CN113158541A - Multi-feature fusion modeling debugging method for microwave filter - Google Patents

Multi-feature fusion modeling debugging method for microwave filter Download PDF

Info

Publication number
CN113158541A
CN113158541A CN202110128530.4A CN202110128530A CN113158541A CN 113158541 A CN113158541 A CN 113158541A CN 202110128530 A CN202110128530 A CN 202110128530A CN 113158541 A CN113158541 A CN 113158541A
Authority
CN
China
Prior art keywords
feature
debugging
microwave filter
frequency response
layer
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Granted
Application number
CN202110128530.4A
Other languages
Chinese (zh)
Other versions
CN113158541B (en
Inventor
曹卫华
郭琳炜
毕乐宇
袁艳
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
China University of Geosciences
Original Assignee
China University of Geosciences
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by China University of Geosciences filed Critical China University of Geosciences
Priority to CN202110128530.4A priority Critical patent/CN113158541B/en
Publication of CN113158541A publication Critical patent/CN113158541A/en
Application granted granted Critical
Publication of CN113158541B publication Critical patent/CN113158541B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F30/00—Computer-aided design [CAD]
    • G06F30/20—Design optimisation, verification or simulation
    • G06F30/27—Design optimisation, verification or simulation using machine learning, e.g. artificial intelligence, neural networks, support vector machines [SVM] or training a model
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/25—Fusion techniques
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/045—Combinations of networks
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/08—Learning methods

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Evolutionary Computation (AREA)
  • Artificial Intelligence (AREA)
  • General Physics & Mathematics (AREA)
  • General Engineering & Computer Science (AREA)
  • Data Mining & Analysis (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Software Systems (AREA)
  • Computing Systems (AREA)
  • Molecular Biology (AREA)
  • General Health & Medical Sciences (AREA)
  • Computational Linguistics (AREA)
  • Biophysics (AREA)
  • Mathematical Physics (AREA)
  • Biomedical Technology (AREA)
  • Health & Medical Sciences (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Bioinformatics & Computational Biology (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • Evolutionary Biology (AREA)
  • Medical Informatics (AREA)
  • Computer Hardware Design (AREA)
  • Geometry (AREA)
  • Measurement Of Resistance Or Impedance (AREA)
  • Radar Systems Or Details Thereof (AREA)

Abstract

The invention provides a multi-feature fusion modeling debugging method for a microwave filter, which comprises the steps of changing the state x of an adjustable component for many times, sampling and measuring S parameters S, constructing an original data set containing x, S and sampling frequency f, preprocessing S and f to obtain original features, and forming a training set by the state x of the adjustable component and the original features; constructing a feature fusion part of the model by using the convolutional layer, the pooling layer and the activation function layer, constructing a feature mapping part of the model by using the full-connection layer, and obtaining a high-precision multi-feature debugging decision model through training; and (3) carrying out the same pretreatment on the S parameters S and f meeting the index requirements, inputting the S parameters S and f into the trained high-precision multi-feature debugging decision model to obtain the corresponding adjustable component state x, and adjusting the adjustable component state of the microwave filter to be debugged to x. The invention has the beneficial effects that: the accuracy of the debugging decision model is improved, and the debugging efficiency is further improved.

