Microwave power transistor X parameter modeling method based on LSTM technology

Abstract

The invention discloses a microwave power transistor X parameter modeling method based on a long short term memory (LSTM) technology, which comprises six steps of data preparation, model construction, model training, model testing, prediction model verification and X parameter model generation. The core of the invention adopts an LSTM modeling method, and the method introduces a gate control mechanism into a network structure so as to control the accumulation speed of information and selectively receive or forget known information, which is helpful for avoiding the gradient disappearance problem in the training process. According to the method, the X parameter of the microwave power transistor can be accurately represented based on limited measurement or simulation data, and the problems of uncertainty, long period, large error and the like of a traditional modeling method can be solved, so that the design efficiency and precision of a radio frequency circuit are further improved.

Description

Technical Field

[0001] The present invention relates to the research field of microwave power transistor parameter modeling, and more specifically, to a microwave power transistor X-parameter modeling method based on LSTM technology. Background Art

[0002] The rapid development of the communication microwave industry has put forward higher requirements for RF circuits, that is, while meeting the basic requirements, it is required to have higher design accuracy and performance. However, as a key device in RF circuits, the model accuracy of microwave power devices is directly related to the performance of the circuit. Under large signal excitation conditions, the power amplifier will produce nonlinear characteristics such as fundamental wave compression characteristics, harmonic distortion, and amplitude modulation-phase modulation characteristics when operating in the nonlinear region. Therefore, the traditional small signal S parameter analysis theory can no longer meet the current high-power device design requirements, and the modeling technology of power devices under large signal working conditions has become a research hotspot.

[0003] Existing simulation modeling software and modeling technology are becoming mature. From the perspective of nonlinear modeling technology of RF devices, they can be divided into empirical models, physical models and behavioral models. Among them, the most representative behavioral modeling method is artificial neural network technology. Because of its fast training speed and high accuracy, it has become an important tool for modeling microwave power devices. This technology draws on the characteristics and operating principles of biological nerves, and can analyze and process information quickly, simultaneously and independently, and store information in each neuron. In addition, the entire neural network model has the ability to map the large signal characteristics of microwave power transistors, and can continuously learn and adjust the weights within the network according to environmental adaptive changes. Therefore, it is feasible to use artificial neural network technology to construct an accurate microwave power transistor parameter model. In summary, the core of the present invention adopts a long short-term memory (LSTM) neural network modeling method to characterize the nonlinear relationship between X parameters and microwave power transistor performance, and further improve the design efficiency and performance of RF circuits. Summary of the invention

[0004] The purpose of the present invention is to establish an accurate microwave power transistor parameter model and propose a microwave power transistor X-parameter modeling method based on LSTM technology, so as to provide a more efficient and flexible design method for microwave engineers.

[0005] The technical method and solution of the present invention for solving the above-mentioned microwave power transistor parameter modeling are as follows: a microwave power transistor X-parameter modeling method based on LSTM technology, including six steps of data preparation, model construction, model training, model testing, model verification and prediction, and generation of X-parameter model; Step 1: Data preparation is to build a schematic diagram for extracting X parameters in the software, determine the value of the input vector and input it into the X parameter control. By setting the corresponding bias voltage and frequency, a certain number of X parameters are finally obtained after the simulation is completed, and they are divided into training sets and test sets according to appropriate proportions; Step 2: Model construction is to determine the characteristic parameters that affect the performance of microwave power transistors. X = [ X 1 , X 2 , …, X n ] as input variables, X parameter set as output variables, and set the step size of each input variable. After determining the input and output, create the LSTM structure and set the number of neurons and initial weights in the input layer, hidden layer, and output layer of the neural network; Step 3: training the model. After setting the training parameters such as the maximum number of iterations, the optimal learning rate, and the optimal error tolerance, the model starts training. If the training parameters meet the requirements or the number of iterations reaches the maximum number, the training ends. Step 4: Test the model by using the test set to test the simulation results of the model trained in the previous step. If the mean square error (MSE) between the measured value and the predicted value is greater than 0.01, readjust the number of neurons and weights, and train and predict the LSTM model again until the mean square error (MSE) is less than 0.01. The mean square error MSE is calculated as follows: , Among them, MSE is the result of mean square error, n is the number of prediction samples, P i For the i Prediction values, T i For the corresponding i An expected value.

[0006] Step 5 verifies the prediction model by performing harmonic balance simulation based on the X-parameter model of the expected data and the predicted data to obtain the standard spectrum and the spectrum of the LSTM model; Step 6 generates the X parameter model by calculating the absolute error between the expected data and the predicted data spectrum to determine whether the modeling is successful. If the spectrum error is less than 15, the X parameter model is generated, otherwise return to the second step to readjust the number and weight of neurons.

