CN-121981040-A - Method, system, storage medium and equipment for designing superconductive band-pass filter

CN121981040ACN 121981040 ACN121981040 ACN 121981040ACN-121981040-A

Abstract

The invention discloses a design method, a system, a storage medium and equipment of a high-temperature superconductive band-pass filter. The method comprises the steps of firstly determining design indexes including center frequency, relative bandwidth and the like, and an extensible structure frame formed by cascading a plurality of sections of coupling lines and a plurality of sections of branch lines, respectively establishing differential mode and common mode equivalent circuit models of a filter according to a transmission line theory, then establishing a multi-layer perceptron type circuit level physical information neural network model with residual connection, inputting the design indexes into the model, embedding an equivalent circuit matrix and a target differential mode transmission function into a loss function to complete training, and outputting coupling line even mode, odd mode characteristic impedance and branch line characteristic impedance through the trained model, thereby generating a filter physical layout. The invention integrates the physical information neural network, can rapidly generate the superconducting filter layout, and effectively improves the design efficiency and the prediction precision.

Inventors

WEN PIN
WANG DAN
ZENG YANHUA
CHEN JINGYAO
JIANG YUQIAN
ZHOU SONG
JIA JIE
SHI LIU

Assignees

南昌大学

Dates

Publication Date: 20260505
Application Date: 20260408

Claims (10)