Description

Multi-feature fusion modeling debugging method for microwave filter
Technical Field
The invention relates to the field of intelligent manufacturing, in particular to the field of microwave filter debugging, and particularly relates to a multi-feature fusion modeling debugging method for a microwave filter.
Background
With the international increasing competition of 5G, China successively puts forward important documents to accelerate the construction of 5G base stations. The microwave filter is a core frequency-selecting device in the 5G base station, and the filtering performance of the microwave filter has great influence on the frequency-selecting quality. However, in the production process, due to unavoidable machining tolerance, the microwave filter cannot meet the requirement of the filtering performance index generally, and the debugging process is indispensable. The traditional debugging method relies on experienced debugging workers, calculates the difference between the current performance index and the target performance index according to the measured scattering parameters (S parameters) of the microwave filter, correspondingly adjusts the state of an adjustable component, changes the internal electromagnetic relation, and enables the performance index of the adjusted microwave filter to meet the requirement. However, the manual debugging method is high in cost and low in efficiency, and the construction process of the 5G base station is blocked, so that the intelligent debugging algorithm is urgently needed.
The debugging algorithm based on the debugging decision model is an efficient intelligent debugging algorithm. The debug decision model implements a mapping from the output response to the tunable component state. Since the mapping is difficult to obtain analytically, it is usually implemented using a data-driven neural network. The existing debugging decision model only utilizes S parameters, does not utilize target performance indexes, cannot comprehensively reflect the characteristics of the debugging process, has low precision and affects the debugging efficiency. On the other hand, the dimension difference between the S parameter and the target debugging index is large, and the relation between the S parameter and the target debugging index is difficult to express. Aiming at the problem, the invention designs an effective data processing mode according to the spatial relationship between the S parameter and the target index in the debugging process, performs feature level fusion and finally establishes a high-precision multi-feature debugging decision model.
Disclosure of Invention
In order to solve the above problems, the present invention provides a microwave filter multi-feature fusion modeling debugging method, which mainly comprises the following steps:
s1: changing the state x of the adjustable component on the microwave filter for multiple times, sampling and measuring S parameterNumber s, constructing an original data set D containing x and s and a sampling frequency fraw；
S2: for the original data set DrawCarrying out data preprocessing to obtain an original characteristic RrawConstructing a model containing the tunable part state x and the original feature RrawThe training set of (2);
s3: constructing a feature fusion part of a multi-feature debugging decision model by using a convolutional layer, a pooling layer and an activation function layer in a neural network, and processing original features into fusion features;
s4: using a full connection layer to construct a feature mapping part of a multi-feature debugging decision model, wherein the feature mapping part is used for mapping the fusion features to an adjustable component state x;
s5: training the feature fusion part and the feature mapping part by using a training set to obtain a trained high-precision multi-feature debugging decision model;
s6: and preprocessing data of the S parameter S meeting the index requirement and the sampling frequency f, inputting the data into the trained high-precision multi-feature debugging decision model to obtain a corresponding adjustable component state x, and adjusting the adjustable component state of the microwave filter to be debugged to x, thereby realizing high-precision debugging of the microwave filter.
Further, using the original data set DrawThe S parameter S and the sampling frequency f in the process are calculated to obtain amplitude-frequency response and phase-frequency response, and original characteristics Rraw are obtained by combining target indexes, wherein the process mainly comprises the following three steps:
firstly, obtaining amplitude-frequency response and phase-frequency response; the S parameter is a complex matrix, S11＝a11+b11i denotes the reflectivity of the input signal energy, s21＝a21+b21i denotes the transmission rate of the input signal energy, S11And S21Are all elements of the S parameter, a11And a21Are respectively S11And S21Real part of (a), b11And b21Are respectively S11And S21The imaginary part of (a); calculating according to the formula (1) and the formula (2) to obtain the amplitude and the phase, and further combining the sampling frequency f to obtain the amplitude-frequency response
Figure RE-GDA0003091636330000021
Sum phase frequency response
Figure RE-GDA0003091636330000022
Figure RE-GDA0003091636330000023
Figure RE-GDA0003091636330000024
Then, setting a target index according to the performance index requirement of the microwave filter; the performance index includes a center frequency fcThe bandwidth W and the return loss zeta are obtained through S parameter calculation;
Figure RE-GDA0003091636330000025
the sampling frequency corresponding to the time is the upper and lower cut-off frequency f1、f2Then the center frequency fcComprises the following steps:
fc＝(f1+f2)/2 (3)
the bandwidth is as follows:
W＝f1-f2 (4)
the performance index requirement is such that the actual performance index satisfies the following relationship:
Figure RE-GDA0003091636330000031
|W-W*|≤δW (6)
ζ≤ζ* (7)
wherein f isc *、W*And ζ*Respectively target center frequency, target bandwidth and target return loss,
Figure RE-GDA0003091636330000032
and
Figure RE-GDA0003091636330000033
allowable errors for center frequency and bandwidth, respectively;
