Study On Nonlinear Intelligent Model Of Traveling Wave Tube

Traveling Wave Tube(TWT)is a vacuum electronic device used to amplify microwave signals,which has important applications in millimeter band and terahertz band.The large signal model of TWT can be used to calculate the output power,efficiency,gain,etc.Deep learning,which obtains functional relationship between inputs and outputs by learning intrinsic features from large amounts of data,is currently achieving excellent results in several fields.In this thesis,the study of large signal model of traveling wave tube based on deep learning is carried out and the output power can be calculated by seven parameters: phase velocity,coupling impedance,current,voltage,frequency,input power,and tube length.The main contents and research results of this thesis are as follows:(1)The neural network models of 6GHz-9GHz single-segment helix TWT and208GHz-212 GHz single-segment coupled cavity TWT are trained.Firstly,the data are calculated respectively by one-dimensional large signal program.Then the data are pre-processed,including kernel function,feature selection,normalization,data storage and so on.Finally,the neural network is built and trained,and the hyperparameters are adjusted according to the training results to obtain the optimal network model.The results calculated by the trained model are compared with those by the one-dimensional large signal program,verifying that the output power curves calculated by the two methods are basically the same.(2)A secondary training algorithm is proposed.Since there are errors between the results of one-dimensional large signal program and the experimental data,the trained model needs to be corrected.The correction coefficient of the coupling impedance is calculated according to the experimental data.The neural network model used to calculate the correction coefficient is trained.Then the corrected parameters are input into the one-dimensional large signal program to calculate the output power of TWT,which reduces the error between the predicted value and the experimental data.(3)An application with a graphical interface based on Qt,python and C++programming languages is developed.The python language is responsible for calling the Py Qt5 library to design the interface.The C++ language is mainly responsible for encapsulating and calling the function functions,and interacting with python through the dynamic link library.The application integrates the function of storing data,training the neural network model and calculating the output power of TWT.