Wiener type dynamic neural network modeling method of transistor and related equipment

Abstract

The invention discloses a Wiener-type dynamic neural network modeling method of a transistor and related equipment, and belongs to the technical field of microwave circuit and device modeling. Firstly, a microwave state equation and a self-heating state equation of the transistor are obtained, and an initial Wiener-type dynamic neural network model of the transistor is established; training the initial Wiener type dynamic neural network model based on the pulse direct current signal, the pulse small signal and the large signal harmonic wave to obtain a Wiener type dynamic neural network model, and verifying the initial Wiener type dynamic neural network model based on the non-pulse direct current signal, the non-pulse small signal and the large signal harmonic wave; a final Wiener type dynamic neural network model is obtained, the modeling efficiency is improved through a training algorithm of the Wiener type dynamic neural network of the high-power transistor under the conditions of direct current, small signals and large signals, the nonlinear characteristic of the self-heating effect of the high-power transistor can be accurately predicted by introducing a self-heating state equation, and the self-heating effect of the high-power transistor can be accurately predicted. And the accuracy of establishing the model is improved.

Description

Technical Field

[0001] The invention belongs to the technical field of microwave circuit and device modeling, and specifically relates to a transistor Wiener-type dynamic neural network modeling method and related equipment. Background Art

[0002] High-power transistors are key devices in microwave communication systems, and their performance directly affects the system's power output, efficiency, and linearity. Under high-power conditions, high-power transistors will produce significant self-heating effects, that is, the device generates heat due to power consumption during operation, causing the temperature of the transistor to rise. The self-heating effect can cause changes in transistor parameters, resulting in nonlinear distortion and affecting its normal operation. Therefore, accurately modeling the self-heating effect is of great significance for improving the performance of high-power transistors.

[0003] Traditional transistor models mainly include physical models and equivalent circuit models. The physical model can accurately describe the internal structure and working mechanism of the transistor, but its establishment and calculation process is complex and time-consuming. The equivalent circuit model simplifies the circuit structure of the transistor and builds a model with similar effects to the actual circuit based on Kirchhoff's law. However, with the continuous advancement of semiconductor technology, traditional physical models and equivalent circuit models often require a lot of experiments and adjustments when facing new materials and processes. The modeling process is cumbersome and the cycle is long. Therefore, the innovation of simulation modeling technology for high-power transistors in wireless communication systems and the improvement of modeling speed while maintaining high accuracy are important research directions in the field of microwave RF device modeling.

[0004] In recent years, artificial neural networks have shown great potential in circuit and device modeling. Artificial neural networks can establish effective behavioral models without understanding the specific internal structure and mechanism of the circuit by learning the nonlinear relationship between input and output data, and have a certain generalization ability for similar circuit structures. As a neural network model containing dynamic characteristics, the Wiener-type dynamic neural network can better approximate nonlinear functions and has strong learning ability. The widely studied dynamic linear or static nonlinear theory can be directly applied to the modeling process instead of using dynamic nonlinear methods such as time delay neural network (TDNN) and state space dynamic neural network (SSDNN). The linear dynamic equations used in the Wiener-type dynamic neural network (Wiener-type DNN) model are simplified and decoupled, where each equation contains only one or two state variables, which is more concise than the traditional Wiener model. By combining linear dynamic systems and static nonlinear systems, the self-heating effect and nonlinear characteristics of high-power transistors can be effectively described. Although some studies have applied dynamic neural networks to the modeling of high-power transistors, under high-power working conditions, the thermal effect becomes obvious and has a greater impact on the performance of the device, resulting in low modeling accuracy and poor stability. Summary of the invention

[0005] The present invention provides a Wiener-type dynamic neural network modeling method for transistors and related equipment, which solves the problem that under high-power working conditions, thermal effects become obvious and have a great impact on device performance, resulting in low modeling accuracy and poor stability.

[0006] To achieve the above object, the present invention provides the following technical solutions: A transistor Wiener-type dynamic neural network modeling method, comprising: Construct the microwave state equation and self-heating state equation of the transistor, and establish the initial Wiener-type dynamic neural network model of the transistor; The initial Wiener type dynamic neural network model is trained based on pulsed DC signals, pulsed small signals and large signal harmonics to obtain the Wiener type dynamic neural network model, and the Wiener type dynamic neural network model is verified based on non-pulsed DC signals, non-pulsed small signals and large signal harmonics to obtain the final Wiener type dynamic neural network model; The final Wiener-type dynamic neural network model is finally verified based on the ADS software, and remodeling or completion of modeling is selected based on the verification results.

