CN116663476A - A High-Order High-Temperature Superconducting Filter and Its Circuit Topology Rapid Design Method - Google Patents

Abstract

The invention provides a high-order high-temperature superconducting filter and a circuit topology structure rapid design method thereof, which adopt a twice optimization method and combine sonnet electromagnetic simulation software and a neural network, so that the circuit topology structure design optimization time of the high-order superconducting filter is shortened, and the efficiency is improved. The first optimization method optimizes the circuit topology by comparing the difference between the ideal coupling matrix of the filter and the coupling matrix extracted from the actual circuit. And secondly, optimizing, namely generating samples by determining the initial value and the range of a circuit topological structure and utilizing sonnet software to train a neural network, wherein a trained model can accurately predict the electromagnetic simulation result of an actual circuit within 0.1s, so that the design efficiency of the high-order high-temperature superconducting filter is greatly improved. The example proves that the method carries out the design of the high-order superconducting filter, and the obtained result has good matching degree with the simulation result of electromagnetic simulation software under the complex coupling condition including non-adjacent coupling.

Description

A High-Order High-Temperature Superconducting Filter and Its Circuit Topology Rapid Design Method

technical field

The invention belongs to the field of radio frequency microwave circuit design and the field of microwave communication, in particular to an efficient and rapid design method for a circuit topology of a high-order high-temperature superconducting filter with an L-band and a bandwidth within 3%.

Background technique

Filters are a key component of electronic systems. Microwave filters select desired useful microwave signals and suppress out-of-band interference. With the continuous development of microwave systems, the performance requirements of filters are getting higher and higher. However, in the reported work, the traditional microstrip filter is usually 2nd or 4th order, because the insertion loss increases significantly with the increase of the filter order, resulting in the high-order microstrip filter insertion loss is too large to be able to This is because the Q value of microstrip resonators designed based on conventional materials is low, and when the relative bandwidth of the filter is less than 5%, the insertion loss increases significantly. The Q value of the microstrip resonator realized by superconducting technology can reach more than 10,000, and the filter can achieve high order, such as 8th order, 10th order or higher, and has very low insertion loss and higher selectivity.

The design of high-order superconducting filters requires long-term optimization of circuit topology parameters through electromagnetic simulation to achieve ideal performance. However, the optimization of the filter topology is highly dependent on the designer's experience. Especially as the filter order increases, the circuit topology becomes more and more complex, and the traditional electromagnetic simulation optimization takes longer and longer energy and time, even up to several months, which is unacceptable in practical engineering applications. There is an urgent need for an efficient and accurate method for rapid design of high-order HTS filter topology.

An effective way to reduce filter topology optimization time is to find a more efficient model to replace the time-consuming electromagnetic simulation process. The present invention combines the Sonnet electromagnetic simulation software with the artificial neural network method, proposes a high-efficiency high-order superconducting filter circuit topology design method based on two optimizations, and solves the problem of low design efficiency of high-order complex superconducting filter circuits .

Contents of the invention

The purpose of the present invention is to propose a high-order high-temperature superconducting filter and its circuit topology rapid design method to solve the problem of high-order high-temperature superconducting filter (the order is 8 or 10 or more) circuit topology design process. rely on design

The problem of slow speed and low efficiency of electromagnetic simulation design based on the experience of the teacher.

The high-order high-temperature superconducting filter is composed of an input feeder and an output feeder with a characteristic impedance of 50 ohms, and multiple half-wavelength resonators; the form of the resonator can be single helix or double helix. The combination of multiple resonators forms the basic circuit topology of HTS filters.

A rapid circuit topology design method for high-order high-temperature superconducting filters is based on two optimization processes, combining sonnet simulation software with artificial neural networks, replacing part of the electromagnetic simulation with a more efficient model, and greatly improving the high-order high-temperature Design efficiency of superconducting filters.

