Microwave filter shape coupling relationship hybrid modeling method, device and storage device

Abstract

The application provides a microwave filter shape-coupling relationship hybrid modeling method, device and storage device, constructs original data sets D1 and D2, extracts Y parameters y1 in D1 through a vector fitting method, constructs a data set Y1 containing resonance screw length x1 and Y parameters y1, extracts Y parameters y2 in D2, constructs a data set Y2 containing coupling screw length x2 and Y parameters y2, designs a microwave filter shape-coupling hybrid model, a shape-coupling relationship model in a serious detuning state and a shape-coupling relationship model in a slight detuning state, and realizes mapping from resonance screw length x1 and coupling screw length x2 to Y parameters y1 and y2. The application has the beneficial effects that the microwave filter is divided into two states of slight detuning and serious detuning, different structure geometric parameters are selected to establish shape-coupling relationship models, the complexity of establishing shape-coupling relationship models is reduced, and the accuracy of shape-coupling relationship models in various states is improved.

Description

Technical Field

[0001] This invention relates to the field of microwave filters, and in particular to a hybrid modeling method, device, and storage device for shape-coupled relationships in microwave filters. Background Technology

[0002] As international competition intensifies in the 5G era, the construction of 5G base stations is accelerating. Microwave filters are core passive components in 5G base stations that control the frequency response of transmitted signals. They transmit useful signal frequency components and block unwanted signal frequency components. High-performance microwave filters are crucial for optimizing spectrum resource allocation and improving the quality of communication systems. However, in the production process of microwave filters, machining tolerances are unavoidable, and microwave filters mass-produced on the production line cannot meet the factory filtering performance specifications. Therefore, performance debugging is essential.

[0003] Debugging workers use a vector network analyzer to measure scattering parameters (S-parameters) in real time and calculate filtering performance. When the filtering performance does not meet factory requirements, they select structural geometric parameters (shapes) that affect the filtering performance based on experience, determine their adjustment amounts, and change the technical indicators until they meet the requirements. However, manual debugging is inefficient, costly, and has a high scrap rate, hindering the construction of 5G base stations. Intelligent debugging has become a bottleneck that urgently needs to be overcome for the high-quality, large-scale production of microwave filters.

[0004] Accurate prediction of microwave filter performance is the foundation of intelligent debugging. It avoids frequent and blind adjustments to the structural geometric parameters of the actual microwave filter, greatly reducing debugging time, costs, and scrap rates. The structural geometric parameters ("shape") and product performance ("property") exhibit strong coupling, strong nonlinearity, and strong dynamic characteristics. Constructing a shape-property coupling model to accurately predict product performance is the basis of intelligent debugging.

[0005] During the commissioning process, the filtering performance of microwave filters changes frequently. Hybrid modeling based on the resonant state can help improve the modeling accuracy. However, existing modeling methods mostly use data-driven modeling methods such as support vector regression and neural networks to establish a shape-correlation relationship model between a certain characteristic parameter of the microwave filter and the structural geometric parameters. They do not consider the influence of the resonant state, making it difficult to fully reflect the dynamic characteristics of the microwave filter throughout the entire commissioning process, resulting in low modeling accuracy. Summary of the Invention

[0006] To address the aforementioned issues, this invention provides a hybrid modeling method, device, and storage device for shape-coupled relationships in microwave filters. Based on filtering performance, the tuning process is divided into two resonance states: severe detuning and good tuning. For different resonance states, different structural geometric parameters are selected to establish shape-coupled relationship models, thereby improving the accuracy of the shape-coupled relationship models under various states.

[0007] A hybrid modeling method for shape-coupled relationships of microwave filters, mainly including:

[0008] S1: Constructing the original datasets D1 and D2: The structural geometric parameters of the microwave filter include the resonant screw length x1 and the coupling screw length x2; by changing x1 multiple times on the microwave filter, the S-parameter s1 is sampled and measured to construct the original dataset D1 containing x1, s1 and the sampling frequency f; by changing x2 multiple times on the microwave filter, the S-parameter s2 is sampled and measured to construct the original dataset D2 containing x2, s2 and the sampling frequency f.