1. A method of designing a high temperature superconducting bandpass filter, the method comprising: Determining design indexes of a filter and an extensible structural framework, wherein the design indexes comprise center frequency, relative bandwidth, filter order, differential mode return loss and common mode rejection, and the structural framework is formed by cascading a plurality of sections of coupling lines and a plurality of sections of branch lines; based on a transmission line theory, respectively constructing a differential mode equivalent circuit model and a common mode equivalent circuit model of the filter structure; constructing a circuit-level physical information neural network model adopting a multi-layer perceptron architecture, comprising setting a plurality of hidden layers, configuring a corresponding number of neurons, and simultaneously adding residual connection; Inputting the design index into the circuit-level physical information neural network model, embedding a matrix derived by an equivalent circuit and a target differential mode transfer function into a loss function of the model, and completing model training; Outputting the even mode characteristic impedance, the odd mode characteristic impedance and the characteristic impedance of the branch line of the coupling line by using the trained circuit-level physical information neural network model; and generating a physical layout of the filter based on the output characteristic impedance.
2. The method for designing a high-temperature superconductive bandpass filter according to claim 1, wherein the target differential mode transfer function adopts an equiripple chebyshev form, and the specific implementation includes: determining the pass band ripple level and the cut-off frequency of the chebyshev response, and calculating the corresponding electrical length; constructing a first class chebyshev polynomial function of corresponding degree; And generating a target differential mode transfer function based on the Chebyshev polynomial function, and embedding a loss function of the circuit-level physical information neural network model.
3. The method for designing a high-temperature superconductive bandpass filter according to claim 2, wherein the loss function of the circuit-level physical information neural network model includes a frequency weighted loss term, and the specific implementation of the frequency weighted loss term includes: identifying a passband frequency interval and an out-of-band frequency interval of the filter; Setting high weight coefficients for frequency points in the passband frequency interval and setting low weight coefficients for frequency points in the out-of-band frequency interval; Multiplying the corresponding weight coefficient with the scattering parameter prediction error to obtain a frequency weighting loss term, and embedding the frequency weighting loss term into the total loss function of the circuit-level physical information neural network model.
4. The method for designing a high-temperature superconductive bandpass filter according to claim 2, wherein the circuit-level physical information neural network model satisfies a passive boundary condition, and the implementation includes: Invoking condition information of a passive boundary condition, and determining amplitude constraint which needs to be met by a scattering parameter according to the condition information; converting the passive constraint condition into a punishment item, and embedding the punishment item into a loss function of the circuit-level physical information neural network model; And in the training process of the circuit-level physical information neural network model, penalty is applied to the prediction result which violates the passivity constraint so as to realize boundary condition constraint.
5. The method of designing a high temperature superconducting bandpass filter according to any one of claims 1-4, further comprising: constructing a layout-level physical information neural network model based on the physical layout, and adopting a one-dimensional convolution self-encoder model, wherein the encoder is a convolution neural network, and the decoder is a multi-layer full-connection network; Constructing a mixed loss function of the layout-level physical information neural network model, wherein the mixed loss function comprises a mean square error term of an actual geometric parameter and a predicted geometric parameter, a mean square error term of an actual scattering parameter and a reconstructed scattering parameter under frequency weighting, and simultaneously adding a physical constraint term of passivity and port matching; Acquiring an initial data set, training the layout-level physical information neural network model, and establishing a mapping relation from the layout geometric parameters to differential mode and common mode scattering parameters through the layout-level physical information neural network model; Setting the number of search agents and the maximum iteration times of an artificial lemming optimization algorithm by taking a decoder of the layout-level physical information neural network model as an electromagnetic agent model, and finishing algorithm initialization; And carrying out global search and local refinement on the layout geometric parameters by adopting the artificial lemming optimization algorithm, continuously optimizing the layout-level physical information neural network model by combining with an automatic data increment training strategy, and outputting the optimal layout geometric parameters meeting design indexes.
6. The method of designing a high temperature superconductor bandpass filter according to claim 5, wherein the step of performing global search and local refinement on the layout geometry parameters by using the artificial lemming optimization algorithm, and continuously optimizing the layout-level physical information neural network model in combination with an automatic data increment training strategy, and outputting the optimal layout geometry parameters satisfying design indexes comprises: Constructing a loss function of the artificial lemming optimization algorithm, wherein the loss function is constructed based on a return loss level and an insertion loss level of a differential mode response and an insertion loss level of a common mode response; setting the number of search agents and the maximum iteration times of the manual lemming optimization algorithm, and initializing the position parameters of each agent; Predicting scattering parameters corresponding to each agent position by using a decoder of the layout-level physical information neural network model, and calculating corresponding loss function values; And updating the proxy position according to the loss function value, executing global search, locally refining parameters near the optimal solution, and outputting the geometric parameters of the optimal layout.
7. The method of designing a high temperature superconductor bandpass filter according to claim 6, wherein the step of predicting the scattering parameters corresponding to each agent location and calculating the corresponding loss function value using the decoder of the layout-level physical information neural network model comprises: Inputting layout geometric parameters corresponding to each agent position in an artificial lemming optimization algorithm to an encoder of the trained layout-level physical information neural network model, and extracting deep feature vectors of the layout geometric parameters; Inputting the deep feature vector to the decoder, and obtaining corresponding differential mode scattering parameters and common mode scattering parameters through up-sampling and dimensional transformation of a multi-layer fully connected network; Extracting a return loss level and an insertion loss level corresponding to the differential mode scattering parameter and an insertion loss level corresponding to the common mode scattering parameter to obtain a level parameter; substituting the level parameters into a pre-constructed artificial lemming optimization algorithm loss function, and calculating the loss function value corresponding to each agent position.
8. A high temperature superconducting bandpass filter design system, characterized by being applied to the method of any one of claims 1-7, the system comprising: The device configuration module is used for determining design indexes of the filter and an extensible structural frame, wherein the design indexes comprise center frequency, relative bandwidth, filter order, differential mode return loss and common mode rejection, and the structural frame is formed by cascading a plurality of sections of coupling lines and a plurality of sections of branch lines; The first construction module is used for respectively constructing a differential mode equivalent circuit model and a common mode equivalent circuit model of the filter structure based on a transmission line theory; The second construction module is used for constructing a circuit-level physical information neural network model adopting a multi-layer perceptron architecture, and comprises the steps of setting a plurality of hidden layers, configuring a corresponding number of neurons and adding residual connection; the model training module is used for inputting the design index into the circuit-level physical information neural network model, embedding a matrix derived by an equivalent circuit and a target differential mode transfer function into a loss function of the model, and completing model training; The impedance classification module is used for outputting the even mode characteristic impedance, the odd mode characteristic impedance and the characteristic impedance of the branch line of the coupling line by using the trained circuit-level physical information neural network model; The layout generation module is used for generating the physical layout of the filter based on various characteristic impedance of the output.
9. A readable storage medium having stored thereon computer instructions which, when executed by a processor, implement the steps of the method of any of claims 1-7.
10. An electronic device comprising a memory, a processor and a computer program stored on the memory and executable on the processor, characterized in that the processor implements the steps of the method according to any one of claims 1-7 when the program is executed by the processor.

