A method for automatically solving parameters of a coupled microstrip line structure based on an MLP neural network

Abstract

The application discloses a kind of based on MLP neural network's coupled microstrip line structure parameter automatic solution method, comprising the following steps: by simulation to obtain S parameter set file, read S parameter set file, calculate to form "image impedance-structure parameter" sample pair dataset;Using dataset trains MLP neural network model;Set target image impedance value, input target image impedance value into trained MLP neural network, and output coupled line structure parameter.The application utilizes API interface to automatically realize model creation, parameter adjustment and data export function, solves the problem of time-consuming and labor-consuming of artificial repetitive operation;At the same time, with the powerful numerical calculation function of MATLAB and the accurate data fitting function of MLP neural network, the initial value of asymmetric and symmetric coupled microstrip line structure parameters approaching the optimal value can be quickly provided, which significantly reduces the iteration number and difficulty of coupled microstrip line structure parameter optimization.

Description

Technical Field

[0001] This invention relates to the field of microwave engineering, and in particular to an automatic method for solving the structural parameters of coupled microstrip lines. Background Technology

[0002] Coupled lines, as fundamental units in microwave circuit design, are widely used in filters, power dividers, antennas, and power amplifiers due to their compact structure and broadband characteristics. Based on whether the two linewidths are the same, coupled lines are further divided into symmetrical and asymmetrical coupled lines. Compared to symmetrical coupled lines, asymmetrical coupled lines add an adjustable variable, i.e., an additional degree of freedom, offering greater design flexibility but also presenting greater design challenges.

[0003] For the design of symmetric coupled-line devices, empirical formulas and commercial tools such as ADS LineCalc are typically used to determine the initial structural parameters of the coupled line. However, because empirical formulas use approximate simplified solutions, and the model dimensions and material parameters provided by commercial tools deviate significantly from the optimal values ​​of the full-wave simulation platform model, subsequent designs still face a considerable burden of iterative optimization. For the design of asymmetric coupled-line devices, the theoretical analysis is even more complex, and the lack of support from empirical formulas and commercial tools has hindered the promotion and application of asymmetric coupled lines to some extent.

[0004] Therefore, to minimize the design complexity and optimization iterations of coupled-line devices, an accurate method for initializing coupled-line structural parameters is needed. This involves directly starting from the telegraph equations describing the behavior of the coupled line, without approximations or simplifications, and establishing an accurate mapping between the analytical solution of the equations and the coupled-line structural parameters. This significantly reduces the number of iterations and the complexity of optimizing coupled microstrip line structural parameters. Summary of the Invention

[0005] To achieve the above objectives, this invention provides an automatic solution method for coupled microstrip line structure parameters based on an MLP neural network, comprising the following steps:

[0006] S1: The S-parameter set file is obtained through simulation. The S-parameter is a two-dimensional matrix whose elements represent the signal transmission between different ports. The S-parameter set file is a collection of multiple S-parameters.

[0007] S2: Read the S-parameter set file and calculate to form a "mirror impedance-structural parameter" sample pair dataset;

[0008] S3: Train an MLP neural network model using a dataset;

[0009] S4: Set the target mirror impedance value, input the target mirror impedance value into the trained MLP neural network, and output the coupled line structure parameters.

[0010] Furthermore, the method for obtaining the S-parameter set file in step S1 is as follows: write a MATLAB vbs script, call the HFSS software to create an initial model of the coupled microstrip line, use the vbs script to automatically change the HFSS microstrip line model, traverse all parameter combinations and simulate to obtain the S-parameter set file.

[0011] Furthermore, the parameter combination includes the linewidths W1 and W2 of the coupled microstrip line, the line spacing P, and the length of the coupled line. The width D and thickness H of the dielectric substrate, and the thickness T of the metal conductor.

[0012] Further, step S2 includes: using Matlab to extract the distributed characteristic parameters of the coupled line, including resistance R, inductance L, reactance G and capacitance C per unit length, calculating the mode impedance to mode voltage ratio of the coupled line, obtaining the image impedance value, and forming a "image impedance-structural parameter" sample pair dataset.

[0013] Furthermore, the step of extracting the distributed feature parameters of the coupling line includes:

[0014] S51: Read the S-parameter set file data and solve for the Z-parameter matrix of the coupled microstrip line, where Z0 is a 4×4 diagonal matrix composed of port reference impedances, I is a 4×4 identity matrix, S is a 4×4 S-parameter matrix, and Z... ij (i, j = 1, 2) are the 2×2 submatrices that constitute matrix Z;

[0015]

[0016] S52: The ABCD parameter matrix of the coupled microstrip line is obtained by solving the Z parameters, where A, B, C, and D are 2×2 submatrices.