[0007] Compared with the prior art, the technical solution of the present invention has the following beneficial effects: The core of the present invention adopts the LSTM modeling method, which introduces the gate control mechanism into the network structure to control the accumulation speed of information and selectively receive or forget known information, which helps to avoid the gradient vanishing problem during the training process. The present invention can accurately characterize the X parameters of microwave power transistors based on limited measurement or simulation data, and can solve the problems of uncertainty, long cycle, large error, etc. in traditional modeling methods, thereby further improving the design efficiency and accuracy of radio frequency circuits. BRIEF DESCRIPTION OF THE DRAWINGS

[0008] Figure 1 The overall modeling method flow of the present invention; Figure 2 is the GaN HEMT transistor X parameter prediction model of the present invention; Figure 3 For the present invention f =1 GHz, P in =36 dBm, V gs =–1.9 V, V ds =12 V GaN HEMT transistor X parameter prediction model fitting results; Figure 4 For the present invention f =1 GHz, P in =36 dBm, V gs =–1.9 V, V ds =12 V GaN HEMT transistor X parameter absolute error curve; Figure 5 For the present invention f =3 GHz, P in =30 dBm, V gs =–3.9 V, V ds =8V GaN HEMT transistor X parameter prediction model fitting results; Figure 6 For the present invention f =3 GHz, P in =30 dBm, V gs =–3.9 V, V ds =8 V GaN HEMT transistor X parameter absolute error curve; Figure 7 For the present inventionf =6 GHz, P in =36 dBm, V gs =–0.9 V, V ds =12 V GaN HEMT transistor X parameter prediction model fitting results; Figure 8 For the present invention f =6 GHz, P in =36 dBm, V gs =–0.9 V, V ds =12 V GaN HEMT transistor X parameter absolute error curve. DETAILED DESCRIPTION

[0009] The principles and features of the present invention are described below in conjunction with the accompanying drawings. The examples given are only used to explain the present invention and are not used to limit the scope of the present invention.

[0010] like Figure 1 As shown, a microwave power transistor X-parameter modeling method based on LSTM technology includes six steps: data preparation, model construction, model training, model testing, prediction model verification, and X-parameter model generation; Step 1: Data preparation is to build a schematic diagram for extracting X parameters in the software, determine the value of the input vector and input it into the X parameter control. By setting the corresponding bias voltage and frequency, a certain number of X parameters are finally obtained after the simulation is completed, and they are divided into training sets and test sets according to appropriate proportions; Step 2: Model construction is to determine the characteristic parameters that affect the performance of microwave power transistors. X = [ X 1 , X 2 , …, X n ] as input variables, X parameter set as output variables, and set the step size of each input variable. After determining the input and output, create the LSTM structure and set the number of neurons and initial weights in the input layer, hidden layer, and output layer of the neural network; Step 3: training the model. After setting the training parameters such as the maximum number of iterations, the optimal learning rate, and the optimal error tolerance, the model starts training. If the training parameters meet the requirements or the number of iterations reaches the maximum number, the training ends. Step 4: Test the model by using the test set to test the simulation results of the model trained in the previous step. If the mean square error (MSE) between the measured value and the predicted value is greater than 0.01, readjust the number of neurons and weights, and train and predict the LSTM model again until the mean square error (MSE) is less than 0.01. Step 5 verifies the prediction model by performing harmonic balance simulation based on the X-parameter model of the expected data and the predicted data to obtain the standard spectrum and the spectrum of the LSTM model; Step 6 generates the X parameter model by calculating the absolute error between the expected data and the predicted data spectrum to determine whether the modeling is successful. If the spectrum error is less than 15, the X parameter model is generated, otherwise return to the second step to readjust the number and weight of neurons.

[0011] In step 4, the mean square error MSE and mean absolute error MAE are calculated as follows: , Among them, MSE is the result of mean square error, n is the number of prediction samples, P i For the i Prediction values, T i For the corresponding i An expected value.

[0012] Embodiments of the present invention.

[0013] Based on the X-parameters of Gallium Nitride High Electron Mobility Transistor (GaN HEMT), a microwave power transistor X-parameter modeling method based on LSTM technology proposed in the present invention is described in detail.