In order to realize the purpose of the present invention, adopt following technical scheme: the circuit topology fast design method of high-order high-temperature superconducting filter combines sonnet electromagnetic simulation and neural network, carries out modeling, by two optimization processes, to the filter The circuit topology is adjusted to quickly and efficiently find the optimal circuit topology parameters.

The process flow of the method is shown in FIG. 1 . Firstly, the theoretically calculated ideal coupling matrix is obtained according to the filter synthesis design theory, and the initial filter circuit topology is constructed according to the ideal coupling matrix, and geometric variables are defined for the topology. Then call Sonnet for electromagnetic simulation on the initial topology structure, and extract the circuit coupling matrix of the actual circuit according to the results of Sonnet electromagnetic simulation. The first optimization is performed by comparing the gap between the ideal coupling matrix and the actual circuit coupling matrix. By adjusting the geometric variables of the circuit topology, the difference between the actual coupling matrix and the ideal coupling matrix is minimized. Afterwards, the neural network model is constructed, and the training samples come from the result response obtained from the electromagnetic simulation of the circuit topology using Sonnet, and the value range of the circuit topology parameters is predefined. Afterwards, the fully trained neural network model is optimized for the second time until the preset conditions are met.

The optimization process is divided into two steps: the first optimization (preliminary optimization) and the second optimization.

In the first optimization, the circuit coupling matrix of the real filter circuit topology is extracted from the sonnet electromagnetic simulation response by using the vector fitting method and matrix similarity transformation. Compare the extracted real coupling matrix with the theoretically calculated ideal coupling matrix to optimize the circuit topology parameters for the first time. The circuit coupling coefficient is the key optimization parameter, and the circuit coupling coefficient corresponds to the designed circuit structure one by one.

The adjustment of the circuit coupling coefficient can be realized by changing the distance between the corresponding resonances, and the matrix elements with the largest gap are preferentially adjusted until the difference between them is the smallest, and the first optimization of the extracted circuit coupling matrix is completed.

The second optimization, using the samples generated by sonnet electromagnetic simulation results, establishes and trains the artificial neural network model to provide accurate and fast electromagnetic response prediction, and implements the condition of S 11≤-20 dB in the passband, for circuit geometry Variables are iterated until a condition is met.

The neural network is shown in Figure 2. It consists of a three-layer neural network, and a transformation from the coupling matrix to the filter frequency response. The vector of the geometric variable x of the filter circuit topology is selected as the input of the neural network, and the coupling matrix [ M ] is output. The basis of this approach is that all coupling matrix elements, including non-adjacent couplings, are determined by the topology of the filter circuit.

The definition of the geometry variable of the filter circuit topology is as follows, the geometry variable is an l- dimensional vector:

Coupling element m _ij can be expressed as a function of topological geometric variables in l- dimensional vector space. Use the artificial neural network to learn the relationship between the geometric variable x and the coupling matrix m _ij :

After obtaining the coupling matrix [ M ], the [ S ] parameter of the circuit topology can be calculated through a theoretical formula, and the formula is as follows:

where [ M ] is the actual circuit coupling matrix extracted from the filter topology, [ I ] is the (n+2)×(n+2) identity matrix, s is the complex frequency s = jw , and [ R ] is a full Zero matrix, except that R ₁₁ = R _{n+2, where n+2} = 1.

The extraction of the coupling matrix mainly includes the following main steps: obtain the [ S ] parameter through sonnet electromagnetic simulation; carry out phase de-embedding to the [ S ] parameter; convert the [ S ] parameter into the admittance parameter [ Y ] parameter; Legally extract the poles and residues of the [ Y ] parameter rational function; find the matrix element values of the N+2 order transverse matrix; convert the transverse matrix into the coupling structure of the actual filter through similar transformation, and find the actual coupling matrix; finally pass Find the [ S ] parameter from the coupling matrix.