[0009] S2: Using vector fitting, extract the Y parameter y1 of s1 in D1 and construct a dataset Y1 containing the resonant screw length x1 and the Y parameter y1; Using vector fitting, extract the Y parameter y2 of s2 in D2 and construct a dataset Y2 containing the coupling screw length x2 and the Y parameter y2.

[0010] S3: Based on the debugging characteristics of the microwave filter and the characteristics of Y parameters y1 and y2, design a shape-coupled hybrid model of the microwave filter;

[0011] S4: Use a BP neural network to construct a shape-correlation relationship model M1 under severe detuning, and realize the mapping from the resonant screw length x1 to the Y parameter y1;

[0012] S5: Use a BP neural network to construct a shape-coupled relationship model M2 under good tuning conditions, and realize the mapping from the coupling screw length x2 to the Y parameter y2;

[0013] S6: Based on the obtained Y parameters y1 and y2, convert them into S parameters to predict the filtering performance index.

[0014] Further, in step S2, the Y matrix is ​​obtained from the S-parameters. It is a complex matrix, S 11 S represents the reflectivity of the input signal energy. 21 S represents the energy transfer rate of the input signal. 12 S represents the energy transfer rate of the output signal. 21 The reflectivity represents the energy of the output signal. The S-parameters are converted into the Y matrix using equations (1) to (4):

[0015] (1)

[0016] (2)

[0017] (3)

[0018] (4)

[0019] Where Y0 is the identity matrix, Y 11 Y represents the relationship between the input signal voltage and the input signal current. 12 This indicates the relationship between the output signal voltage and the input signal current, Y. 21 This represents the relationship between the input signal voltage and the output signal current, Y. 22 This indicates the relationship between the output signal voltage and the input signal current.

[0020] Furthermore, based on the vector fitting method, the Y parameters in the Y matrix are extracted:

[0021] (5)

[0022] Where s = jω, ω represents the angular frequency corresponding to the sampling frequency f, and λ k Let r be the k-th pole of the Y matrix. ijk Let be the residue corresponding to the k-th pole, i,j = 1,2,k = 1,2,……,N, where N represents the order of the microwave filter. The poles of the Y matrix are denoted as l=[λ1,λ2, …, λ N The residue of the Y matrix is ​​denoted as r. ij =[r ij1,…, r ijN The poles and residues of the Y matrix are called Y parameters.

[0023] Furthermore, in step S2, the Y parameter also needs to be dimensionality reduced. After dimensionality reduction, Y is selected. 11 Imaginary part of poles Im(l), Y 11 The real part of the residue Re(r) 11 ), Y 21 The real part of the residue Re(r) 21 The Y parameters are represented by y = [Im(l), Re(r)] after dimensionality reduction. 11 ), Re(r 21 )], r 11 Y represents 11 Residue, r 21 Y represents 21 Remaining numbers.

[0024] Furthermore, in step S3, the modeling process is mainly divided into two steps: based on the BP neural network, a shape-coupled relationship model of the microwave filter under severe detuning and a shape-coupled relationship model of the microwave filter under good tuning are established. The shape-coupled relationship model of the microwave filter under severe detuning and the shape-coupled relationship model of the microwave filter under good tuning are combined to obtain a hybrid shape-coupled model of the microwave filter.

[0025] A storage device that stores instructions and data for implementing a hybrid modeling method for the shape-coupled relationship of a microwave filter.

[0026] A hybrid modeling device for shape-coupled relationships of microwave filters includes: a processor and a storage device; the processor loads and executes instructions and data in the storage device to implement a hybrid modeling method for shape-coupled relationships of microwave filters.