[0007] Preferably, the steps of constructing the microwave state equation and the self-heating state equation of the transistor and establishing the initial Wiener-type dynamic neural network model of the transistor are specifically as follows: The pole information of the Wiener-type dynamic neural network model of high-power transistors is obtained by using vector fitting method, and then the structure of microwave state equation and the initial values ​​of microwave state equation coefficients in the model are determined; Obtain the frequency range of the self-heating effect of the transistor and the frequency range of other RF signals, obtain the initial value of the coefficient of the self-heating state equation according to the frequency range of the self-heating effect and the frequency range of other RF signals, and then use the pulsed DC data to train the neural network at different temperatures; Under pulse conditions, a self-heating state variable is obtained by performing a DC simulation from a known non-pulsed DC condition using a trained neural network, and another initial value of the self-heating state equation coefficient is determined according to the self-heating state variable, the self-heating state equation coefficient and an input signal of the self-heating state equation; Set the number of neurons in the hidden layer of the neural network, initialize their weights, and build an initial Wiener-type dynamic neural network model.

[0008] Preferably, the initial Wiener type dynamic neural network model is trained based on the pulsed DC signal, the pulsed small signal and the large signal harmonics to obtain the Wiener type dynamic neural network model, and the model is verified based on the non-pulsed DC signal, the non-pulsed small signal and the large signal harmonics. The steps of obtaining the Wiener type dynamic neural network model are specifically as follows: At different temperatures, the initial Wiener-type dynamic neural network model is trained with pulsed DC signals and pulsed small signal parameter data. During the training process, the coefficients of the linear dynamic equations and the weights in the multilayer perceptron neural network are adjusted so that the output of the constructed network model fitted under DC and small signals is close to the actual measured data of high-power transistors. The model is verified by using non-pulse DC signal and non-pulse small signal parameter data, and the error of the Wiener-type dynamic neural network between input and output under non-pulse DC signal and non-pulse small signal is verified by using test data. If the error does not meet the accuracy requirement, the Wiener-type dynamic neural network is trained or retrained on the original basis. If the test error meets the accuracy requirement, the Wiener-type dynamic neural network model under DC and small signal conditions is obtained, and the next step is carried out; Use pulsed DC signals, pulsed small signal parameter data and large signal harmonic data to simultaneously train the Wiener-type dynamic neural network model, and use the results of the linear dynamic equation coefficients in the Wiener-type dynamic neural network model under DC and small signal conditions as the initial training values, so that the output of the constructed network model fitted under pulsed DC signals, pulsed small signals and large signal harmonic data is close to the actual measured data of the transistor; The model is verified by using non-pulse DC signal, non-pulse small signal parameter data and large signal harmonic test data. The error of the Wiener type dynamic neural network between input and output under non-pulse DC signal, non-pulse small signal parameter data and large signal harmonic data is verified by using the test data. If the error does not meet the accuracy requirements, the Wiener type dynamic neural network is continued to be trained or retrained on the original basis. If the test error meets the requirements, the Wiener type dynamic neural network model is obtained.

[0009] Preferably, the DC and pulse small signal formulas for training the initial Wiener type dynamic neural network model using pulse DC signal and pulse small signal parameter data are:

[0010]

[0011] in, and Represents the input voltage and output current under non-pulsed DC conditions, where , , and They represent the gate voltage, drain voltage, gate current and drain current under non-pulsed DC respectively; Indicates the input ambient temperature; , , and represents the coefficients of the microwave state equation; and denote the microwave state variables and self-heating state variables in the DC case respectively; and represents the coefficient of the self-heating state equation; let represents the neural network between input voltage and output current, is a vector containing all the internal weights of the neural network; , , represents the number of real poles in the model, represents the number of complex conjugate pole pairs in the model; in addition, in the DC case, and The derivatives with respect to time are all 0; , and They represent the first-order partial derivatives of the neural network function with respect to the input voltage, the real poles of the microwave state variables, and the complex conjugate poles of the microwave state variables under DC bias, respectively. , ; , Indicates frequency.

[0012] Preferably, the large signal harmonic data training formula for transistor self-heating effect modeling training is:

[0013] in , , represents the number of real poles in the model, represents the number of complex conjugate pole pairs in the model, represents a common harmonic frequency, where Indicates the harmonic number, , is the number of harmonics considered in the harmonic balance simulation, and They represent the input voltage and output current in the frequency domain, and Represents microwave state variables and self-heating state variables in the frequency domain; , , and are the coefficients of the microwave state equation; and Represents the coefficients of the self-heating state equation.

