Microwave transistor small signal modeling method and system
By establishing a small-signal model of microwave transistors using a Kolmogorov-Arnold network, the problems of gradient vanishing and training difficulty in microwave transistor modeling using multilayer perceptron neural networks were solved, achieving higher modeling accuracy and simplicity.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-12-23
- Publication Date
- 2026-04-03
AI Technical Summary
In existing technologies, multilayer perceptron neural networks require a large number of hidden layers and hidden neurons when modeling microwave transistors, which leads to the vanishing gradient problem and increased training difficulty, limiting the improvement of model accuracy. In particular, its application in the microwave field has not been fully explored.
A Kolmogorov-Arnold network was used to establish a model training architecture based on small-signal S-parameter data under DC and multi-bias conditions. A microwave transistor small-signal model was built through a network structure with 2 inputs, 4 outputs, and 2 hidden layers, and the network parameters were adjusted by training error.
It improves the accuracy and generalization ability of small-signal modeling of microwave transistors, outperforming the performance of traditional multilayer perceptron neural networks, and achieving higher fitting accuracy and a simpler network structure.
Smart Images
Figure CN121787346A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of microwave circuit and device modeling technology, specifically a small-signal modeling method for microwave transistors. Background Technology
[0002] Accurate microwave transistor models are crucial for optimizing circuit performance and reducing design iterations. Multilayer perceptron (MLP) neural networks, with their superior nonlinear fitting capabilities, have been widely used in transistor modeling. However, when microwave transistors exhibit strong nonlinear characteristics in the modeling scenario, MLPs require a large number of hidden layers and hidden neurons to effectively capture complex input-output mappings. This structural complexity not only leads to the vanishing gradient problem but also significantly increases the difficulty of model training, ultimately becoming a key bottleneck restricting the improvement of model accuracy.
[0003] In recent years, Kolmogorov-Arnold networks (KANs) have emerged as a groundbreaking paradigm in neural network architecture, providing a highly promising and efficient alternative to traditional microprocessor-based power logic (MLPs). This network replaces traditional fixed activation functions with learnable unary functions, significantly improving model expressiveness and interpretability while achieving higher fitting accuracy with a more streamlined network structure. In the compact modeling task of DC current and charge characteristics of FinFETs, KANs have demonstrated significantly better performance than MLPs and industry-standard models. Furthermore, their application potential has been successfully extended to multiple circuit fields such as digital predistortion (DPD) for RF power amplifiers. However, research on KANs in the microwave field is still in its early exploratory stages, undoubtedly representing an important direction worthy of further exploration. No methods for building microwave transistor models using KANs have been reported.
[0004] Therefore, the purpose of this invention is to propose a novel small-signal modeling method for microwave FET transistors based on Kolmogorov-Arnold networks, which aims to further improve the model accuracy. Summary of the Invention
[0005] This invention aims to at least solve one of the technical problems existing in the prior art. Therefore, one objective of this invention is to propose a small-signal modeling method and system for microwave transistors. By establishing a model training architecture based on small-signal S-parameter data under DC and multi-bias conditions, small-signal modeling of microwave transistors is achieved, thereby further improving modeling accuracy.
[0006] To address the above problems, this invention provides a microwave transistor small-signal modeling method, comprising the following steps: S1: Obtain the core characteristic data of the microwave transistor to be modeled through actual measurement, including small signal S-parameter data under DC and multiple bias points, divide the data into training dataset and test dataset, and set the training error threshold. S2: Construct a Kolmogorov-Arnold network with 2 inputs, 4 outputs, and 2 hidden layers, and its adjoint network with 2 inputs and 8 outputs. The inputs of the Kolmogorov-Arnold network are the gate voltage u1 and the drain voltage u2, and the outputs are the gate current i1, the drain current i2, the gate charge q1, and the drain charge q2. The output of the adjoint network is the partial derivative of the output of the Kolmogorov-Arnold network with respect to the input. S3: Build the training architecture for the small-signal model. The inputs to this architecture are the gate voltage u1, drain voltage u2, and frequency freq. The outputs are the DC output gate current I1 and drain current I2, as well as the small-signal S-parameters and the input reflection coefficient S. 11 1. Reverse transmission coefficient 2. Forward transmission coefficient S 21 and output reflection coefficient S 22 ; S4: Train the training architecture based on the small-signal S-parameter training dataset under DC and multiple bias points, and calculate the training error E after training. train ; S5: If the training error E train Greater than the training error threshold If the condition is met, the number of hidden layer neurons is increased, and the process returns to S2 to rebuild the network; otherwise, proceed to S6. S6: Test the training architecture using the test dataset and calculate the test error E. test ; S7: If the test error E test Greater than the training error threshold If the training data is correct, then add training data and return to S4 for retraining; otherwise, proceed to S8. S8: Update weights to obtain And derive the trained microwave transistor small-signal model.