[0027] The beneficial effects of the technical solution provided by this invention are: dividing the microwave filter into two states, slight detuning and severe detuning, and selectively choosing different structural geometric parameters to establish a shape-coupled relationship model, thereby reducing the complexity of constructing the shape-coupled relationship model and improving the accuracy of the shape-coupled relationship model under various states. Attached Figure Description

[0028] The present invention will be further described below with reference to the accompanying drawings and embodiments. In the accompanying drawings:

[0029] Figure 1 This is a flowchart of a hybrid modeling method for shape-coupled relationships of microwave filters in an embodiment of the present invention.

[0030] Figure 2 This is a schematic diagram of the shape-coupled hybrid model of the microwave filter designed in the embodiment of the present invention.

[0031] Figure 3 This is a schematic diagram of the electromagnetic model of a sixth-order microwave filter established in HFSS in an embodiment of the present invention.

[0032] Figure 4 This is a comparison diagram of the shape-correlation coupling relationship model under the severely detuned microstate in the embodiments of the present invention.

[0033] Figure 5 This is a comparison diagram of the shape-correlation coupling relationship model under a well-tuned state in the embodiments of the present invention.

[0034] Figure 6 This is a schematic diagram of the hardware device working in an embodiment of the present invention. Detailed Implementation

[0035] To provide a clearer understanding of the technical features, objectives, and effects of the present invention, specific embodiments of the present invention will now be described in detail with reference to the accompanying drawings.

[0036] Embodiments of the present invention provide a hybrid modeling method, device, and storage device for the shape-correlation relationship of microwave filters.

[0037] Please refer to Figure 1 , Figure 1This is a flowchart of a hybrid modeling method for shape-coupled relationships of microwave filters according to an embodiment of the present invention, specifically including:

[0038] S1: During the data acquisition phase, construct the original datasets D1 and D2. The structural geometric parameters of the microwave filter include the resonant screw length x1 and the coupling screw length x2. Change x1 multiple times on the microwave filter, sample and measure the S-parameter s1, and construct the original dataset D1 containing x1, s1, and the sampling frequency f. Change x2 multiple times on the microwave filter, sample and measure the S-parameter s2, and construct the original dataset D2 containing x2, s2, and the sampling frequency f.

[0039] S2: Using vector fitting, extract the Y parameter y1 of s1 in D1 and construct a dataset Y1 containing the resonant screw length x1 and the Y parameter y1; using vector fitting, extract the Y parameter y2 of s2 in D2 and construct a dataset Y2 containing the coupling screw length x2 and the Y parameter y2.

[0040] First, the S-parameters (including s1 and s2) are used to transform the Y-matrix (the Y-parameters are the poles and residues of the Y-matrix; after dimensionality reduction, the Y-parameters are the Y-matrix). 11 Imaginary part of pole Im(λ), Y 11 The real part of the residue Re(r) 11 ) and Y 21 The real part of the residue Re(r) 21 S-parameters It is a complex matrix, S 11 S represents the reflectivity of the input signal energy. 21 S represents the energy transfer rate of the input signal. 12 S represents the energy transfer rate of the output signal. 22 The reflectivity represents the output signal energy, where i is the imaginary unit. The S-parameters are converted into a Y matrix using equations (1) to (4):

[0041] (1)

[0042] (2)

[0043] (3)

[0044] (4)

[0045] In the formula, Y0 is the expression related to Y. 11 Identity matrices of the same dimension, Y 11 Y represents the relationship between the input signal voltage and the input signal current. 12 This indicates the relationship between the output signal voltage and the input signal current, Y. 21 This represents the relationship between the input signal voltage and the output signal current, Y.22 This indicates the relationship between the output signal voltage and the input signal current.

[0046] Secondly, based on the vector fitting method, the Y parameters are extracted from the Y matrix. The microwave filter Y matrix can be expressed in polynomial form as follows:

[0047] (5)

[0048] In equation (5), s = jω, where ω represents the angular frequency corresponding to the sampling frequency f, and λ k Let r be the k-th pole of the Y matrix. ijk Let be the residue corresponding to the k-th pole, i,j = 1,2, and let l = [λ1,λ2, …, λ]. The poles of the Y matrix are denoted as l = [λ1,λ2, …, λ]. N ,],k = 1,2,……,N,N represents the order of the microwave filter, and the residue of the Y matrix is ​​denoted as r ij =[r ij1,…, r ijN The poles and residues of the Y matrix are called Y parameters, which are complex numbers. In this embodiment, the Y parameters are obtained using the most commonly used vector fitting method.