[0014] Preferably, the Wiener-type dynamic neural network model is finally verified based on the ADS software, and the steps of remodeling or completing the modeling are selected according to the verification results: A verification circuit for the Wiener-type dynamic neural network model is established in ADS. The trained linear dynamic equation coefficients and neural network formulas are placed in the SDD device. The verification circuit is analyzed within the same range. The model output results are compared with the actual measured data of the high-power transistor. Remodeling or completion of modeling is selected based on the comparison results.

[0015] Preferably, the formula of the initial Wiener-type dynamic neural network model is:

[0016] in, , , represents the number of real poles in the model, Represents the number of complex conjugate pole pairs in the model. In the model, the linear dynamic equation is divided into two parts: the microwave state equation is composed of the first two sets of equations in (1) that describe the microwave information, and the self-heating state equation is composed of the third equation in (1) that represents the self-heating effect information, where , and Respectively represent the input voltage, input ambient temperature and output current of the model; t Indicates time; , , and represents the coefficients of the microwave state equation; and represents microwave state variables; and Respectively and About time The derivative of and They represent the self-heating state variable and the self-heating state variable with respect to time. The derivative of and represents the coefficients of the self-heating state equation; represents a neural network between input and output, is a vector containing all the internal weights of the neural network.

[0017] A transistor Wiener-type dynamic neural network modeling system, comprising: Building module: used to build the microwave state equation and self-heating state equation of the transistor, and establish the initial Wiener-type dynamic neural network model of the transistor; Training and verification module: used to train the initial Wiener-type dynamic neural network model based on pulsed DC signals, pulsed small signals and large signal harmonics to obtain the Wiener-type dynamic neural network model, and to verify the Wiener-type dynamic neural network model based on non-pulsed DC signals, non-pulsed small signals and large signal harmonics to obtain the final Wiener-type dynamic neural network model; Verification module: used to perform final verification of the Wiener-type dynamic neural network model based on ADS software, and choose to remodel or complete the modeling based on the verification results.

[0018] A computer device comprises a memory, a processor and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, the steps of a transistor Wiener-type dynamic neural network modeling method are implemented.

[0019] A computer-readable storage medium stores a computer program, which, when executed by a processor, implements the steps of a transistor Wiener-type dynamic neural network modeling method.

[0020] Compared with the prior art, the present invention has the following beneficial effects: the present invention provides a Wiener-type dynamic neural network modeling method for a transistor, firstly obtaining the microwave state equation and the self-heating state equation of the transistor, establishing an initial Wiener-type dynamic neural network model of the transistor, then training the initial Wiener-type dynamic neural network model based on pulsed DC signals, pulsed small signals and large signal harmonics to obtain a Wiener-type dynamic neural network model, and verifying the initial Wiener-type dynamic neural network model based on non-pulsed DC signals, non-pulsed small signals and large signal harmonics to obtain a final Wiener-type dynamic neural network model, and the training algorithm of the Wiener-type dynamic neural network of a high-power transistor under DC, small signal and large signal conditions improves the modeling efficiency, and by introducing the self-heating state equation, the nonlinear characteristics of the self-heating effect of the high-power transistor can be accurately predicted, thereby improving the accuracy of modeling. BRIEF DESCRIPTION OF THE DRAWINGS

[0021] Figure 1 A flow chart of a transistor Wiener-type dynamic neural network modeling method according to the present invention; Figure 2 This is a structural block diagram of the model of an embodiment of the present invention; Figure 3 is a flow chart for modeling a high-power transistor according to an embodiment of the present invention; Figure 4 is a comparison diagram of the DC curves of the original data of the embodiment of the present invention, the output of the proposed model and the output of the equivalent circuit model; Figure 5is a comparison diagram of non-pulse small signal curves of the original data of the embodiment of the present invention, the output of the proposed model and the output of the equivalent circuit model; Figure 6 is a large signal curve comparison diagram of the original data of the embodiment of the present invention, the output of the proposed model and the output of the equivalent circuit model; Figure 7 This is a block diagram of a transistor Wiener-type dynamic neural network modeling system of the present invention. DETAILED DESCRIPTION

[0022] In order to make the purpose, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Generally, the components of the embodiments of the present invention described and shown in the drawings here can be arranged and designed in various different configurations.

[0023] Therefore, the following detailed description of the embodiments of the present invention provided in the accompanying drawings is not intended to limit the scope of the invention claimed for protection, but merely represents selected embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0024] It should be noted that similar reference numerals and letters denote similar items in the following drawings, and therefore, once an item is defined in one drawing, further definition and explanation thereof is not required in subsequent drawings.