[0017]

[0018] S53: Perform similarity diagonalization decomposition on matrix A to obtain the eigenvector matrix E and the eigenvalue diagonal matrix Λ, and then solve for the propagation constant matrix Γ and the characteristic impedance matrix Z of the coupled microstrip line. c ,in, Given the coupling line length, the operators cosh-1 and sinh are applied to matrices Λ and Γ, respectively. The elements;

[0019]

[0020]

[0021]

[0022] S54: Obtain the distributed parameters of the coupled microstrip line, where the angular frequency ω = 2πf, and R, L, G, C are 2×2 matrices.

[0023]

[0024] Furthermore, the method for calculating the mode impedance to mode voltage ratio of the coupled line is as follows: the mode impedance Z of the asymmetric coupled microstrip line c1 Z c2 Z π1 Z π2 Inter-line mode voltage ratio R c R π The image impedance Z is obtained from the distributed characteristic parameters R, L, G, and C. I1 Z I2 Z I3 With Z c1 Z c2 Z π1 Z π2 R c R π The relationship is:

[0025]

[0026]

[0027] .

[0028] Furthermore, the method for training the MLP neural network model in step S3 is as follows:

[0029] S71: Randomly split the sample pair dataset into training and validation sets;

[0030] S72: Construct an MLP neural network model and set training parameters;

[0031] S73: Train the MLP neural network model using the training set, establish the mapping relationship between mirror impedance and structural parameters, and use the sample data of the validation set to verify the prediction results of the neural network model.

[0032] Furthermore, the training parameters include the number of hidden layers, the number of hidden layer neurons, the activation function, the maximum number of training epochs, the training target error, and the learning rate.

[0033] Furthermore, the method for constructing the MLP neural network model is as follows:

[0034] S71: The hyperbolic tangent sigmoid function is chosen as the activation function of the hidden layer, compressing the element values ​​to the interval [-1, 1]. N is defined as the input vector, and the function expression is:

[0035]

[0036] S72: Weight initialization, using random numbers from the standard normal distribution to initialize the weight matrix;

[0037] S73: Bias initialization, initializes the bias using random numbers from the standard normal distribution, and sets the bias matrix;

[0038] S74: Maximum training epochs are 1000, training objective error is 10⁻⁵, and learning rate (lr) is 0.1.

[0039] Furthermore, in step S3, the input dimension of the dataset is 3, the target dimension is 3, and the Levenberg-Marquardt algorithm is used to train the MLP neural network model to optimize and update the weights and bias values.

[0040] The beneficial effects of this invention are as follows:

[0041] 1) This invention automatically realizes the functions of model creation, parameter adjustment, and data export, solving the problem of time-consuming and laborious manual repetitive operations.

[0042] 2) With the help of MATLAB’s powerful numerical computing capabilities and the accurate data fitting capabilities of MLP neural networks, the initial parameters of asymmetric and symmetric coupled microstrip line structures can be quickly provided to approximate the optimal values, which significantly reduces the number of iterations and the difficulty of optimizing the parameters of coupled microstrip line structures.

[0043] 3) Using the Levenberg-Marquardt algorithm to train MLP neural network models improves training efficiency and avoids getting trapped in local minima, thereby obtaining better model performance and generalization ability. Attached Figure Description

[0044] Figure 1 This is a flowchart of an automatic solution method for coupled microstrip line structure parameters based on MATLAB-HFSS co-simulation and MLP neural network according to an embodiment of the present invention;

[0045] Figure 2 This is a diagram showing the cross-section and model structure parameter definition of the coupled microstrip line in an embodiment of the present invention;

[0046] Figure 3 This is a flowchart illustrating the extraction of distributed characteristic parameters of coupled microstrip lines according to an embodiment of the present invention.

[0047] Figure 4 This is a block diagram illustrating the construction, training, and use of the MLP neural network model according to an embodiment of the present invention. Detailed Implementation

[0048] To make the technical means, creative features, objectives and effects of this invention easier to understand, the invention will be further described below in conjunction with specific embodiments.

[0049] This embodiment provides an automatic solution method for coupled microstrip line structure parameters based on MLP neural networks, such as... Figure 1 As shown, firstly, a MATLAB VBScript is written to call the HFSS software to create the initial model of the coupled microstrip line.

[0050] In this project, HFSS parametric processing was performed on the basic parameters of the coupled microstrip line model using VBScript, avoiding the need for remodeling after parameter values ​​were changed. For example... Figure 2 As shown, the basic parameters include the linewidths W1 and W2 of the coupled microstrip lines, the line spacing P, and the length of the coupled lines. The width D and thickness H of the dielectric substrate, and the thickness T of the metal conductor.