finally, the original characteristic R is obtainedraw(ii) a By utilizing the spatial relationship of the three characteristics, namely the corresponding relationship of the amplitude-frequency response and the phase-frequency response at the same frequency, the target index can be visually reflected in the amplitude-frequency response, and the amplitude-frequency response, the phase-frequency response and the target index are combined.
Further, the original data set DrawIs processed into a raw feature RrawTogether with the corresponding adjustable component state x as a training sample, a plurality of training samples form a training set Dtrain。
Furthermore, the feature fusion part comprises four convolution layers, and each convolution layer is sequentially connected with an activation function layer and a pooling layer; the size of the first convolution layer convolution kernel is 5 x 5, the number of channels is 16, and the first convolution layer convolution kernel is used for extracting line and contour information; the last three convolution layers gradually reduce the convolution kernel size and increase the number of channels, namely 32@3 x 3, 64@2 x 2 and 64@2 x 2, respectively, for extracting echo information; the activation function layer adopts a Relu function, the pooling layer adopts maximum pooling, the pooling size is 2 x 2, and the step size is 2 x 2.
Further, the feature mapping portion includes three fully-connected layers, a first fully-connected layer has 128 channels, a second fully-connected layer has 64 channels, and the number of channels of a third fully-connected layer is equal to the dimension of the tunable component state x. Further, the loss function used in training the multi-feature debugging decision model is as follows:
Figure RE-GDA0003091636330000034
wherein N is the number of samples in the training set, m is the dimension of the adjustable component state x of the microwave filter, t represents the error of a specific state in m states, N represents the nth sample, m, N and t are positive integers which are more than or equal to 1,
Figure RE-GDA0003091636330000035
error of the t adjustable component state and the model output for the n sample:
Figure RE-GDA0003091636330000036
wherein
Figure RE-GDA0003091636330000041
Is the output result of the model for the t-th state of the nth sample,
Figure RE-GDA0003091636330000042
is the sample value of the t-th state of the nth sample.
The technical scheme provided by the invention has the beneficial effects that: the method utilizes various characteristics of the microwave filter debugging process, improves the precision of the debugging decision model, further improves the debugging efficiency, and has practicability. The method specifically comprises the following steps:
(1) various data of the debugging process are utilized, including amplitude-frequency response, phase-frequency response and target indexes; the amplitude-frequency response can visually reflect performance indexes, the phase-frequency response can reflect the phase change condition of signals after passing through the filter, the target indexes can provide guidance for debugging, and the three characteristics contain information helpful for debugging.
(2) A debugging process data feature level fusion method is provided. The method performs feature level fusion based on the spatial relationship among the amplitude-frequency response, the phase-frequency response and the target index, and finally obtains fusion features.
(3) A multi-feature debugging decision model is established, so that a high-precision mapping relation can be realized, and the debugging efficiency is improved.
Drawings
The invention will be further described with reference to the accompanying drawings and examples, in which:
FIG. 1 is a flowchart of a method for modeling and debugging multi-feature fusion of a microwave filter according to an embodiment of the present invention;
FIG. 2 is a schematic diagram of an amplitude frequency response and a phase frequency response in an embodiment of the present invention;
FIG. 3 is a graph of raw features in an embodiment of the invention;
FIG. 4 is a diagram of a multi-feature debug decision model in an embodiment of the present invention;
FIG. 5 is an electromagnetic model of a dielectric filter built in HFSS according to an embodiment of the invention;
fig. 6 is a graph showing the center frequency, bandwidth and return loss of four sets of experimental results in the example of the present invention.
Detailed Description
For a more clear understanding of the technical features, objects and effects of the present invention, embodiments of the present invention will now be described in detail with reference to the accompanying drawings.
The embodiment of the invention provides a multi-feature fusion modeling and debugging method for a microwave filter, which establishes a multi-feature debugging decision model and mainly comprises five parts, namely, data acquisition, data preprocessing, feature fusion construction, feature mapping construction and training model construction, and is shown in figure 1.
In the data acquisition phase, an original data set D is constructedraw. And randomly changing the state x of the adjustable part of the microwave filter to measure different S parameters. The adjustable component state x, the acquired S parameter S and the sampling frequency f form a sample at each time, and a plurality of samples form an original data set Draw。
In the data preprocessing stage, the original data set D is usedrawThe S parameter S and the sampling frequency f are calculated to obtain amplitude-frequency response and phase-frequency response, and the original characteristic R is obtained by combining target indexesrawBuilding a training set Dtrain. The process mainly comprises three steps:
first, an amplitude-frequency response and a phase-frequency response are obtained. S parameter ═ S11,s21]Is a complex matrix, s11＝a11+b11i denotes the reflectivity of the input signal energy, s21＝a21+b21i denotes the transmission rate of the input signal energy. Calculating according to the formula (1) and the formula (2) to obtain the amplitude and the phase, and further combining the sampling frequency f to obtainAmplitude frequency response
Figure RE-GDA0003091636330000051
Sum phase frequency response
Figure RE-GDA0003091636330000052
As shown in fig. 2 (a) and (b), fig. 2 (a) is an amplitude-frequency response diagram, and fig. 2 (b) is a phase-frequency response diagram.
Figure RE-GDA0003091636330000053
Figure RE-GDA0003091636330000054