[0025] In order to enable those skilled in the art to better understand the technical solution of the present invention, the present invention will be further described in detail below with reference to the accompanying drawings.

[0026] like Figure 1 As shown, an embodiment of the present invention provides a Wiener-type dynamic neural network modeling method for a transistor, comprising: S101 constructs the microwave state equation and self-heating state equation of the transistor, and establishes the initial Wiener-type dynamic neural network model of the transistor; S102 trains the initial Wiener type dynamic neural network model based on the pulsed DC signal, the pulsed small signal and the large signal harmonics to obtain the Wiener type dynamic neural network model, and verifies the Wiener type dynamic neural network model based on the non-pulsed DC signal, the non-pulsed small signal and the large signal harmonics to obtain the final Wiener type dynamic neural network model; S103 performs final verification of the Wiener-type dynamic neural network model based on ADS software, and chooses to remodel or complete the modeling based on the verification results.

[0027] The detailed steps are: Step 1: Use the vector fitting method to obtain the pole information of the high-power transistor Wiener-type dynamic neural network model and then determine the structure of the microwave state equation and the initial values ​​of the microwave state equation coefficients in the model; Step 2: Determine the initial value of the coefficient of the self-heating state equation in the high-power transistor model. Under the condition that the time is approximately 1 millisecond, the self-heating effect of the high-power transistor is within the kHz frequency range, and other RF signals are within the GHz frequency range, so the coefficient of the self-heating state equation is The initial value of is set to -1 / 1000, and then the neural network is trained using pulsed DC data at different temperatures; Step 3: Under pulse conditions, the self-heating state variables The value is equal to the ambient temperature , the self-heating state variables are obtained by DC simulation from known non-pulsing DC conditions using a trained neural network , and considering that the derivative of the self-heating state variable with respect to time is 0, the coefficient of the self-heating state equation in the self-heating state equation is , self-heating state variables and the three input signals are known, the coefficients of the self-heating state equation can be determined The initial value of Step 4: Set the number of neurons in the hidden layer of the multilayer perceptron neural network, initialize their weights, and construct a Wiener-type dynamic neural network model of the self-heating effect of high-power transistors; Step 5: Use pulsed DC and pulsed small signal at different temperatures The parameter data is used to train the Wiener-type dynamic neural network model of the high-power transistor. During the training process, the coefficients of the linear dynamic equation and the weights in the multilayer perceptron neural network are adjusted so that the output of the constructed network model fitted under DC and small signals is close to the actual measured data of the high-power transistor. Step 6: Using non-pulsed DC and non-pulsed small signals Parameter data verification model, using test data to verify the error between input and output of the Wiener-type dynamic neural network under non-pulse DC and non-pulse small signal. If the error does not meet the accuracy requirements, continue to train or retrain the Wiener-type dynamic neural network on the original basis. The training can be adjusted by analyzing the sensitivity and updating the weights and coefficients of the neural network. If the test error remains within a reasonable range, proceed to the next step; Step 7: Use pulsed DC and pulsed small signal The parameter data and large signal harmonic data are used to train the Wiener-type dynamic neural network model of the high-power transistor at the same time. The result of the linear dynamic equation coefficient obtained from step 6 can be used as the initial value of the training in step 7, so that the output of the constructed network model fitted under pulsed DC, pulsed small signal and large signal is close to the actual measured data of the high-power transistor; Step 8: Use non-pulse DC and non-pulse small signal The parameter data and large signal harmonic test data verify the model. The test data is used to verify the error of the Wiener-type dynamic neural network between input and output under non-pulse DC, non-pulse small signal and large signal. If the error does not meet the accuracy requirement, the Wiener-type dynamic neural network is trained or retrained on the original basis. The training can be adjusted by analyzing the sensitivity and updating the weights and coefficients of the neural network. If the test error remains within a reasonable range, the modeling of the self-heating effect of the high-power transistor is completed. Step 9: Establish a verification circuit of the Wiener-type dynamic neural network model of the high-power transistor in ADS, place the trained linear dynamic equation coefficients and neural network formulas in the SDD device, analyze the verification circuit within the same range, and compare the model output results with the actual measurement data of the high-power transistor.