The scattering parameter [ S ] is converted into the admittance parameter [ Y ] using the following formula:

Use the vector fitting algorithm to fit the above [ Y ] parameter curve to obtain a rational function. The mathematical representation of a rational function is as follows:

where the highest degree of the numerator and denominator is equal to the order of the filter. The admittance parameters y ₁₁ , y ₁₂ , y ₂₁ and y ₂₂ of the filter share the same root and should be fitted in one run. Rewrite the rational function above the expression in matrix form:

Since a typical two-port microwave filter can be equivalent to a horizontal network connected in parallel. Its admittance can be written as a transverse array connected in parallel:

Among them, M _Sk and M _Lk represent the coupling between resonator k and the source/load port, M _SL is the coupling between source and load, and B _k is the self-coupling of resonator k . Comparing the admittance Y (s) transformed from the [ S ] parameter of the actual filter with the rational function of the transverse network admittance array Y _N (s) , each matrix element of the N+2 order transverse matrix can be obtained value of:

As in the above moment [ M _p ], it represents an N+2 coupling matrix, there is no coupling between resonators, and there is only coupling between the input and output ports. However, the coupling structure of the actual filter circuit is different from it, and there will be coupling between adjacent resonators. In order to transform the transverse coupling structure into the actual coupling structure, the matrix [ M _p ] performs a similar transformation rotation operation, and the rotation operation The mathematical representation looks like this:

where R ^θ _{(i, j)} is the rotation matrix. By selecting the appropriate rotation axis [ i, j ] and rotation angle θ , the non-coupling matrix elements in the transverse matrix [ M _p ] can be eliminated and transformed into the actual coupling structure of the filter.

The vector x composed of geometric variables of the selected filter topology is used as the input of the model, and the sonnet simulation software is called to obtain the simulation response curve of the circuit topology, and the simulation result is used to train the neural network. During the training process, the connection weights and bias parameters in the neural network are updated using the backpropagation algorithm to minimize the training error as follows:

in is the coupling matrix element value output by the neural network, and m _ij is the training coupling matrix element value extracted from the sonnet simulation response. E _k is the training error (squared error) of the kth sample, and N is the order of the filter. E is the mean square error of all training samples, and n is the total number of training samples. In a well-trained neural network, the output value /> is expected to be equal to the training value m _ij .

High-temperature superconducting filters generally use a 500 nm thick YBCO film deposited on a 0.5 mm thick LaAlO3 substrate with a relative permittivity of 23.75.

Features and beneficial effects of the present invention:

The present invention proposes a high-order high-temperature superconducting filter and a rapid design method for its circuit topology. It adopts a two-time optimization method, combined with the characteristics of sonnet electromagnetic simulation software and artificial neural network, which greatly improves the design efficiency, and Filters for high-order complex couplings (including non-phase couplings) are fast. The invention is applicable to the superconducting filter design of high-order and complex circuit topological structure in L band and higher frequency. The geometric parameters of the filter circuit topology can be quickly obtained. It is verified by an example that the circuit response obtained by this method is consistent with the simulation result of the electromagnetic simulation software.

Description of drawings

Fig. 1 is a flow chart of the rapid design method of high-order high-temperature superconducting filter circuit topology of the present invention;

The neural network training model based on sonnet electromagnetic simulation of Fig. 2 of the present invention;

The ideal coupling matrix obtained by the theoretical synthesis of the 8th-order high-temperature superconducting filter of the present invention of Fig. 3;

The 8-order high-temperature superconducting filter sonnet simulation connection of Fig. 4 of the present invention extracts the circuit coupling matrix of the actual circuit;

Fig. 5 definition of geometric parameter of circuit topology structure of 8th-order high-temperature superconducting filter of the present invention;

Fig. 6 is the 8th-order high-temperature superconducting filter initialization circuit simulation result and the first optimization result of the present invention;

Fig. 7 compares the final parameter prediction curve of the 8th-order high-temperature superconducting filter of the present invention with the electromagnetic simulation curve;