[0049] Finally, dimensionality reduction is performed on the Y parameters. The extraction process of the Y parameters is based on the principle that the poles of each element in the Y matrix are equal. Therefore, in the modeling process, only the pole of a single element in the Y parameter matrix is ​​needed. In this embodiment, the pole of Y is selected. 11 The poles reflect the pole characteristics of the Y matrix; Y 11 and Y 22 Y 21 and Y 12 With essentially the same properties, this invention selects Y. 11 and Y 21 The residues are used to reflect the residue characteristics of the Y matrix; the poles and residues of the Y matrix are imaginary numbers, while the inputs and outputs of the neural network are real numbers, requiring certain processing. Therefore, the real parts of the Y matrix poles and the imaginary parts of the residues are both 0. After dimensionality reduction, this embodiment selects Y... 11 Imaginary part of poles Im(l), Y 11 The real part of the residue Re(r) 11 ), Y 21 The real part of the residue Re(r) 21 The Y parameter is reflected by y = [Im(l), Re(r)]. 11 ), Re(r 21 )], r 11 Y represents 11 Residue, r 21 Y represents 21 Remaining numbers.

[0050] S3: Overall Design of the Shape-Coupling Hybrid Model for the Microwave Filter: Based on the debugging characteristics and the characteristics of the Y-parameter, the overall design of the shape-coupled hybrid model for the microwave filter is carried out. Since this invention divides the filter into two states, it is necessary to model each state separately. The overall design structure diagram of the shape-coupled hybrid model is shown below. Figure 2 As shown, the modeling process mainly consists of two steps: establishing a shape-coupled relationship model under the severely detuned state of the microwave filter and a shape-coupled relationship model under the well-tuned state of the microwave filter. Finally, the established neural networks are combined, that is, the shape-coupled relationship model M1 under the severely detuned state of the microwave filter and the shape-coupled relationship model M2 under the well-tuned state of the microwave filter are combined to obtain a hybrid shape-coupled relationship model of the microwave filter.

[0051] S4: Constructing a shape-correlation coupling model under severe detuning: Using the most common ten-fold cross method, the dataset Y1 is divided into a training set and a test set. Referring to the block modeling method in patent CN109783905B, a BP neural network is used to construct a shape-correlation coupling model M1 under severe detuning, realizing the mapping from the resonant screw length x1 to the Y parameter y1.

[0052] S5: Constructing a shape-correlation coupling model under slight detuning: Using the most common ten-fold cross method, the dataset Y2 is divided into a training set and a test set. Referring to the block modeling method in patent CN109783905B, a BP neural network is used to construct a shape-correlation coupling model M2 under slight detuning, realizing the mapping from the coupling screw length x2 to the Y parameter y2.

[0053] Finally, the complete Y parameters are obtained by using the shape-coupled hybrid model of the microwave filter. The Y parameters are then converted into S parameters to predict the filter performance index.

[0054] A simulation and debugging platform was built using the 3D electromagnetic software HFSS and MATLAB. A simulation model was established in the 3D electromagnetic software HFSS, such as... Figure 3 The simulation model of a sixth-order non-cross-coupled microwave filter is shown, and its S-parameters are solved. A shape-coupled relationship model is constructed in MATLAB. The designed microwave filter has 6 resonant screws and 5 coupling screws.