[0028] In step 4 of the invention, the formula of the Wiener-type dynamic neural network model suitable for modeling the self-heating effect of high-power transistors is: (1) in, , , represents the number of real poles in the model, Represents the number of complex conjugate pole pairs in the model. In the model, the linear dynamic equation is divided into two parts: the microwave state equation is composed of the first two sets of equations in (1) that describe the microwave information, and the self-heating state equation is composed of the third equation in (1) that represents the self-heating effect information, where , and Respectively represent the input voltage, input ambient temperature and output current of the model; t Indicates time; , , and represents the coefficients of the microwave state equation; and represents microwave state variables; and Respectively and About time The derivative of and They represent the self-heating state variable and the self-heating state variable with respect to time. The derivative of and represents the coefficients of the self-heating state equation; represents a neural network between input and output, is a vector containing all the internal weights of the neural network.

[0029] In step 5 of the invention, the DC and small signal formulas of the Wiener model used for modeling and training of the self-heating effect of high-power transistors are: (2) (3) in, and Represents the input voltage and output current under non-pulsed DC conditions, where , , and They represent the gate voltage, drain voltage, gate current and drain current under non-pulsed DC respectively; Indicates the ambient temperature; , , and represents the coefficients of the microwave state equation; and denote the microwave state variables and self-heating state variables in the DC case respectively; and represents the coefficient of the self-heating state equation; let represents the neural network between input voltage and output current, is a vector containing all the internal weights of the neural network; , , represents the number of real poles in the model, represents the number of complex conjugate pole pairs in the model; in addition, in the DC case, and The derivatives with respect to time are all 0; in the small signal formula (3), 𝒀 represents the small signal transmission equation of the system; , and They represent the DC bias (where , and ) The first-order partial derivatives of the neural network function with respect to the input voltage, the real poles of the microwave state variables, and the complex conjugate poles of the microwave state variables, where , ; , Indicates frequency.

[0031] In step 7 of the invention, the large signal formula of the Wiener model trained on the self-heating effect modeling of the high-power transistor is: (4) in , , represents the number of real poles in the model, represents the number of complex conjugate pole pairs in the model, represents a common harmonic frequency, where Indicates the harmonic number, , is the number of harmonics considered in the harmonic balance simulation, and They represent the input voltage and output current in the frequency domain, and Represents microwave state variables and self-heating state variables in the frequency domain; , , and are the coefficients of the microwave state equation; and Represents the coefficients of the self-heating state equation.

[0032] (5) (6) (7) in For the A point in time, For the time point and Fourier coefficients at subharmonic frequency points, superscript * indicates conjugate complex number, , and They are the time domain information of input voltage, microwave state variables and self-heating state variables respectively.

[0033] (8) in represents the output current in the frequency domain, Indicates the total number of time points.

[0034] The method constructs a new Wiener-type dynamic neural network model structure by introducing ambient temperature, self-heating state variables and self-heating state equations, proposes a training algorithm suitable for this new structure, and derives a new Wiener-type dynamic neural network training formula suitable for DC, small signal and large signal data formats. The real and imaginary data of the small signal parameters of the high-power transistor are processed by the vector fitting method, and the initial values ​​of the microwave state equation coefficients are determined by simplifying the linear dynamic equations according to the obtained pole information, and the initial values ​​of the self-heating state equation coefficients are determined by non-pulse DC simulation. The Wiener-type dynamic neural network of the high-power transistor is constructed in combination with a multilayer perceptron. The method is suitable for modeling high-power transistors under the influence of self-heating effects. The present invention proposes a training algorithm for the Wiener-type dynamic neural network of high-power transistors under DC, small signal and large signal conditions, which improves the modeling efficiency. The high-power transistor example proves that the model has good accuracy and convergence, and can accurately predict the nonlinear characteristics of the self-heating effect of high-power transistors. Another embodiment of the present invention provides a transistor Wiener-type dynamic neural network modeling method, comprising: like Figure 2 As shown in the figure, in the model, the linear dynamic equation is divided into two parts: the microwave state equation, which consists of the first two equations in (1) and describes the microwave information; the self-heating state equation, which consists of the third equation in (1) and represents the self-heating effect information. In addition, the neural network function represents the nonlinear static module.