Fig. 8 is a schematic diagram of a high temperature superconducting filter of the present invention;

Fig. 9 8th-order and high-temperature superconducting filter actual measurement results of the present invention;

Fig. 10 Definition of geometrical parameters of the circuit topology of the 10th-order high-temperature superconducting filter of the present invention;

Fig. 11 is the ideal coupling matrix obtained by theoretical synthesis of the 10th-order high-temperature superconducting filter of the present invention;

Fig. 12 is the 10th-order high-temperature superconducting filter initialization circuit simulation result and the first optimization result of the present invention;

Fig. 13 compares the final parameter prediction curve of the 10th-order high-temperature superconducting filter of the present invention with the electromagnetic simulation curve;

Fig. 14 is the measured result of the 10-order high-temperature superconducting filter of the present invention.

Detailed ways

The present invention will be described in detail below in conjunction with the accompanying drawings and specific embodiments, but not as a limitation of the present invention.

The invention proposes a high-order high-temperature superconducting filter and a rapid design method for its circuit topology. Using conventional electromagnetic simulation, relying on the designer's experience to design the topology parameters of the circuit, it usually takes several weeks to obtain the S-parameter response curve that meets the requirements. The present invention is based on two optimization processes, and can be used for fast prediction of filter response and function of geometric variables from sonnet electromagnetic simulation and neural network methods. In order to prove the effectiveness of the method, the present invention will be further described below in conjunction with the accompanying drawings and specific embodiments.

Embodiment 1: 8-order high-temperature superconducting filter and its circuit topology rapid design method;

The 8th-order high-temperature superconducting filter consists of an input feeder and an output feeder with a characteristic impedance of 50 ohms, and 8 half-wavelength resonators.

The center frequency and bandwidth of this high-order filter are set to f ₀ = 508 MHz and BW = 8 MHz, respectively.

Step 1, the initial value of the circuit topology of the high-order superconducting filter is determined: according to the requirements of the index, the ideal coupling matrix is obtained through the filter synthesis theory, as shown in Figure 3. According to the ideal coupling matrix, determine the initial topology of the filter circuit topology, as shown in Figure 5, select six geometric variables of the circuit topology as input variables, namely x = { q ₁ , L ₁ , g ₁ , g ₂ , g ₃ , g ₄ }. q ₁ is the length of the feeder. L ₁ represents the parameters of the first resonator. g ₁ , g ₂ , g ₃ , g ₄ are the distances between adjacent resonators.

Step 2, filter circuit topology matrix extraction: extract the actual circuit coupling matrix of the circuit topology according to the above method.

Step 3, the initial optimization of the geometric parameters of the circuit topology: compare the ideal coupling matrix obtained in step 1 with the actual circuit coupling matrix extracted in step 2, and optimize the circuit coupling matrix by adjusting the circuit coupling coefficient.

The adjustment of the circuit coupling coefficient is realized by changing the spacing between the corresponding resonances. The matrix elements with the largest gap are adjusted preferentially until the difference between them is the smallest. After the first optimization of the extracted circuit coupling matrix is completed, as shown in Figure 4 Show

Step 4, determine the range of sample training, and generate training samples through sonnet simulation response results. The initial value set of parameters in this example is {3.32, 5.04, 0.98, 1.58, 1.46}, and the range of parameters is shown in Table 1. 81 samples were generated by Sonnet software and randomly scattered in a grid form within the training range.

Table 1 Geometric parameter range of 8th-order HTS filter circuit topology

Step five, extract all sample coupling matrices.

Step 6, training the neural network based on the samples obtained in steps 4 and 5: using a three-layer neural network including 11 neurons in the hidden layer neural network to learn the relationship between the geometric variables and the coupling matrix. Among all samples, 65 samples are used for training and 16 samples are used for testing.