[0051] It should be noted that the method in this embodiment can be applied to both asymmetric coupled microstrip line structures, i.e., the two line widths are different, and symmetric coupled line structures, i.e., the two line widths are the same. For symmetric coupled line structures, W1 and W2 are combined into one variable for parameter setting.

[0052] Furthermore, the range of variation of the coupled line structure parameters is set, and the HFSS microstrip line model is automatically modified using a VBScript. All parameter combinations are traversed and simulation is performed to obtain the S-parameter set file.

[0053] Those skilled in the art will know that, as Figure 3 As shown, the S-parameters of a coupled microstrip line structure are used to describe signal transmission and reflection in the circuit. The S-parameter matrix is ​​a two-dimensional matrix, where each element represents the signal transmission between different ports. S-parameters can be converted to Z-parameters, which are measures of impedance transmission between different ports in the circuit. Converting S-parameters to Z-parameters reflects the impedance changes during signal transmission.

[0054] Furthermore, MATLAB is used to read the S-parameter set file and extract the distributed characteristic parameters of the coupling line. The specific steps include:

[0055] S41: Read the S-parameter set file data and solve for the Z-parameter matrix of the coupled microstrip line, where Z0 is a 4×4 diagonal matrix composed of port reference impedances, I is a 4×4 identity matrix, S is a 4×4 S-parameter matrix, and Z... ij (i, j = 1, 2) are the 2×2 submatrices that constitute matrix Z;

[0056]

[0057] S42: The ABCD parameter matrix of the coupled microstrip line is obtained by solving the Z parameters, where A, B, C, and D are 2×2 submatrices.

[0058]

[0059] S43: Perform similarity diagonalization decomposition on matrix A to obtain the eigenvector matrix E and the eigenvalue diagonal matrix Λ, and then solve for the propagation constant matrix Γ and the characteristic impedance matrix Z of the coupled microstrip line. c ,in, For the length of the coupling line, the operator cosh -1 sinh and sinh act on matrices Λ and Γ, respectively. The elements;

[0060]

[0061]

[0062]

[0063] S44: Obtain the distributed parameters of the coupled microstrip line, where the angular frequency ω = 2πf, and R, L, G, C are 2×2 matrices.

[0064]

[0065] Furthermore, the mode impedance to mode voltage ratio of the coupled line is calculated to obtain the image impedance value, specifically: the mode impedance Z of the asymmetric coupled microstrip line. c1 Z c2 Z π1 Z π2 Inter-line mode voltage ratio R c R π The image impedance Z is obtained from the distributed characteristic parameters R, L, G, and C. I1 Z I2 Z I3 With Z c1 Z c2 Z π1 Z π2 R c R π The relationship is:

[0066]

[0067]

[0068] .

[0069] This embodiment traversed 18 parameter combinations and obtained 18 pairs of "image impedance-structural parameter" sample pairs, the specific parameters of which are shown in Table 1.

[0070] Table 1

[0071]

[0072] Furthermore, a "mirror impedance-structural parameter" sample pair dataset is formed, and the dataset is randomly divided into a training set and a validation set. The data in the training set is used to train the MLP neural network model, and then the data in the validation set is used to further validate the prediction results of the MLP neural network model.

[0073] Furthermore, the hidden layer number is set to 1, the number of hidden layer neurons is 10, the activation function is the hyperbolic tangent sigmoid function, the maximum training epochs are 1000, and the training objective error is 10. -5 Learning rate l r The initial value is 0.1. The weight matrix and bias matrix are initialized with random numbers from the standard normal distribution to construct the MLP neural network.

[0074] like Figure 4 As shown, an MLP neural network model is trained using the training set and the Levenberg-Marquardt algorithm, and the weights and biases are optimized and updated. The mapping relationship between mirror impedance and structural parameters is established, and the prediction results of the neural network model are verified using sample data from the validation set.

[0075] As those skilled in the art will recognize, the Levenberg-Marquardt algorithm is an iterative optimization algorithm that combines the Levenberg-Marquardt method with gradient descent. This algorithm minimizes the error between the network output and the target output by continuously adjusting the weights and biases of the neural network. During each iteration, the algorithm calculates the gradient of the network and uses gradient descent to update the network weights and biases. Simultaneously, the algorithm adaptively adjusts the step size based on the properties of the error function to accelerate convergence and avoid getting trapped in local minima.

[0076] Furthermore, in this embodiment, five sets of mirror impedance values ​​are set, and the target mirror impedance values ​​are input into the trained MLP neural network to automatically output the coupling line structure parameters as shown in the table below.