Description

Method, system, storage medium and equipment for designing superconductive band-pass filter Technical Field The invention relates to the technical field of microwave devices and circuits, in particular to a method, a system, a storage medium and equipment for designing a high-temperature superconductive band-pass filter. Background The balanced band-pass filter can inhibit common mode interference and improve the anti-interference capability of the system, and is widely applied to communication, radar and satellite systems. As systems evolve towards broadband and high sensitivity, broadband balanced bandpass filters need to achieve high level common mode rejection over a wide frequency range while achieving good differential mode passband response. To reduce insertion loss and improve out-of-band rejection, high temperature superconducting materials are introduced into the filter design. The broadband balance high-temperature superconductive band-pass filter generally adopts a complex transmission line network formed by multistage coupling lines and branch lines, the number of structural parameters is obviously increased along with the increase of the orders, the design is extremely dependent on full-wave electromagnetic simulation and manual parameter adjustment, the design period is long, and the rapid engineering is difficult to realize. In the prior art, a common data driving proxy model comprises an artificial neural network, a support vector machine, gaussian process regression and the like, and is used for replacing part of electromagnetic simulation. However, such models generally do not show the physical laws of the introduced transmission lines and electromagnetic fields, belong to the "black box" regression, have limited generalization capability, and are prone to non-physical or distortion predictions when training data is insufficient or design parameters are out of training range. Disclosure of Invention The invention aims to overcome the defects of the prior art and provide a design method, a system, a storage medium and equipment of a high-temperature superconductive band-pass filter, which aims to solve the problems described in the prior art. A first aspect of the present invention is to provide a method for designing a high-temperature superconducting bandpass filter, the method comprising: Determining design indexes of a filter and an extensible structural framework, wherein the design indexes comprise center frequency, relative bandwidth, filter order, differential mode return loss and common mode rejection, and the structural framework is formed by cascading a plurality of sections of coupling lines and a plurality of sections of branch lines; based on a transmission line theory, respectively constructing a differential mode equivalent circuit model and a common mode equivalent circuit model of the filter structure; constructing a circuit-level physical information neural network model adopting a multi-layer perceptron architecture, comprising setting a plurality of hidden layers, configuring a corresponding number of neurons, and simultaneously adding residual connection; Inputting the design index into the circuit-level physical information neural network model, embedding a matrix derived by an equivalent circuit and a target differential mode transfer function into a loss function of the model, and completing model training; Outputting the even mode characteristic impedance, the odd mode characteristic impedance and the characteristic impedance of the branch line of the coupling line by using the trained circuit-level physical information neural network model; and generating a physical layout of the filter based on the output characteristic impedance. According to an aspect of the above technical solution, the target differential mode transfer function adopts an isopipe chebyshev form, and the specific implementation includes: determining the pass band ripple level and the cut-off frequency of the chebyshev response, and calculating the corresponding electrical length; constructing a first class chebyshev polynomial function of corresponding degree; And generating a target differential mode transfer function based on the Chebyshev polynomial function, and embedding a loss function of the circuit-level physical information neural network model. According to an aspect of the foregoing technical solution, the loss function of the circuit-level physical information neural network model includes a frequency weighted loss term, and a specific implementation of the frequency weighted loss term includes: identifying a passband frequency interval and an out-of-band frequency interval of the filter; Setting high weight coefficients for frequency points in the passband frequency interval and setting low weight coefficients for frequency points in the out-of-band frequency interval; Multiplying the corresponding weight coefficient with the scattering parameter prediction error to obtain a frequ