[0007] Preferably, the expression for the Kolmogorov-Arnold network described in S2 is: (1) In the formula: o1, o2, o3 and o4 are the outputs i1, i2, q1 and q2 of the Kolmogorov-Arnold network, respectively; Let represent the k-th order B-spline curves between the j-th hidden neuron in the first hidden layer and the i-th input, the k-th hidden neuron in the second hidden layer and the j-th hidden neuron in the first hidden layer, and the k-th hidden neuron in the second hidden layer and the l-th output, respectively. The k-th order B-spline curves are formed by... The calculation yields c. m Let c be the learnable coefficients of the B-spline curve. m The vector formed is denoted as The symbol m represents the index of the learnable coefficient; the symbol M represents the total number of learnable coefficients, where M = G + k, and G is the number of grids in the B-spline function.
[0008] Preferably, the expression for the adjoint network in S2 is: (2) in, yes The derivative of .
[0009] Preferably, the DC output gate current I1 and drain current I2 mentioned in S3 are obtained by formula (1) under DC conditions: u1=U1, u2=U2.
[0010] Preferably, the small-signal S-parameter output in S3 is transmitted through the intermediate variable admittance parameter Y. 11 Y 12 Y 21 Y 22 After YS transformation, the admittance parameter is calculated using the following formula: (3) in, Angular frequency, equal to Multiply by the third input freq of the Kolmogorov-Arnold network.
[0011] Preferably, the training error E in S4 train Calculated using the following formula: (4) in, , and These are the DC output and small-signal S-parameter output of the training architecture, respectively. D and S D For measured data, G and M are diagonal matrices, with scaling factors equal to the reciprocals of the corresponding data ranges. k and j are the bias point and frequency point indices, respectively. train and J train These represent the total number of biases and frequencies in the training set, respectively.
[0012] Preferably, the test error E in S6 test Calculated using the following formula: (5) Among them, K test and J testThese represent the total number of biases and frequencies in the test set, respectively.
[0013] Preferably, in the small-signal model of the microwave transistor derived in S8, the voltage between the gate and the source is u1, and the current flowing into the gate is denoted as u1. The voltage between the drain and source is u2, and the current flowing into the drain is denoted as u2. A current source and a charge element are connected in parallel between the gate and the source, respectively, and f in formula (1) is built in. KAN1 and f KAN3 The expression for the current source and charge element are connected in parallel between the drain and source, with each having a built-in f. KAN2 and f KAN4 The expression for the current is: (6) (7) A microwave transistor small-signal modeling system, comprising: The data acquisition module is used to acquire the core characteristic data of the microwave transistor to be modeled through actual measurement, including small-signal S-parameter data under DC and multiple bias points, and divide the data into training dataset and test dataset. The network construction module is used to build a Kolmogorov-Arnold network with 2 inputs, 4 outputs, and 2 hidden layers, as well as its companion network with 2 inputs and 8 outputs. The inputs of the Kolmogorov-Arnold network are the gate voltage u1 and the drain voltage u2, and the outputs are the gate current i1, the drain current i2, the gate charge q1, and the drain charge q2. The output of the companion network is the partial derivative of the Kolmogorov-Arnold network output with respect to the input. The training architecture module is used to build the training architecture for the small-signal model. The inputs to this architecture are the gate voltage u1, the drain voltage u2, and the frequency. freq The output consists of DC output gate current I1 and drain current I2, as well as small-signal S-parameters and input reflection coefficient S. 11 Reverse transmission coefficient S 12 Forward transmission coefficient S 21 and output reflection coefficient S 22 ; The training module is used to train the training architecture based on the training dataset and calculate the training error E. train And adjust network parameters or data according to the error; The testing module is used to test the training architecture using a test dataset and calculate the test error E. test And adjust the training data based on the error; The model export module is used to update the weights and export the trained microwave transistor small-signal model.