Step 7: Use the neural network for the second optimization: For the fully trained neural network, use the index condition max (S 11) ≤ -20 dB (504~512 MHz) for the second optimization. Geometry parameters are further tuned. The optimal solution is x = {3.92, 4.88, 0.92, 1.56, 1.44, 1.64}.

Step 8: Bring the geometric parameters of the final circuit topology into Sonnet software for electromagnetic simulation, and compare with the prediction results of the neural network.

In this example, as shown in Table 2, the total time to train the neural network model is 10 h, however, once the model is well trained, it takes less than 0.1 seconds to predict the response of the neural network model. Then the obtained circuit topology parameters are simulated by Sonnet, as shown in Figure 7, the simulation response is in good agreement with the model response and meets the design specifications. The high-temperature superconducting filter uses a 500 nm thick YBCO film deposited on a 0.5 mm thick LaAlO3 substrate with a relative permittivity of 23.75. A schematic diagram of the filter is shown in Figure 8. The filter was measured with an Agilent E5072A network analyzer at 65 K, and the measured response is shown in Figure 9.

Table 2 Comparison of the time used by the method of the present invention and conventional electromagnetic simulation in the filter design of section 8

Embodiment 2: 10-order non-adjacent coupled high-temperature superconducting filter topology fast optimization model;

The 10th-order high-temperature superconducting filter consists of an input feeder and an output feeder with a characteristic impedance of 50 ohms, and 10 half-wavelength resonators.

In order to improve the selectivity of the passband, the order of the filter is increased to 10, and a negative cross-coupling is added between resonator 2 and resonator 9 to introduce a transmission zero. As _shown in Figure 10, set the parameters v1 and h1 to _control the negative cross-coupling. The filter has a center frequency of f ₀ = 1.025 GHz and a bandwidth of 0.02 GHz.

Its quick optimization steps are the same as Example 1.

Firstly, according to the filter synthesis theory, the ideal coupling matrix is obtained as shown in Figure 11, and the initial circuit topology of the filter is established, and nine circuit topology geometric variables x = { q ₁ , L ₁ , L ₂ , h ₁ , g ₁ , g ₂ , g ₃ , g ₄ , g ₅ } are input vectors, as shown in Figure 10. _The variable h1 is used to adjust the coupling strength between resonators 2 and 9. Through sonnet simulation, the frequency response is obtained, the coupling matrix of the actual circuit is extracted, and compared with the ideal coupling matrix, the filter structure parameters are initially optimized. The result is that the maximum return loss is -13 dB, which cannot meet the application requirements. The results of the first optimization are shown in Figure 12.

Continue to carry out the second optimization, determine the range of samples according to the results of the first optimization, and obtain the response curves of all samples through Sonnet, and extract the coupling matrix to train the neural network. The range definition of the training data is shown in Table 4. Within this range, a total of 80 training samples and 20 testing samples are randomly distributed. Train a three-layer neural network with 9 input neurons, 18 hidden layer neurons and 144 output neurons to map the input-output relationship. In this example, it took 21 hours to generate training and test samples and train the neural network model. Add the index requirement, implement the condition of S11<-20 dB in the passband 1.015-1.035 GHz. For each iteration of the geometric variables, the results obtained using the neural network model only need 0.06 s. The optimal geometric parameters of the topological structure of the 10th-order high-temperature superconducting filter given by the model are x = {1.36, 1.16, 1.36, 1.14, 1.3, 0.92, 1.76, 2.2, 4.2}, which are verified by electromagnetic simulation, as shown in the figure As shown in 13, the electromagnetic simulation result is consistent with the optimized result of the present invention. Finally, the schematic diagram of the filter made of the same high-temperature superconducting material as in Example 1 is shown in Figure 8, and the measured performance is shown in Figure 14. As shown in Table 3, by adopting the method of the present invention, the design time of the topological structure of the 10-order high-temperature superconducting filter circuit is greatly shortened, and the efficiency is obviously improved.

