Research On Intelligent Design Methods For Broadband Filters

Currently,microwave filters are widely used in civil and military RF fields.In view of the current trend of rapid development of wireless communication technology,it is necessary to quickly and accurately complete circuit simulation in microwave design simulation.Therefore,it is necessary to establish a set of efficient and high-precision filter design methods.This paper mainly focuses on broadband filters,using machine learning and deep learning to construct suitable electromagnetic substitution models as a computer-aided design method for microwave filters,thereby improving the efficiency and accuracy of the size and structure design of broadband filters.In this paper,the design theory of filter is briefly introduced firstly.There are three parts:(1)Obtaining the system function.(2)The implementation of filter.(3)Simulation and optimization of filter.Based on the theoretical analysis of four kinds of structure design with parallel coupling lines,the general framework of intelligent filter design method is given.Secondly,in order to shorten the Design cycle of the filter,it is proposed that the joint programming of electromagnetic simulation ADS(Advanced Design System)and MATLAB can automatically generate batch data for network model training,which is very important for data-driven optimization algorithm.As a data-driven optimization algorithm,network model needs a lot of training data to improve its performance.This method directly establishes the internal relation between the characteristic response of wideband filter and the physical dimension parameters,without in-depth theoretical reasoning or solving nonlinear equations,and effectively improves the efficiency and practicability of wideband filter design.Third,The design of single stage filter based on K-nearest neighbor regression algorithm is presented.This design method takes the characteristic response curve of the single stage wideband filter collected by the co-simulation as the input and the physical size corresponding to the characteristic response curve as the output variable for network modeling.It can accurately and efficiently realize the same design function of wideband filter as the traditional theoretical analysis method.At the same time,the trained network structure can be used to realize the design of single-stage wideband filter under different design indexes of bandwidth in the data acquisition range.Finally,a comparison between BP neural network and Auto ML machine learning is presented to design a two-stage wideband filter.Select a network model with good results and then conduct further analysis training.The experimental results show that the design efficiency and accuracy of wideband filter can be significantly improved by using automatic machine learning method,and compared with traditional filter design,automatic machine learning method can find the desired physical size of filter design within the error range more quickly.Through a large number of experimental data and verification,it is proved that the method has wide applicability and practical value in practical design.In the future,the method can be further improved,including using more data sets and optimization algorithms to improve the accuracy and robustness of the model,as well as applying the method to the design of other microwave devices.