[0014] Preferably, the system also includes an accuracy comparison module, which is used to compare the accuracy of the model based on the Kolmogorov-Arnold network with the model based on the multilayer perceptron neural network and output the comparison results.
[0015] The advantages of this invention compared to the prior art are: This invention proposes a small-signal modeling method and system for microwave transistors based on Kolmogorov-Arnold networks, aiming to further improve modeling accuracy. A novel Kolmogorov-Arnold network is introduced, a model training architecture based on small-signal S-parameter data under DC and multi-bias conditions is established, a model training algorithm is developed, and the architecture is trained. This method outperforms existing methods in terms of modeling accuracy and generalization ability. Attached Figure Description
[0016] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.
[0017] Figure 1 This is a schematic diagram of the Kolmogorov-Arnold network of the present invention.
[0018] Figure 2 This invention relates to the Kolmogorov-Arnold network.
[0019] Figure 3 This is a diagram of the model training architecture based on DC and small-signal S-parameter data of the present invention.
[0020] Figure 4 This is a structural diagram of the microwave transistor derived from the present invention.
[0021] Figure 5 This is a comparison chart of the accuracy of the present invention and the small-signal model of microwave transistors based on MLP.
[0022] Figure 6 A comparison chart of the DC output of the model established for this invention and the modeling data.
[0023] Figure 7 The small-signal S-parameter output of the model established for this invention is compared with the modeling data. (a) S11, (b) S12, (c) S21, (d) S22. Detailed Implementation
[0024] The embodiments of this application are described in detail below. Examples of the embodiments are shown in the accompanying drawings, wherein the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the accompanying drawings are exemplary and are only used to explain this application, and should not be construed as limiting this application.
[0025] In the description of this application, it should be noted that, unless otherwise expressly specified and limited, the terms "installation," "connection," and "joining" should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral connection; they can refer to a mechanical connection or an electrical connection; they can refer to a direct connection or an indirect connection through an intermediate medium; they can refer to the internal communication of two components or the interaction between two components. Those skilled in the art can understand the specific meaning of the above terms in this application according to the specific circumstances.
[0026] The present invention will now be described in further detail with reference to the accompanying drawings.
[0027] This invention provides a small-signal modeling method for microwave transistors, comprising the following steps: S1: Obtain the core characteristic data of the microwave transistor to be modeled through actual measurement, including small signal S-parameter data under DC and multiple bias points, divide the data into training dataset and test dataset, and set the training error threshold. S2: Construct a Kolmogorov-Arnold network with 2 inputs, 4 outputs, and 2 hidden layers, and its adjoint network with 2 inputs and 8 outputs. The inputs of the Kolmogorov-Arnold network are the gate voltage u1 and the drain voltage u2, and the outputs are the gate current i1, the drain current i2, the gate charge q1, and the drain charge q2. The output of the adjoint network is the partial derivative of the output of the Kolmogorov-Arnold network with respect to the input. S3: Build the training architecture for the small-signal model. The inputs to this architecture are the gate voltage u1, drain voltage u2, and frequency freq. The outputs are the DC output gate current I1 and drain current I2, as well as the small-signal S-parameters and the input reflection coefficient S. 11 1. Reverse transmission coefficient 2. Forward transmission coefficient S 21 and output reflection coefficient S 22 ; S4: Train the training architecture based on the small-signal S-parameter training dataset under DC and multiple bias points, and calculate the training error E after training. train ; S5: If the training error E train Greater than the training error threshold If the condition is met, the number of hidden layer neurons is increased, and the process returns to S2 to rebuild the network; otherwise, proceed to S6. S6: Test the training architecture using the test dataset and calculate the test error E. test ; S7: If the test error E test Greater than the training error threshold If the training data is correct, then add training data and return to S4 for retraining; otherwise, proceed to S8. S8: Update weights to obtain And derive the trained microwave transistor small-signal model.