Table 3 Comparison of the time used by the method of the present invention and the conventional electromagnetic simulation in the design of the 8-section filter

The above is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto, any person familiar with the technical field within the technical scope disclosed in the present invention, according to the technical solution of the present invention Any equivalent replacement or change of the inventive concepts thereof shall fall within the protection scope of the present invention.

Claims (9)

1. A circuit topology rapid optimization design method of a high-order high-temperature superconducting filter is characterized in that the circuit topology rapid optimization design method comprises two optimizations, wherein the first optimization compares a circuit coupling matrix actually extracted by a circuit with an ideal coupling matrix, and adjusts the part with the largest phase difference until the difference between the two is the smallest; the second optimization is based on the optimization of a fully trained neural network model, and the optimal solution of the circuit topology structure is obtained by executing the condition that S11 is less than or equal to-20 and dB in the pass band of the filter.

2. The method for rapidly optimizing and designing the circuit topology of the high-order high-temperature superconducting filter according to claim 1, wherein the ideal coupling matrix in the first optimization is compared with the circuit coupling matrix actually extracted by the circuit, modeling and simulating the circuit topology of the high-order superconducting filter by using sonnet software, and extracting the coupling matrix from electromagnetic simulation response; the first optimized comparison parameters are mainly circuit coupling coefficients of a coupling matrix actually extracted by a circuit and ideal coupling coefficients of an ideal coupling matrix.

3. The method for fast optimizing circuit topology of high-order high-temperature superconductive filter according to claim 2, wherein the circuit coupling matrix includes coupling effects of all distribution parameters in the actual filter circuit topology, including adjacent coupling, non-adjacent coupling, and even higher non-adjacent coupling of the filter.

4. The method for rapid optimum design of circuit topology of a high-temperature order superconducting filter according to claim 3, wherein the circuit topology of the actual filter comprises a coupling structure; the coupling structure of the filter may be a simple coupling form, such as a series coupling, where the resonator is coupled only to adjacent resonators, or a more complex coupling form, including not only adjacent occasional sums but also non-adjacent couplings.

5. The method for fast optimizing circuit topology of high-order high-temperature superconductive filter according to claim 2, wherein the first optimization is characterized in that the extracted circuit coupling coefficients are in one-to-one correspondence with the designed circuit structure, the differences between the extracted circuit coupling coefficients and the ideal coupling coefficients are compared, the position of the circuit topology corresponding to the circuit coupling matrix element with the largest difference is found, and the distance between the corresponding resonators is adjusted, so that the difference between the circuit coupling coefficients and the ideal coupling matrix is reduced.

6. The method for rapid optimization design of circuit topology of high-order high-temperature superconducting filter according to claim 1, wherein the second optimization uses sonnet electromagnetic simulation software to perform electromagnetic simulation on the circuit topology to generate samples, trains a neural network, and thus obtains a corresponding relation between geometric parameters and performance response of the circuit topology of the filter, and performs rapid design.

7. The method for rapid optimization design of circuit topology of high-order high-temperature superconducting filter according to claim 1, wherein a design model combining sonnet and neural network is established, and precise modeling and rapid optimization are performed on the circuit topology of the filter.

8. The high-order high-temperature superconducting filter designed by the circuit topology rapid optimization design method of the high-order high-temperature superconducting filter according to any one of claims 1-7, wherein the high-order high-temperature superconducting filter consists of an input feeder line and an output feeder line with characteristic impedance of 30-100 ohms and a plurality of resonators; the resonator may be in the form of a single helix, a double helix.

9. The high-order high-temperature superconducting filter designed by the circuit topology rapid optimization design method of the high-order high-temperature superconducting filter according to claim 8, wherein the high-order high-temperature superconducting filter is formed by depositing a thin film with the thickness of 400-600 nm on a substrate with the thickness of 0.4-0.6 mm, and the relative dielectric constant is 23-25.