Research On Feature Mining And Modeling Of Scattering Communication Channels

Scattering communication is a very widely used method of over-thehorizon communication.It has the characteristics of being less affected by weather and terrain,not easy to be intercepted,and long communication distance.Therefore,it has played an important role in military warfare and disaster relief operations.Among them,the scattering communication channel It has many characteristics such as large path loss,insufficient multipath and small Doppler frequency shift.In recent years,machine learning technology has developed rapidly,its application scenarios have become more extensive,and significant results have also been achieved in the field of communication and channel modeling.Based on the relatively complete SC-FDE system in the laboratory,this paper collects a large amount of field channel data,and uses machine learning technology to mine the characteristics of the channel data.At the same time,because of the many conditions that affect the scattering channel,it is difficult to analyze,and the use of machine learning technology is proposed.A method to simplify the modeling process of the scattering channel.The main work of this paper includes the following two points:First,the random forest model is used to mine the corresponding relationship between the communication distance and the delay power spectrum of tropospheric scattering communication system,so that the receiver can infer the communication distance according to the delay power spectrum.The channel data under 7 different communication distances are collected in the field.First,the channel data is analyzed.It can be obtained that the communication distance of the tropospheric scatter communication system has a strong correlation with the delay power spectrum,so the channel time can be used.The extended power spectrum classifies different distance scenarios.Next,perform data preprocessing on the collected channel information,including data cleaning,normalization,and averaging.Different classification models in machine learning are used to classify different communication distance scenarios,and the performance of different models is evaluated.In contrast,the random forest model has the highest classification accuracy,which can reach more than 90%.Then use the random forest algorithm to build a regression model to complete the prediction of the communication distance,and use Qt to design a stable visualization module.Secondly,the tropospheric scattering channel is modeled by neural network.Without knowing the complicated parameters such as scattering angle,altitude and antenna height,the channel model based on neural network is obtained only according to the channel input and output measurement results to realize the fast modeling of corresponding channel.First,the traditional channel modeling method is introduced.On this basis,a channel model based on the Jakes model and the tapped delay line model is established.By simulating the traditional model,the input data and output data of the channel are obtained.The neural network has a good ability to simulate nonlinear systems,which is very consistent with the actual channel environment.The data obtained above is processed as a training sample of the neural network,and two models of the BP neural network and the RBF neural network are used for training.The constructed RBF neural network model can better learn the characteristics of the channel,and the bit error rate curve obtained by this model is also roughly consistent with the bit error rate curve obtained by the traditional model.In this paper,an in-depth study on the mining and modeling of the characteristics of the scattering communication channel is carried out,and the machine learning method is applied to it,which provides new ideas and development directions for the future research of the scattering channel.