[0028] Preferably, the expression for the Kolmogorov-Arnold network described in S2 is: (1) In the formula: o1, o2, o3 and o4 are the outputs i1, i2, q1 and q2 of the Kolmogorov-Arnold network, respectively; Let represent the k-th order B-spline curves between the j-th hidden neuron in the first hidden layer and the i-th input, the k-th hidden neuron in the second hidden layer and the j-th hidden neuron in the first hidden layer, and the k-th hidden neuron in the second hidden layer and the l-th output, respectively. The k-th order B-spline curves are formed by... The calculation yields c. m Let c be the learnable coefficients of the B-spline curve. m The vector formed is denoted as The symbol m represents the index of the learnable coefficient; the symbol M represents the total number of learnable coefficients, where M = G + k, and G is the number of grids in the B-spline function.
[0029] Preferably, the expression for the adjoint network in S2 is: (2) in, yes The derivative of .
[0030] Preferably, the DC output gate current I1 and drain current I2 mentioned in S3 are obtained by formula (1) under DC conditions: u1=U1, u2=U2.
[0031] Preferably, the small-signal S-parameter output in S3 is transmitted through the intermediate variable admittance parameter Y. 11 Y 12 Y 21 Y 22 After YS transformation, the admittance parameter is calculated using the following formula: (3) in, Angular frequency, equal to Multiply by the third input freq of the Kolmogorov-Arnold network.
[0032] Preferably, the training error E in S4 train Calculated using the following formula: (4) in, , and These are the DC output and small-signal S-parameter output of the training architecture, respectively. D and S D For measured data, G and M are diagonal matrices, with scaling factors equal to the reciprocals of the corresponding data ranges. k and j are the bias point and frequency point indices, respectively. train and J train These represent the total number of biases and frequencies in the training set, respectively.
[0033] Preferably, the test error E in S6 test Calculated using the following formula: (5) Among them, K test and J test These represent the total number of biases and frequencies in the test set, respectively.
[0034] Preferably, in the small-signal model of the microwave transistor derived in S8, the voltage between the gate and the source is u1, and the current flowing into the gate is denoted as u1. The voltage between the drain and source is u2, and the current flowing into the drain is denoted as u2. A current source and a charge element are connected in parallel between the gate and the source, respectively, and f in formula (1) is built in. KAN1 and f KAN3 The expression for the current source and charge element are connected in parallel between the drain and source, with each having a built-in f. KAN2 and f KAN4 The expression for the current is: (6) (7) A microwave transistor small-signal modeling system, comprising: The data acquisition module is used to acquire the core characteristic data of the microwave transistor to be modeled through actual measurement, including small-signal S-parameter data under DC and multiple bias points, and divide the data into training dataset and test dataset. The network construction module is used to build a Kolmogorov-Arnold network with 2 inputs, 4 outputs, and 2 hidden layers, as well as its companion network with 2 inputs and 8 outputs. The inputs of the Kolmogorov-Arnold network are the gate voltage u1 and the drain voltage u2, and the outputs are the gate current i1, the drain current i2, the gate charge q1, and the drain charge q2. The output of the companion network is the partial derivative of the Kolmogorov-Arnold network output with respect to the input. The training architecture module is used to build the training architecture for the small-signal model. The inputs to this architecture are the gate voltage u1, the drain voltage u2, and the frequency. freq The output consists of DC output gate current I1 and drain current I2, as well as small-signal S-parameters and input reflection coefficient S. 11 Reverse transmission coefficient S 12 Forward transmission coefficient S 21 and output reflection coefficient S 22 ; The training module is used to train the training architecture based on the training dataset and calculate the training error E. train And adjust network parameters or data according to the error; The testing module is used to test the training architecture using a test dataset and calculate the test error E. test And adjust the training data based on the error; The model export module is used to update the weights and export the trained microwave transistor small-signal model.
[0035] Preferably, the system also includes an accuracy comparison module, which is used to compare the accuracy of the model based on the Kolmogorov-Arnold network with the model based on the multilayer perceptron neural network and output the comparison results.