[0077]

[0078] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any person skilled in the art can easily conceive of various equivalent modifications or substitutions within the technical scope disclosed in the present invention, and these modifications or substitutions should all be covered within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. A method for automatically solving the parameters of a coupled microstrip line structure based on an MLP neural network, characterized in that, Includes the following steps: S1: The S-parameter set file is obtained through simulation. The S-parameter is a two-dimensional matrix whose elements represent the signal transmission between different ports. The S-parameter set file is a collection of multiple S-parameters. S2: Read the S-parameter set file and calculate to form a "mirror impedance-structural parameter" sample pair dataset; S3: Train an MLP neural network model using a dataset; S4: Set the target mirror impedance value, input the target mirror impedance value into the trained MLP neural network, and output the coupled line structure parameters; Step S2 includes: using Matlab to extract distributed feature parameters of the coupled line, including resistance R, inductance L, reactance G and capacitance C per unit length; calculating the mode impedance to mode voltage ratio of the coupled line; obtaining the image impedance value; and forming a "image impedance-structural parameter" sample pair dataset. The step of extracting the distributed feature parameters of the coupling line includes: S51: Read the S-parameter set file data and solve for the Z-parameter matrix of the coupled microstrip line, where Z0 is a 4×4 diagonal matrix composed of port reference impedances, I is a 4×4 identity matrix, S is a 4×4 S-parameter matrix, and Z... ij (i, j = 1, 2) are the 2×2 submatrices that constitute matrix Z; S52: The ABCD parameter matrix of the coupled microstrip line is obtained by solving the Z parameters, where A, B, C, and D are 2×2 submatrices. S53: Perform similarity diagonalization decomposition on matrix A to obtain the eigenvector matrix E and the eigenvalue diagonal matrix Λ, and then solve for the propagation constant matrix Γ and the characteristic impedance matrix Z of the coupled microstrip line. c ,in, For the length of the coupling line, the operator cosh -1 sinh and sinh act on matrices Λ and Γ, respectively. The elements; S54: Obtain the distributed parameters of the coupled microstrip line, where the angular frequency ω = 2πf, and R, L, G, C are 2×2 matrices; The method for calculating the mode impedance to mode voltage ratio of the coupled line is as follows: Mode impedance Z of the asymmetric coupled microstrip line... c1 Z c2 Z π1 Z π2 Inter-line mode voltage ratio R c R π The image impedance Z is obtained from the distributed characteristic parameters R, L, G, and C. I1 Z I2 Z I3 With Z c1 Z c2 Z π1 Z π2 R c R π The relationship is: 。 2. The automatic solution method for coupled microstrip line structure parameters based on MLP neural network according to claim 1, characterized in that, The method for obtaining the S-parameter set file in step S1 is as follows: write a MATLAB vbs script, call the HFSS software to create an initial model of the coupled microstrip line, use the vbs script to automatically change the HFSS microstrip line model, traverse all parameter combinations and simulate to obtain the S-parameter set file.

3. The automatic solution method for coupled microstrip line structure parameters based on MLP neural network according to claim 2, characterized in that, The parameter combination includes the linewidths W1 and W2 of the coupled microstrip line, the line spacing P, and the length of the coupled line. The width D and thickness H of the dielectric substrate, and the thickness T of the metal conductor.

4. The automatic solution method for coupled microstrip line structure parameters based on MLP neural network according to claim 1, characterized in that, The method for training the MLP neural network model in step S3 is as follows: S71: Randomly split the sample pair dataset into training and validation sets; S72: Construct an MLP neural network model and set training parameters; S73: Train the MLP neural network model using the training set, establish the mapping relationship between mirror impedance and structural parameters, and use the sample data of the validation set to verify the prediction results of the neural network model.

5. The automatic solution method for coupled microstrip line structure parameters based on MLP neural network according to claim 4, characterized in that, The training parameters include the number of hidden layers, the number of neurons in the hidden layers, the activation function, the maximum number of training epochs, the training target error, and the learning rate.

6. The automatic solution method for coupled microstrip line structure parameters based on MLP neural network according to claim 5, characterized in that, The method for constructing the MLP neural network model is as follows: S71: The hyperbolic tangent sigmoid function is chosen as the activation function of the hidden layer, compressing the element values ​​to the interval [-1, 1]. N is defined as the input vector, and the function expression is: S72: Weight initialization, using random numbers from the standard normal distribution to initialize the weight matrix; S73: Bias initialization, initializes the bias using random numbers from the standard normal distribution, and sets the bias matrix; S74: Maximum training epochs are 1000, and the training objective error is 10. -5 Learning rate l r It is 0.

1.

7. The automatic solution method for coupled microstrip line structure parameters based on MLP neural network according to claim 1, characterized in that, In step S3, the input dimension of the dataset is 3, the target dimension is 3, and the Levenberg-Marquardt algorithm is used to train the MLP neural network model to optimize and update the weights and bias values.

Images