Design And Research Of Indoor Localization Algorithm Based On 5G Sub-Base Station

With the development of the Internet of Things,all aspects of social demand have put forward wider and deeper requirements for location-based services.In indoor environment,due to a large number of reinforced concrete structural frames in modern buildings,the signal transmission path is damaged,and GPS positioning system cannot play a role in indoor environment.WiFi signal,as a radio signal,is easily disturbed by a variety of environmental factors in the process of transmission,resulting in serious instability of the signal.As a new mobile communication technology,5G has the characteristics of multiple locations,safety and reliability.In this paper,5G technology is combined with Deep Neural Networks(DNN)and Convolutional Neural Networks(CNN)to propose an indoor location scheme based on 5G NR parameters.It mainly includes three contents:(1)Build 5G location fingerprint database and collect 5G NR signals in indoor environment.The collected 5G NR signals SS-RSRP,SS-SINR and SS-RSRQ are preprocessed and corresponding to the location label to realize the construction of location fingerprint database.(2)The positioning model of deep neural network and convolutional neural network was built,and the training of the positioning algorithm model was completed.The network structure of deep neural network and convolutional neural network,as well as the model parameters and optimization method of model training were determined.(3)For the established 5G fingerprint database,the neural network model is used to carry out localization experiment and analysis.Using DNN and CNN to achieve 5G indoor positioning,The test of the scheme was completed by comparing the parameters of PRF value,namely Precision,Recall,micro F1-score and crossentropy loss,as well as the mean error between real data and test data..The test results show that the positioning accuracy of DNN model gradually converges to about 92%,and that of CNN model gradually converges to 96%.The average errors of DNN model and CNN model were 1.62m and 1.33m,respectively.The experimental scheme of DNN and CNN's 5G-based indoor positioning algorithm is reasonable,with stable function and high positioning accuracy.The convolutional neural network training model is superior to the deep neural network training model to a certain extent.The overall data volume and computational complexity of the scheme are relatively small,and the positioning accuracy is high.