[0036] To more clearly illustrate the specific embodiments of the present invention, an example is provided below; In a specific embodiment of the microwave transistor small-signal modeling method based on Kolmogorov-Arnold networks described in this invention, an actual GaAs HEMT device is used as the modeling object. Modeling DC training data. U 1. Take a sample at 0.2V intervals from -0.2V to 0.8V. U 2. Take a sample from 0V to 6.2V, at 0.1V intervals. Model DC test data. U 1. Take a sample at 0.2V intervals from -0.1V to 0.7V. U 2. Data points were taken at 0.1V intervals from 0.05V to 6.15V. Small-signal S-parameter training data. U 1. Take a sample at 0.2V intervals from -0.2V to 0.8V. U 2. From 0V to 6.2V, take a point at every 0.1V interval, frequency freq Data points were taken at 1 GHz intervals, from 0.1 GHz to 40.1 GHz. Small-signal S-parameter test data. U 1. Take a sample at 0.2V intervals from -0.1V to 0.7V. U 2. From 0.05V to 6.15V, take a point at every 0.1V interval, frequency freq From 0.1GHz to 40.1GHz, a point is taken every 1GHz.
[0037] First, according to Figure 1 A KAN with 2 inputs, 4 outputs, and 2 hidden layers, each with 32 neurons, is constructed. Secondly, based on the KAN, according to... Figure 2 Establish an adjoint KAN. The KAN and the adjoint KAN share weights. w Again, according to Figure 3 A model training architecture based on DC and small-signal S-parameter data was established, and the architecture was trained and tested using DC and small-signal S-parameter training data. After successful training, the updated weights were obtained. w # Finally, according to Figure 4 A small-signal model of a microwave transistor based on KAN was established. For comparison, a small-signal model of a microwave transistor based on MLP was established using the same method. The MLP also includes 2 inputs, 4 outputs, and 2 hidden layers, each with 32 neurons. The comparison results are as follows: Figure 5 As shown in the figure, it can be seen that the accuracy of the KAN-based microwave transistor small-signal model is much higher than that of the MLP-based small-signal model, both in terms of DC characteristics and small-signal S-parameter characteristics.
[0038] To further illustrate the results, we plotted the small-signal S-parameter outputs of the KAN-based microwave transistor small-signal model at DC and six bias points, comparing them with the corresponding modeling data, as shown below. Figure 6 and Figure 7 As shown in the figure, the model fits the microwave transistor measurement data well, verifying the effectiveness of the modeling method.
[0039] The present invention and its embodiments have been described above. This description is not restrictive, and the accompanying drawings are only one embodiment of the present invention; the actual structure is not limited thereto. In conclusion, if those skilled in the art are inspired by this description and design similar structures and embodiments without departing from the spirit of the invention, such designs should fall within the protection scope of the present invention.
Claims
1. A method for small-signal modeling of microwave transistors, characterized in that, Includes the following steps: S1: Obtain the core characteristic data of the microwave transistor to be modeled through actual measurement, including small signal S-parameter data under DC and multiple bias points, divide the data into training dataset and test dataset, and set the training error threshold. S2: Construct a Kolmogorov-Arnold network with 2 inputs, 4 outputs, and 2 hidden layers, and its adjoint network with 2 inputs and 8 outputs. The inputs of the Kolmogorov-Arnold network are the gate voltage u1 and the drain voltage u2, and the outputs are the gate current i1, the drain current i2, the gate charge q1, and the drain charge q2. The output of the adjoint network is the partial derivative of the output of the Kolmogorov-Arnold network with respect to the input. S3: Build the training architecture for the small-signal model. The inputs to this architecture are the gate voltage u1, drain voltage u2, and frequency freq. The outputs are the DC output gate current I1 and drain current I2, as well as the small-signal S-parameters and the input reflection coefficient S. 11 1. Reverse transmission coefficient 2. Forward transmission coefficient S 21 and output reflection coefficient S 22 ; S4: Train the training architecture based on the small-signal S-parameter training dataset under DC and multiple bias points, and calculate the training error E after training. train ; S5: If the training error E train Greater than the training error threshold If the condition is met, the number of hidden layer neurons is increased, and the process returns to S2 to rebuild the network; otherwise, proceed to S6. S6: Test the training architecture using the test dataset and calculate the test error E. test ; S7: If the test error E test Greater than the training error threshold If so, add training data and return to S4 to retrain; Otherwise, proceed to S8; S8: Update weights to obtain And derive the trained microwave transistor small-signal model.