[0014] Step 1: Data preparation is to establish a schematic diagram for extracting X parameters in the software, which mainly consists of a CGH40010F transistor, an input matching circuit, an output matching circuit, a gate bias circuit, and a drain bias circuit. Among them, the gate of the CGH40010F transistor is connected to the input matching circuit and the gate bias circuit, the drain is connected to the output matching circuit and the drain bias circuit, and the source is grounded. Secondly, determine the value of the input vector and input it into the X parameter control. By setting the corresponding bias voltage and frequency, 149×192 groups of X parameters are finally obtained after the simulation is completed, and they are divided into training sets and test sets in a ratio of 8:2; Step 2: The model is constructed by converting [frequency f , P in ，V gs ，Vds ] as input variables, X parameter set as output variables, and set the frequency f 0 GHz - 6 GHz, step size is 1 GHz; P in Set to 30 dBm - 36 dBm in 3 dBm steps; V gs Set to -3.9 V - -0.9 V , the step size is 1 V ; V ds Set to 8 V - 20 V , the step length is 4 V After determining the input and output, create the LSTM structure and set the number of neurons and initial weights in the neural network input layer, hidden layer, and output layer; Step 3: Training the model. After setting the maximum number of iterations to 1000, the optimal learning rate to 0.02, and the optimal error tolerance to 0.005, the model starts training. If each training parameter meets the preset requirements or the number of iterations reaches the maximum number, the training ends. Step 4: Test the model by using the test set to test the simulation results of the model trained in the previous step. If the mean square error (MSE) between the measured value and the predicted value is greater than 0.01, readjust the number of neurons and weights, and train and predict the LSTM model again until the mean square error (MSE) is less than 0.01. That is, the mean square error (MSE) of the X parameter of GaN HEMT predicted by the LSTM model is 0.0056743. Step 5 verifies the prediction model by performing harmonic balance simulation based on the X-parameter model of the expected data and the predicted data to obtain the standard spectrum and the spectrum of the LSTM model; Step 6 generates the X-parameter model by calculating the absolute error between the expected data and the predicted data spectrum to determine whether the modeling is successful. If the spectrum error is less than 15, the X-parameter model is generated, that is, the errors of the third harmonic are 0.801dBm, 7.511 dBm and 13.47 dBm respectively, and the GaN HEMT transistor X-parameter modeling is completed. It can be seen that this method accurately characterizes the X-parameters of the GaN HEMT transistor.

[0015] In summary, this microwave power transistor X-parameter modeling method based on LSTM technology provides an important reference for large signal modeling research and circuit design of microwave power transistors.

[0016] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principle of the present invention should be included in the protection scope of the present invention.

Claims

1. A microwave power transistor X-parameter modeling method based on LSTM technology, including six steps: data preparation, model construction, model training, model testing, prediction model verification, and X-parameter model generation; The data preparation in step 1 is to establish a schematic diagram for extracting X parameters in the software, determine the value of the input vector and input it into the X parameter control. By setting the corresponding bias voltage and frequency, a certain number of X parameters are finally obtained after the simulation is completed, and they are divided into training sets and test sets according to appropriate proportions; The model of step 2 is constructed by: X = [ X 1 , X 2 , …, X n ] as input variables, X parameter set as output variables, and set the step size of each input variable. After determining the input and output, create the LSTM structure and set the number of neurons and initial weights in the input layer, hidden layer, and output layer of the neural network; The step 3 of training the model is to set the training parameters such as the maximum number of iterations, the optimal learning rate and the optimal error tolerance, and then start the model training. If the training parameters meet the requirements or the number of iterations reaches the maximum number, the training ends. The test model in step 4 is to use the test set to test the simulation results of the model trained in the previous step. If the mean square error (MSE) between the measured value and the predicted value is greater than 0.01, the number of neurons and weights are readjusted, and the LSTM model is trained and predicted again until the mean square error (MSE) is less than 0.

01. The step 5 of verifying the prediction model is to perform harmonic balance simulation based on the X-parameter model of the expected data and the predicted data, respectively, to obtain the standard spectrum and the spectrum of the LSTM model; The step 6 of generating the X parameter model is to determine whether the modeling is successful by calculating the absolute error between the expected data and the predicted data spectrum. If the spectrum error is less than 15, the X parameter model is generated, otherwise, the second step is returned to readjust the number and weight of neurons.

2. According to a microwave power transistor X-parameter modeling method based on LSTM technology according to claim 1, it is characterized in that: The mean square error MSE of step 4 can be obtained as follows: ， Among them, MSE is the result of mean square error, n is the number of prediction samples, P i For the i Prediction values, T i For the corresponding i An expected value.

Images