteristics of millimeter wave propagation in different indoor complex scenarios are studied based on convolutional neural network algorithm,and particle swarm optimization(PSO)algorithm is used to further improve the prediction accuracy of the model.In addition,the K-Means algorithm is used to analysis the clustering characteristics of the millimeter wave channel.This thesis provides theoretical support for the signal coverage and optimization of 5G/B5 G indoor millimeter wave,with the following research components:(1)The technical roadmap of the channel propagation characteristics prediction method based on the PSO algorithm optimized convolutional neural network is elaborated,and the actual measured data in a corridor scenario are used to build the CNN-based path loss prediction model.By comparing the predicted values with the measured values,the feasibility and correctness of using CNN to predict the channel propagation characteristics parameters are verified.On this basis,the PSO algorithm is used to automatically find the optimal hyper-parameters combination of the CNN to improve the prediction accuracy of the model,and the optimized path loss predictions are compared with the unoptimized predictions and the actual measured values.The results show that the performance of the PSO-CNN algorithm is significantly better than the CNN,and the predicted values of its path loss are better consistent with the actual measured values,verifying the feasibility of using the PSO-CNN algorithm to predict the propagation characteristic parameters of the millimeter wave channel.(2)In a Line-of-Sight(LoS)scenario of a conference room,the path loss values obtained by the Shooting-and-Bouncing Raying Tracing(SBR)method are compared with the actual measured values in the literature to verify the correctness of the SBR method.On this basis,the influence of human occlusion on the propagation characteristics of the millimeter wave channel is considered,and the PSO-CNN algorithm is used to predict and compare the propagation characteristics parameters of the indoor Single-Input-Multiple-Output(SIMO)channel for both human and non-human cases.The results show that the PSO-CNN algorithm can effectively predict channel propagation characteristics parameters such as path loss and delay spread in a LoS scenario.In addition,in order to provide a more comprehensive understanding of the millimeter wave propagation characteristics and radio wave coverage in the whole LoS scenario,an indoor received power map is constructed based on environmental information,and the PSO-CNN algorithm was also used for prediction and analysis.(3)In a None-Line-of-Sight(NLoS)scenario of a stairwell,channel modeling and simulation are carried out using the SBR method,and the influence of different antenna polarization methods on the received power of multipath signals in a multi-layer stairwell scenario is explored to select the most suitable combination of antenna polarization methods for different received paths.Subsequently,the PSO-CNN algorithm is used to predict the path loss,root mean square delay spread and angle of arrival of the channel.The predicted values are compared with simulated values,verifying the feasibility and correctness of using the PSO-CNN algorithm to predict channel propagation characteristics parameters in a complex NLoS scenario.Meanwhile,the Power Delay Angle Profile(PDAP)characteristics of the angle of arrival are studied in combination with the received power and delay.Finally,the K-Means algorithm is used to analysis the channel clustering characteristics of millimeter waves in a stairwell scenario,where the optimal K value is determined jointly by the elbow method and the contour coefficient.