2. The microwave transistor small-signal modeling method according to claim 1, characterized in that: The expression for the Kolmogorov-Arnold network described in S2 is: （1） In the formula: o1, o2, o3 and o4 are the outputs i1, i2, q1 and q2 of the Kolmogorov-Arnold network, respectively; Let represent the k-th order B-spline curves between the j-th hidden neuron in the first hidden layer and the i-th input, the k-th hidden neuron in the second hidden layer and the j-th hidden neuron in the first hidden layer, and the k-th hidden neuron in the second hidden layer and the l-th output, respectively. The k-th order B-spline curves are formed by... The calculation yields c. m Let c be the learnable coefficients of the B-spline curve. m The vector formed is denoted as The symbol m represents the index of the learnable coefficient; the symbol M represents the total number of learnable coefficients, where M = G + k, and G is the number of grids in the B-spline function.
3. The microwave transistor small-signal modeling method according to claim 1, characterized in that: The expression for the adjoint network described in S2 is: （2） in, yes The derivative of .
4. The microwave transistor small-signal modeling method according to claim 1, characterized in that: The DC output gate current I1 and drain current I2 mentioned in S3 are obtained by formula (1) under DC conditions: u1=U1, u2=U2.
5. The microwave transistor small-signal modeling method according to claim 1, characterized in that: The small-signal S-parameter output described in S3 is obtained through the intermediate variable admittance parameter Y. 11 Y 12 Y 21 Y 22 After YS transformation, the admittance parameter is calculated using the following formula: （3） in, Angular frequency, equal to Multiply by the third input freq of the Kolmogorov-Arnold network.
6. The microwave transistor small-signal modeling method according to claim 1, characterized in that: The training error E mentioned in S4 train Calculated using the following formula: （4） in, , and These are the DC output and small-signal S-parameter output of the training architecture, respectively. D and S D For measured data, G and M are diagonal matrices, with scaling factors equal to the reciprocals of the corresponding data ranges. k and j are the bias point and frequency point indices, respectively. train and J train These represent the total number of biases and frequencies in the training set, respectively.
7. The microwave transistor small-signal modeling method according to claim 1, characterized in that: The test error E mentioned in S6 test Calculated using the following formula: （5） Among them, K test and J test These represent the total number of biases and frequencies in the test set, respectively.
8. The microwave transistor small-signal modeling method according to claim 1, characterized in that: In the small-signal model of the microwave transistor derived in S8, the voltage between the gate and the source is u1, and the current flowing into the gate is denoted as u1. The voltage between the drain and source is u2, and the current flowing into the drain is denoted as u2. A current source and a charge element are connected in parallel between the gate and the source, respectively, and f in formula (1) is built in. KAN1 and f KAN3 The expression for the current source and charge element are connected in parallel between the drain and source, with each having a built-in f. KAN2 and f KAN4 The expression for the current is: （6） （7）。 9. A microwave transistor small-signal modeling system, characterized in that, include: The data acquisition module is used to acquire the core characteristic data of the microwave transistor to be modeled through actual measurement, including small-signal S-parameter data under DC and multiple bias points, and divide the data into training dataset and test dataset. The network construction module is used to build a Kolmogorov-Arnold network with 2 inputs, 4 outputs, and 2 hidden layers, as well as its companion network with 2 inputs and 8 outputs. The inputs of the Kolmogorov-Arnold network are the gate voltage u1 and the drain voltage u2, and the outputs are the gate current i1, the drain current i2, the gate charge q1, and the drain charge q2. The output of the companion network is the partial derivative of the Kolmogorov-Arnold network output with respect to the input. The training architecture module is used to build the training architecture for the small-signal model. The inputs to this architecture are the gate voltage u1, the drain voltage u2, and the frequency. freq The output consists of DC output gate current I1 and drain current I2, as well as small-signal S-parameters and input reflection coefficient S. 11 Reverse transmission coefficient S 12 Forward transmission coefficient S 21 and output reflection coefficient S 22 ; The training module is used to train the training architecture based on the training dataset and calculate the training error E. train And adjust network parameters or data according to the error; The testing module is used to test the training architecture using a test dataset and calculate the test error E. test And adjust the training data based on the error; The model export module is used to update the weights and export the trained microwave transistor small-signal model.
10. The system according to claim 9, characterized in that: The system also includes an accuracy comparison module, which compares the accuracy of the model based on the Kolmogorov-Arnold network with that based on the multilayer perceptron neural network and outputs the comparison results.