Computer Aided Design Method Of Microwave Filter Based On Convolutional Neural Network

Microwave filters are currently widely used in civil and military communication systems.With the continuous development of microwave technology,it is necessary to quickly and accurately complete circuit simulation in microwave design simulation,which is difficult to achieve in traditional microwave CAD.aims.Under normal circumstances,due to assembly errors and other factors in actual production and processing,the performance of the actual microwave filter and the circuit design will be quite different.To meet the actual performance requirements,experienced debuggers are required to filter the microwave based on experience.The adjustment of the physical size parameters of the microwave filter greatly increases the production time and production cost of the microwave filter.To solve this problem,this paper proposes the prediction of the physical size parameters of the microwave filter based on the convolutional neural network model.This article first briefly describes the construction of the data set required for the algorithm model.The electromagnetic simulation software Sonnet is used to model the microwave filter and obtain the microwave filter scattering parameter data and the corresponding physical size parameter data.Because the direct use of the scattering parameters for training has insufficient accuracy Therefore,the coupling matrix parameter extraction method is used to extract the coupling matrix to optimize the loss in the scattering parameter by using the coupling matrix parameter extraction method,and finally the coupling matrix data of the microwave filter and its corresponding physical size parameter data set are obtained.Secondly,use the multiple linear regression in statistics for training on the constructed data set.When evaluating the training results of the model trained by the multiple linear regression model,because its statistical quantitative indicators and actual simulation waveforms are not good enough,in response to this problem,we assume that the microwave filter coupling matrix data and the physical size parameter data show a nonlinear relationship,Introduced the support vector regression model that showed better performance for nonlinear problems in the regression model to train on the data set,and found that it performed better than the multiple linear regression model in both statistical and quantitative indicators and model prediction simulation waveforms.Finally,based on the analysis of the results of multiple linear regression and support vector regression models,a convolutional neural network model is proposed to fit the correspondence between the coupling matrix of the microwave filter and the physical size parameters.Finally,it is found that whether it is through the quantitative analysis of statistical indicators or the intuitive experience of simulated waveforms,the convolutional neural network model has shown better performance in predicting physical size parameters.The feasibility and superiority of the convolutional neural network model in predicting the physical size of the microwave filter are verified.The research in this article will provide an efficient and accurate design method for the circuit design of microwave components represented by microwave filters,which has important application value and research significance.