[0035] like Figure 3 As shown, the modeling method of the present invention first uses a vector fitting method to process Parameter data, simplify the system matrix according to the pole information obtained from the vector fitting method, determine the number of microwave state variables and the structure of the microwave state equation, and then determine the initial values ​​of the microwave state equation coefficients. Then determine the initial values ​​of the thermal parameters related to the self-heating effect of the high-power transistor. Under the condition that the time is approximately 1 millisecond, the self-heating effect of the high-power transistor is within the kHz frequency range, and other RF signals are within the GHz frequency range, so the self-heating state equation coefficients The initial value of is set to -1 / 1000. Then the neural network is trained using pulsed DC data at different temperatures. The self-heating state variables are obtained by DC simulation under known non-pulsed DC conditions using the trained neural network. . Considering that the derivative of the self-heating state variable with respect to time is 0, the coefficient of the self-heating state equation in the self-heating state equation is , self-heating state variables and the three input signals are known, the coefficients of the self-heating state equation can be determined The initial value of . Set the number of neurons in the hidden layer of the multilayer perceptron neural network and initialize its weights to build a Wiener-type dynamic neural network model of the self-heating effect of high-power transistors. The parameter data is used to train the Wiener-type dynamic neural network model of high-power transistors. During the training process, the coefficients of the linear dynamic equations and the weights in the multilayer perceptron neural network are adjusted so that the output of the high-power transistor model fitted under pulsed DC and pulsed small signals is close to the actual measured data of the high-power transistor. Parameter data verification model. Use test data to verify the error of the high-power transistor model between input and output under non-pulse DC and non-pulse small signal. If the error does not meet the accuracy requirements, continue to train or retrain the high-power transistor model on the original basis. The training can be adjusted by analyzing the sensitivity and updating the weights and coefficients of the neural network. If the test error remains within a reasonable range, proceed to the next step.

[0036] Use pulsed DC, pulsed small signal The parameter data and large signal harmonic data are used to train the new Wiener-type dynamic neural network at the same time. The linear dynamic equation coefficients obtained from the previous step can be used as the initial value of this step of training, so that the output of the new Wiener-type dynamic neural network model under pulsed DC, pulsed small signal and large signal fits close to the actual measured data of high-power transistors under self-heating effect. The parameter data and large signal harmonic test data verify the high-power transistor model. The test data is used to verify the error of the high-power transistor Wiener-type dynamic neural network between input and output under non-pulse DC, non-pulse small signal and large signal harmonic data. If the error does not meet the accuracy requirements, the high-power transistor model is trained or retrained on the original basis. The training can be adjusted by analyzing the sensitivity and updating the weights and coefficients of the neural network. If the test error remains within a reasonable range, the modeling of the self-heating effect of the high-power transistor is completed.

[0037] Figures 4 to 6 This is a comparison diagram of the output characteristic curve of the model built by the modeling method of the present invention for a high-power transistor, the original measurement data and the equivalent circuit model. Figure 4 A comparison diagram of the output of the model built by the modeling method of the present invention for a high-power transistor, the original high-power transistor measurement data and the DC curve output of the equivalent circuit model; Figure 5 The non-pulse small signal curve comparison diagram of the output of the model built by the modeling method of the present invention for the high-power transistor, the original high-power transistor measurement data and the output of the equivalent circuit model, respectively comparing the four small signal S parameters S11, S12, S21 and S22; Figure 6This is a curve comparison diagram of the output of the model built by the modeling method of the present invention for the high-power transistor, the original high-power transistor measurement data and the output power baseband signal output by the equivalent circuit model. It can be seen that the output DC, small signal and large signal curves of the high-power transistor Wiener-type dynamic neural network model have a high degree of match with the high-power transistor measured data.

[0038] Among them, Device Data is the original high-power transistor measurement data; Equivalent Circuit Model is an equivalent circuit model; Proposed Wiener-type DNN Model is the proposed Wiener-type dynamic neural network model.

[0039] like Figure 7 As shown, an embodiment of the present invention further provides a transistor Wiener-type dynamic neural network modeling system, comprising: Building module: used to build the microwave state equation and self-heating state equation of the transistor, and establish the initial Wiener-type dynamic neural network model of the transistor; Training and verification module: used to train the initial Wiener-type dynamic neural network model based on pulsed DC signals, pulsed small signals and large signal harmonics to obtain the Wiener-type dynamic neural network model, and to verify the Wiener-type dynamic neural network model based on non-pulsed DC signals, non-pulsed small signals and large signal harmonics to obtain the final Wiener-type dynamic neural network model; Verification module: used to perform final verification of the Wiener-type dynamic neural network model based on ADS software, and choose to remodel or complete the modeling based on the verification results.

[0040] A terminal device is provided in one embodiment of the present invention. The terminal device of this embodiment includes: a processor, a memory, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, the steps in the above-mentioned method embodiments are implemented. Alternatively, when the processor executes the computer program, the functions of the modules / units in the above-mentioned device embodiments are implemented.

[0041] The computer program may be divided into one or more modules / units, and the one or more modules / units are stored in the memory and executed by the processor to accomplish the present invention.