[0055] like Figure 2 As shown, in cases of severe detuning, the resonant screw length, the imaginary part data set of the poles of the Y parameters, and the Y... 11 and Y 21The residue real part dataset is used to train neural networks 1, 2, and 3. After all three neural networks (i.e., BP neural network 1, BP neural network 2, and BP neural network 3) are trained, the modeling process of the severely detuned filter is measured using a test set. Given the length of the resonant screw, the Y parameters are obtained by combining the outputs of the three neural networks (i.e., the output of the shape-coupled relationship model under severe detuning), and then converted into S parameters for display. The amplitude and phase response curves of the ideal S parameters are compared with the S parameters output by the model. m The relationship between amplitude and phase response curves is as follows: Figure 4 As shown, both the amplitude-frequency curve and the phase-frequency curve fit well, indicating that the shape-coupled relationship model constructed in this invention for the severely detuned state is effective. When the microwave filter is severely detuned, the resonant screw needs to be adjusted. At this time, the shape-coupled relationship model M1 under the severely detuned state can predict the filtering performance.

[0056] In the case of slight detuning, the imaginary part of the poles of the coupled screw length and the Y parameter are used as the basis for the calculation. 11 and Y 21 The residue real part dataset is used to train neural networks 4, 5, and 6. After all three neural networks (i.e., BP neural network 4, BP neural network 5, and BP neural network 6) are trained, the modeling process of the slightly detuned filter is measured using a test set. Given the length of the coupling screw, the Y parameters are obtained by combining the outputs of the three neural networks (i.e., the output of the shape coupling relationship model under slightly detuned state), and then converted into S parameters for display. The amplitude and phase response curves of the ideal S parameters are compared with the S parameters of the model output. m The relationship between amplitude and phase response curves is as follows: Figure 5 As shown, both the amplitude-frequency curve and the phase-frequency curve fit well, indicating that the shape-coupled relationship model constructed in this invention for the slightly detuned state is effective. When the microwave filter is slightly detuned, the coupling screw needs to be adjusted. At this time, the shape-coupled relationship model M2 under the slightly detuned state can predict the filtering performance.

[0057] Please see Figure 6 , Figure 6 This is a schematic diagram of the hardware device in operation according to an embodiment of the present invention. The hardware device specifically includes: a microwave filter shape-coupled relationship hybrid modeling device 401, a processor 402, and a storage device 403.

[0058] A hybrid modeling device 401 for shape-coupled relationships of microwave filters: The hybrid modeling device 401 for shape-coupled relationships of microwave filters implements the hybrid modeling method for shape-coupled relationships of microwave filters.

[0059] Processor 402: The processor 402 loads and executes the instructions and data in the storage device 403 to implement the hybrid modeling method for shape-coupled relationship of microwave filters.

[0060] Storage device 403: The storage device 403 stores instructions and data; the storage device 403 is used to implement the hybrid modeling method for shape-coupled relationship of microwave filters.

[0061] The beneficial effects of this invention are: it divides microwave filters into two states, slight detuning and severe detuning, and selects different structural geometric parameters to establish shape-correlation relationship models, thereby reducing the complexity of constructing shape-correlation relationship models and improving the accuracy of shape-correlation relationship models under various states.

