Research On Environment Recognition Of Radio Propagation Model Based On Deep Learning

The rapid development of economy and society promotes the explosive growth of radio service demand,but it also brings some problems.It is reflected not only in the shortage of radio spectrum resources,but also in the increasingly complex radio propagation environment,which brings new challenges to the prediction and coverage of radio propagation.A large number of studies and experiments show that the environmental variables in the process of radio propagation path have an important impact on the accurate prediction of signal fading and coverage.Especially with the advent of 5G era,with the increase of frequency,the wavelength will be further reduced,and the influence of environmental variables will be further deepened.Therefore,a method that can identify the radio propagation environment quickly and accurately will provide effective reference and key support for the prediction of which radio propagation model is used in a certain area.This paper first introduces the typical types of radio propagation models,including empirical models based on actual tests: free space loss model,Okumura Hata model,cost-231 Hata model,Lee model.The above models are a lot of data fitting and testing,which are widely used in the field of radio propagation prediction and network optimization,and a lot of them have been obtained Verification;as well as the theoretical model: ray tracing,FDTD(finite difference time domain).By summarizing the fitting calculation method and application scope of the above model,including the applicable frequency band and other parameter information,the main causes and types of signal attenuation are analyzed,and the judgment that the environmental variables are the main factors affecting the signal attenuation is obtained,which provides the basis for the following research.Secondly,in order to solve the problems of low efficiency and low accuracy in the traditional radio communication process,which rely on the artificial recognition of the communication environment,an intelligent recognition method based on deep learning is proposed.Firstly,deep learning and convolution neural network are introduced,and its development process and basic principle are described.After that,two basic operations(down sampling and pooling)and three basic methods(based on region recommendation,regression and multi task model)are introduced to solve the problems encountered in the process of target recognition of radio propagation model.It provides a model and direction for the identification of radio propagation environment variables.Then,eight kinds of convolutional neural networks are selected as the basic network training,resnet50 with the highest recognition efficiency and accuracy is selected as the basic network,which integrates the high and low level features and enhances the recognition ability of deformation model.At the same time,the improved non maximum linear suppressionmethod is used to deal with the problem of feature frame overlap.Combined with the radio propagation model,an improved algorithm is proposed to improve the radio propagation The recognition rate of environmental variables is 95.8%,and the efficiency of single recognition is 0.43 s.After that,the theoretical calculation and actual measurement results of the radio propagation model suitable for the region are compared and analyzed,and the overall results are consistent with the expectation.The validity of this method is verified when it is applied to the environment identification of radio propagation model.Finally,the paper summarizes the selection of radio propagation model and the problems in the process of network training,analyzes the gains and losses in the experiment and data processing,summarizes the experimental results,and puts forward the next work plan and prospect.