[0042] The terminal device may be a computing device such as a desktop computer, a notebook, a PDA, a cloud server, etc. The terminal device may include, but is not limited to, a processor and a memory.

[0043] The processor may be a central processing unit (CPU), or other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), field-programmable gate arrays (FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc.

[0044] The memory may be used to store the computer programs and / or modules, and the processor implements various functions of the terminal device by running or executing the computer programs and / or modules stored in the memory and calling the data stored in the memory.

[0045] If the module / unit integrated in the terminal device is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the present invention implements all or part of the processes in the above-mentioned embodiment method, and can also be completed by instructing the relevant hardware through a computer program. The computer program can be stored in a computer-readable storage medium. When the computer program is executed by the processor, the steps of the above-mentioned various method embodiments can be implemented. Among them, the computer program includes computer program code, and the computer program code can be in source code form, object code form, executable file or some intermediate form. The computer-readable medium may include: any entity or device capable of carrying the computer program code, recording medium, U disk, mobile hard disk, disk, optical disk, computer memory, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), electric carrier signal, telecommunication signal and software distribution medium. It should be noted that the content contained in the computer-readable medium can be appropriately increased or decreased according to the requirements of legislation and patent practice in the jurisdiction. For example, in some jurisdictions, according to legislation and patent practice, computer-readable media do not include electric carrier signals and telecommunication signals.

Claims

1. A transistor Wiener-type dynamic neural network modeling method, characterized in that: include: Construct the microwave state equation and self-heating state equation of the transistor, and establish the initial Wiener-type dynamic neural network model of the transistor; The initial Wiener type dynamic neural network model is trained based on pulsed DC signals, pulsed small signals and large signal harmonics to obtain the Wiener type dynamic neural network model, and the Wiener type dynamic neural network model is verified based on non-pulsed DC signals, non-pulsed small signals and large signal harmonics to obtain the final Wiener type dynamic neural network model; The final Wiener-type dynamic neural network model is finally verified based on the ADS software, and remodeling or completion of modeling is selected based on the verification results.

2. The transistor Wiener-type dynamic neural network modeling method according to claim 1, characterized in that: The specific steps of constructing the microwave state equation and self-heating state equation of the transistor and establishing the initial Wiener-type dynamic neural network model of the transistor are: The pole information of the Wiener-type dynamic neural network model of high-power transistors is obtained by using vector fitting method, and then the structure of microwave state equation and the initial values ​​of microwave state equation coefficients in the model are determined; Obtain the frequency range of the self-heating effect of the transistor and the frequency range of other RF signals, obtain the initial value of the coefficient of the self-heating state equation according to the frequency range of the self-heating effect and the frequency range of other RF signals, and then use the pulsed DC data to train the neural network at different temperatures; Under pulse conditions, a self-heating state variable is obtained by performing a DC simulation from a known non-pulsed DC condition using a trained neural network, and another initial value of the self-heating state equation coefficient is determined according to the self-heating state variable, the self-heating state equation coefficient and an input signal of the self-heating state equation; Set the number of neurons in the hidden layer of the neural network, initialize their weights, and build an initial Wiener-type dynamic neural network model.

3. The transistor Wiener-type dynamic neural network modeling method according to claim 1, characterized in that: The initial Wiener-type dynamic neural network model is trained based on pulsed DC signals, pulsed small signals and large signal harmonics to obtain the Wiener-type dynamic neural network model, and the model is verified based on non-pulsed DC signals, non-pulsed small signals and large signal harmonics. The specific steps for obtaining the Wiener-type dynamic neural network model are: At different temperatures, the initial Wiener-type dynamic neural network model is trained with pulsed DC signals and pulsed small signal parameter data. During the training process, the coefficients of the linear dynamic equations and the weights in the multilayer perceptron neural network are adjusted so that the output of the constructed network model fitted under DC and small signals is close to the actual measured data of high-power transistors. The model is verified by using non-pulse DC signal and non-pulse small signal parameter data, and the error of the Wiener-type dynamic neural network between input and output under non-pulse DC signal and non-pulse small signal is verified by using test data. If the error does not meet the accuracy requirement, the Wiener-type dynamic neural network is trained or retrained on the original basis. If the test error meets the accuracy requirement, the Wiener-type dynamic neural network model under DC and small signal conditions is obtained, and the next step is carried out; Use pulsed DC signals, pulsed small signal parameter data and large signal harmonic data to simultaneously train the Wiener-type dynamic neural network model, and use the results of the linear dynamic equation coefficients in the Wiener-type dynamic neural network model under DC and small signal conditions as the initial training values, so that the output of the constructed network model fitted under pulsed DC signals, pulsed small signals and large signal harmonic data is close to the actual measured data of the transistor; The model is verified by using non-pulse DC signal, non-pulse small signal parameter data and large signal harmonic test data. The error of the Wiener type dynamic neural network between input and output under non-pulse DC signal, non-pulse small signal parameter data and large signal harmonic data is verified by using the test data. If the error does not meet the accuracy requirements, the Wiener type dynamic neural network is continued to be trained or retrained on the original basis. If the test error meets the requirements, the Wiener type dynamic neural network model is obtained.