[0062] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A hybrid modeling method for shape-correlation coupling relationships of microwave filters, characterized in that: include: S1: Constructing the original datasets D1 and D2: The structural geometric parameters of the microwave filter include the resonant screw length x1 and the coupling screw length x2; by changing x1 multiple times on the microwave filter, the S-parameter s1 is sampled and measured to construct the original dataset D1 containing x1, s1 and the sampling frequency f; by changing x2 multiple times on the microwave filter, the S-parameter s2 is sampled and measured to construct the original dataset D2 containing x2, s2 and the sampling frequency f. in, It is a complex matrix, S 11 S represents the reflectivity of the input signal energy. 21 S represents the energy transfer rate of the input signal. 12 S represents the energy transfer rate of the output signal. 22 Represents the reflectivity of the output signal energy; S2: Using vector fitting, extract the Y parameter y1 of s1 in D1 and construct a dataset Y1 containing the resonant screw length x1 and the Y parameter y1; Using vector fitting, extract the Y parameter y2 of s2 in D2 and construct a dataset Y2 containing the coupling screw length x2 and the Y parameter y2. The Y matrix is ​​obtained by transforming the S parameters, where the Y parameters are the poles and residues of the Y matrix; S3: Based on the debugging characteristics of the microwave filter and the characteristics of Y parameters y1 and y2, design a shape-coupled hybrid model of the microwave filter; The modeling process of the shape-coupled hybrid model of microwave filter is as follows: Based on the BP neural network, a shape-coupled relationship model of microwave filter under severe detuning state and a shape-coupled relationship model of microwave filter under good tuning state are established. The shape-coupled relationship model of microwave filter under severe detuning state and the shape-coupled relationship model of microwave filter under good tuning state are combined to obtain the shape-coupled hybrid model of microwave filter. S4: Use a BP neural network to construct a shape-correlation relationship model M1 under severe detuning, and realize the mapping from the resonant screw length x1 to the Y parameter y1; S5: Use a BP neural network to construct a shape-coupled relationship model M2 under good tuning conditions, and realize the mapping from the coupling screw length x2 to the Y parameter y2; S6: Based on the obtained Y parameters y1 and y2, convert them into S parameters to predict the filtering performance index.

2. The hybrid modeling method for shape-coupled relationships of microwave filters as described in claim 1, characterized in that: In step S2, the Y matrix is ​​obtained by transforming the S parameters. It is a complex matrix, S 11 S represents the reflectivity of the input signal energy. 21 S represents the energy transfer rate of the input signal. 12 S represents the energy transfer rate of the output signal. 22 The reflectivity represents the energy of the output signal. The S-parameters are converted into the Y matrix using equations (1) to (4): (1) (2) (3) (4) Where Y0 is the identity matrix, Y 11 Y represents the relationship between the input signal voltage and the input signal current. 12 This indicates the relationship between the output signal voltage and the input signal current, Y. 21 This represents the relationship between the input signal voltage and the output signal current, Y. 22 This indicates the relationship between the output signal voltage and the input signal current.

3. The hybrid modeling method for shape-coupled relationships of microwave filters as described in claim 1, characterized in that: In step S2, the Y parameters in the Y matrix are extracted based on the vector fitting method: (5) Among them, Y 11 Y represents the relationship between the input signal voltage and the input signal current. 12 This indicates the relationship between the output signal voltage and the input signal current, Y. 21 This represents the relationship between the input signal voltage and the output signal current, Y. 22 This represents the relationship between the output signal voltage and the input signal current, s = jω, where ω represents the angular frequency corresponding to the sampling frequency f, and λ k Let r be the k-th pole of the Y matrix. ijk Let be the residue corresponding to the k-th pole, i, j = 1,2, k = 1,2,……,N, where N represents the order of the microwave filter. The poles of the Y matrix are denoted as l=[λ1, λ2, …, λ N The residue of the Y matrix is ​​denoted as r. ij =[r ij1,…, r ijN The poles and residues of the Y matrix are called Y parameters.

4. The hybrid modeling method for shape-coupled relationships of microwave filters as described in claim 3, characterized in that: In step S2, the Y parameter also needs to be dimensionality reduced. After dimensionality reduction, Y is selected. 11 Imaginary part of poles Im(l), Y 11 The real part of the residue Re(r) 11 ), Y 21 The real part of the residue Re(r) 21 The Y parameters are represented by y = [Im(l), Re(r)] after dimensionality reduction. 11 ), Re(r 21 )], r 11 Y represents 11 Residue, r 21 Y represents 21 Remaining numbers.

5. A storage device, characterized in that: The storage device stores instructions and data to implement the hybrid modeling method for shape-coupled relationships of microwave filters as described in any one of claims 1 to 4.

6. A hybrid modeling device for the shape-correlation relationship of microwave filters, characterized in that: include: A processor and a storage device; the processor loads and executes instructions and data in the storage device to implement the hybrid modeling method for shape-coupled relationships of microwave filters as described in any one of claims 1 to 4.

Images