Research On Modeling And Parameter Extraction Of RF Microwave Devices Based On Neural Network

In recent years,5G wireless communication technology has become more and more popular,and the wireless communication technology will be closely related to everyone,which will inevitably lead to the rapid growth of radio frequency microwave equipment,and transmitters and receivers are essential in equipment such as base stations.Low noise amplifier and power amplifier play an important role.The power,noise,efficiency and working stability of a wireless communication system like a base station can be mainly determined by the design of low noise amplifier and power amplifier.Therefore,designing RF microwave devices such as low noise amplifier and power amplifier is a good subject worthy of study.At the same time,5G communication technology requires higher performance of semiconductor devices,and high electron mobility transistors made of GaAs and Ga N have the characteristics of high temperature,high frequency and high power compared with transistors made of other materials,so it is perfect to use GaAs and Ga N materials to manufacture high power,high frequency and ultra-wideband RF devices.At the same time,the modeling method of semiconductor devices combined with neural network can solve the problems of uncertainty,long period and large error in testing,and can extract parameters quickly and accurately.In this paper,the appropriate low noise amplifier and power amplifier transistors are selected by consulting relevant literature and materials,and the neural network model is established.Firstly,the low noise amplifier is modeled by neural network,and the scattering parameters,noise parameters,gain and optimal reflection coefficient are extracted from ATF-XX1M4 and MGA-16x16 series products datasheet.BP neural network is trained by data,and two neural networks,reflection coefficient neural network and propagation coefficient neural network,are innovatively proposed to train and obtain scattering parameter values,while gain and noise coefficient are also two neural networks.The accuracy of scattering parameters extracted in this way is better.Finally,compared with the conventional single neural network structure,the average relative error is increased by 31.3%,which shows that the model in this paper has better accuracy.At the same time,a neural network model is established for the power amplifier.Ga N transistor of CGH40010 F is selected as the research object of the power amplifier.after DC scanning analysis,bias circuit,stability analysis and input-output matching circuit in ADS software,the gain bandwidth performance of the designed power amplifier is S21>13d B at the frequency of 2ghz-3ghz.when the input power is 26 d Bm,the output power is more than 39 d Bm at the frequency of 2ghz-3ghz.Finally,the scattering parameters,output power,power gain,efficiency,third-order intermodulation point and other parameter data are extracted.ADS simulation data are used to train BP neural network and RBF neural network,and the two neural network models are evaluated to find the best neural network model.the results show that the average relative error of scattering parameters of BP neural network model is 0.23%.The average relative error of harmonic balance parameters is 0.788%.The average relative error of RBF neural network model is 0.79%.The average relative error of harmonic balance parameters is1.378%.The results show that the model has good accuracy and reliability,and can be applied to the field of radio frequency devices of wireless communication transmitters with wide band,high efficiency and high power.