4. The transistor Wiener-type dynamic neural network modeling method according to claim 3, characterized in that: The DC and pulse small signal formulas for training the initial Wiener type dynamic neural network model using pulse DC signal and pulse small signal parameter data are: in, and Represents the input voltage and output current under non-pulsed DC conditions, where , , and They represent the gate voltage, drain voltage, gate current and drain current under non-pulsed DC respectively; Indicates the input ambient temperature; , , and represents the coefficients of the microwave state equation; and denote the microwave state variables and self-heating state variables in the DC case respectively; and represents the coefficient of the self-heating state equation; let represents the neural network between input voltage and output current, is a vector containing all the internal weights of the neural network; , , represents the number of real poles in the model, represents the number of complex conjugate pole pairs in the model; in addition, in the DC case, and The derivatives with respect to time are all 0; , and They represent the first-order partial derivatives of the neural network function with respect to the input voltage, the real poles of the microwave state variables, and the complex conjugate poles of the microwave state variables under DC bias, respectively. , ; , Indicates frequency.

5. The transistor Wiener-type dynamic neural network modeling method according to claim 3, characterized in that: The large signal harmonic data training formula for transistor self-heating effect modeling training is: in , , represents the number of real poles in the model, represents the number of complex conjugate pole pairs in the model, represents a common harmonic frequency, where Indicates the harmonic number, , is the number of harmonics considered in the harmonic balance simulation, and They represent the input voltage and output current in the frequency domain, and Represents microwave state variables and self-heating state variables in the frequency domain; , , and are the coefficients of the microwave state equation; and Represents the coefficients of the self-heating state equation.

6. The transistor Wiener-type dynamic neural network modeling method according to claim 1, characterized in that: The Wiener-type dynamic neural network model is finally verified based on the ADS software. The steps for remodeling or completing the modeling are as follows: A verification circuit for the Wiener-type dynamic neural network model is established in ADS. The trained linear dynamic equation coefficients and neural network formulas are placed in the SDD device. The verification circuit is analyzed within the same range. The model output results are compared with the actual measured data of the high-power transistor. Remodeling or completion of modeling is selected based on the comparison results.

7. The transistor Wiener-type dynamic neural network modeling method according to claim 1, characterized in that: The formula of the initial Wiener-type dynamic neural network model is: in, , , represents the number of real poles in the model, represents the number of complex conjugate pole pairs in the model, where , and Respectively represent the input voltage, input ambient temperature and output current of the model; t Indicates time; , , and represents the coefficients of the microwave state equation; and represents microwave state variables; and Respectively and About time The derivative of and They represent the self-heating state variable and the self-heating state variable with respect to time. The derivative of and represents the coefficients of the self-heating state equation; represents a neural network between input and output, is a vector containing all the internal weights of the neural network.

8. A transistor Wiener-type dynamic neural network modeling system, characterized in that: include: Building module: used to construct the microwave state equation and self-heating state equation of the transistor, and establish the initial Wiener-type dynamic neural network model of the transistor; Training and verification module: used to train the initial Wiener-type dynamic neural network model based on pulsed DC signals, pulsed small signals and large signal harmonics to obtain the Wiener-type dynamic neural network model, and to verify the Wiener-type dynamic neural network model based on non-pulsed DC signals, non-pulsed small signals and large signal harmonics to obtain the final Wiener-type dynamic neural network model; Verification module: used to perform final verification of the Wiener-type dynamic neural network model based on ADS software, and choose to remodel or complete the modeling based on the verification results.

9. A computer device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that: When the processor executes the computer program, the steps of the Wiener-type dynamic neural network modeling method for a transistor as described in any one of claims 1 to 7 are implemented.

10. A computer-readable storage medium storing a computer program, characterized in that: When the computer program is executed by a processor, the steps of a transistor Wiener-type dynamic neural network modeling method as described in any one of claims 1 to 7 are implemented.

Images