Parameter Extraction Of Equivalent Circuit Of High Frequency Field Effect Tube Based On Support Vector Regression

The characterization of the equivalent circuit model of the gan high electron mobility transistor is the prerequisite for the design of radio frequency integrated circuits,and the key to establishing the equivalent circuit model is whether the extracted parameters have physical meaning,so the equivalent circuit parameter extraction research is the key to device modeling.At present,the extraction accuracy of equivalent circuit parameters is limited by the working frequency band of the measured data and the bias conditions of the device under test.The traditional equivalent circuit model lacks the simulation capabilities of frequency extension and offset point interpolation.As the feature size of devices continues to decrease and high-frequency characteristics continue to improve,it becomes more difficult to accurately extract gan hemt small-signal equivalent circuit parameters,which increases the cost of circuit research and development.To solve the above problems,this paper uses the prominent frequency extension and offset point interpolation capabilities of support vector regression to carry out the following research work.First of all,selecting accurate and reliable training samples and test samples is the basis for constructing an support vector regression model.The optimal hyperparameters of the model are sought by optimizing algorithm iterations and training samples,so that the model has a good learning ability and can accurately characterize the nonlinear function relationship between the working conditions of the device and the small signal characteristic parameters.For this reason,this paper studies the direct extraction algorithm of gan hemt high frequency small signal equivalent circuit parameters.The dual-port network conversion technology is used to extract the equivalent circuit parameters,so as to provide accurate broadband S-parameter simulation data for the scattering parameter model built in this paper and provide effective sample data for the intrinsic parameter model.Secondly,s-parameters can correctly reflect the small signal characteristics of the device,as a priori knowledge to ensure the validity of the intrinsic parameter model based on SVR.With reference to the support vector regression modeling experience of radio frequency devices,the selection of key parameter values determines the support vector regression model learning and generalization capabilities.In this paper,particle swarm optimization algorithm is used as a key parameter optimization tool of the model to avoid the problems of slow convergence speed and poor generalization ability caused by hyperparameter variation during model training.By introducing noise factors,the antiinterference ability of the model is improved.Compared with the literature ann model,the sparameter support vector regression model built has a good frequency extension capability in the [13GHz-18GHz] frequency band after testing and verification.Therefore,adopting the built s-parameter support vector regression model can solve the problem of omission of frequency point scanning,and it is easy to continuously obtain s-parameters at any bias point and frequency range,avoiding complicated s-parameter measurement and tedious extraction process.Finally,the intrinsic parameters are the key to reflect the physical characteristics of gan hemt.This paper establishes an intrinsic parameter model based on support vector regression and extracts the intrinsic parameters.Using the prior knowledge injection method,the sparameter model is combined with the support vector regression algorithm as prior knowledge to solve the accuracy problem caused by insufficient data.The built model avoids the cumbersome steps of measuring parameters in complex physical property analysis and modeling and characterization,and improves the offset point interpolation capability that traditional models lack,and can quickly extract intrinsic parameters in a wider frequency band and offset range,So as to realize the effective characterization of the gan hemt small signal equivalent circuit.The intrinsic parameters extracted based on the support vector regression intrinsic parameter model and the literature data were embedded in simulation software for comparison and verification.The results show that the support vector regression intrinsic parameter model has less error at high frequency points,indicating that the built support vector regression intrinsic parameter model is in It has obvious advantages in modeling efficiency and modeling accuracy.