Characterization And Extraction Method Of Microwave Semiconductor Device Dual Frequency Nonlinear Network Model

In recent years,with the rapid development of the third generation of semiconductors,communication electronics technology has undergone tremendous changes.In particular,its application scenarios tend to be multi-frequency and high-power environments,and a multifrequency communication system capable of supporting multiple frequency bands at the same time has become a key means to improve communication efficiency.Establishing a multi-frequency nonlinear model of a semiconductor device and applying it to device simulation design is of great significance for improving communication quality and reducing communication costs.Therefore,research on multi-frequency nonlinear systems,especially dual-frequency communication technology,has become an important direction at present.Because the traditional S-parameter linear model has natural deficiencies in characterizing nonlinear systems,building an accurate and efficient nonlinear network model is essential to the development of communication technology.This paper discusses the nonlinear characteristics and modeling of high-power semiconductor devices,and proposes two research directions for modeling dual-frequency nonlinear systems.The main contents are as follows:In the first part.By studying the basic theory and extraction scheme of X-parameters,we find that the conventional X-parameter measurement system is very complicated and has the disadvantages of long test period and many measurement data.The strong learning ability and accurate prediction ability of the ELM neural network algorithm can make up for the shortcomings of the conventional X parameter measurement system.Based on this,this article uses single-frequency X parameters as the starting point,and uses the X parameters extracted from ADS as a training set to train the ELM neural network to establish a singlefrequency X-parameter prediction model,and then uses the same method to establish a dualfrequency X-parameter prediction model.Finally,the accuracy of the training model was verified.The results showed that 97% of the prediction data of the single-frequency Xparameter prediction model had a maximum relative error of no more than 0.03,and 95% of the prediction frequency of the dual-frequency X-parameter prediction model had a maximum relative error of no more than 0.05.Finally,the accuracy of the model was verified by using harmonic circuit simulation.In summary,the X-parameter prediction model established in this paper can accurately characterize nonlinear system characteristics;The second part is because the theory of X parameters is more complicated,it is more difficult to directly use X parameters to optimize the circuit during simulation.Taking dual frequency input as an example,the X parameters used to characterize the intermodulation components of each order are very complicated and it is difficult to understand the relationship between the parameters.Therefore,a new path is established in this paper to establish a dual-frequency nonlinear network W-parameter model,which directly links the intermodulation components of each order with their fundamental and harmonic components.The W parameter is not only concise and clear in characterizing the intermodulation components of each order,but also easy to optimize and adjust during the simulation process.After elaborating the basic theory and model formula of W parameters in detail,this paper proposes two measurement schemes for extracting W parameters,and introduces the calibration of measurement schemes,fixture design,and fixture de-embedding methods;In summary,this paper proposes a single-frequency / dual-frequency X-parameter prediction model based on the ELM extreme learning machine,and verifies the accuracy and practicability of the model by designing a power amplifier.In addition,this paper proposes another dual-frequency nonlinear network W parameter model,and designs a corresponding measurement scheme.