Then, a target index is set according to the microwave filter performance index requirement. The performance index includes a center frequency fcThe bandwidth W and the return loss ζ can be calculated by the S parameter.
Figure RE-GDA0003091636330000055
The sampling frequency corresponding to the time is the upper and lower cut-off frequency f1、f2Then the center frequency is
fc＝(f1+f2)/2, (3)
A bandwidth of
W＝f1-f2. (4)
Return loss ζ of
Figure RE-GDA0003091636330000056
Of the maximum amplitude of all echoes. Typically, the debugging goal is to have the actual performance indicator satisfy the following relationship:
Figure RE-GDA0003091636330000057
|W-W*|≤δW， (6)
ζ≤ζ*. (7)
wherein f isc *、W*And ζ*Respectively target center frequency, target bandwidth and target return loss,
Figure RE-GDA0003091636330000058
and
Figure RE-GDA0003091636330000061
the allowable error of the center frequency and bandwidth, respectively.
Finally, the original characteristic R is obtainedraw(ii) a By utilizing the spatial relationship of the three characteristics, namely the corresponding relationship of the amplitude-frequency response and the phase-frequency response at the same frequency, the target index can be visually reflected in the amplitude-frequency response, and the amplitude-frequency response, the phase-frequency response and the target index are combined. The method selects two indexes of center frequency and return loss, and processes the two indexes, amplitude-frequency response and phase-frequency response into original characteristics R shown in figure 3raw。
Construction of training set Dtrain. The original data set DrawIs processed into a raw feature RrawTogether with the corresponding adjustable component state x as a training sample, a plurality of training samples form a training set Dtrain。
In the stage of constructing the feature fusion part, a neural network is used, wherein the neural network comprises a convolution layer, a pooling layer, an activation function layer and a full connection layer. Constructing a feature fusion part of the model by using the convolution layer, the pooling layer and the activation function layer to convert the original features R into original features RrawProcessed into a fused feature Rff. The characteristic fusion part designed by the invention comprises four convolution layers, and each convolution layer is sequentially connected with an activation function layer and a pooling layer. The size of the first convolution layer convolution kernel is 5 x 5, the number of channels is 16, and the first convolution layer convolution kernel is used for extracting information such as lines, outlines and the like; the last three convolution layers gradually reduce the size of convolution kernels, increase the number of channels, namely 32@3 x 3, 64@2 x 2 and 64@2 x 2 respectively, and are used for extracting information with higher-level semantics such as echoes and the like. The activation function layer employs a Relu function. The pooling layers all adopt maximum pooling, and the size of the pooling is 2 x 2The step size is 2 x 2.
In the feature mapping stage, the feature mapping part of the model is constructed by using the full connection layer, and the mapping from the fusion feature Rff to the adjustable component state x is realized. The feature mapping part designed by the invention comprises three full-connection layers, wherein the first full-connection layer is provided with 128 channels, the second full-connection layer is provided with 64 channels, and the number of the channels of the third full-connection layer is equal to the dimension of the state x of the adjustable component.
In the model training stage, a training set Dtrain training model is used to obtain a high-precision multi-feature debugging decision model. The task to be completed here belongs to the regression task, and therefore, the use of l is considered1Loss function or2A loss function. In practical application, the l1 loss function and the l2 loss function have little difference in regression accuracy, but since the l2 loss function has a square term and is easy to be derived, the l2 loss function is better than the l1 loss function in terms of calculation and convergence, and therefore, the l2 loss function is adopted, as shown in formula (8).
Figure RE-GDA0003091636330000062
Wherein N is the number of samples in the training set, m is the dimension of the state of an adjustable component of the microwave filter, t represents the error of a specific state in m states, N represents the nth sample, m, N and t are positive integers which are more than or equal to 1,
Figure RE-GDA0003091636330000071
error of the t adjustable component state and the model output for the n sample:
Figure RE-GDA0003091636330000072
wherein
Figure RE-GDA0003091636330000073
Is the output result of the model for the t-th state of the nth sample,
Figure RE-GDA0003091636330000074
is the sample value of the t-th state of the nth sample.
After the training hyperparameters are set, the model can be trained until the loss function value is smaller than a set value or the maximum training times are reached.
In summary, as shown in fig. 4, in the multi-feature debugging decision model, in the debugging stage, an S parameter S meeting the index requirement is selected, and is input into the multi-feature debugging decision model after data preprocessing, so as to obtain a corresponding adjustable component state x, and adjust the adjustable component state of the microwave filter to be debugged to x.
3.2 Key points
The key technical points of the invention are as follows:
(1) various data of the debugging process are utilized, including amplitude-frequency response, phase-frequency response and target indexes; the amplitude-frequency response can visually reflect performance indexes, the phase-frequency response can reflect the phase change condition of signals after passing through the filter, the target indexes can provide guidance for debugging, and the three characteristics contain information helpful for debugging.
(2) A debugging process data feature level fusion method is provided. The method performs feature level fusion based on the spatial relationship among the amplitude-frequency response, the phase-frequency response and the target index, and finally obtains fusion features.
(3) A multi-feature debugging decision model is established, so that a high-precision mapping relation can be realized, and the debugging efficiency is improved.
3.3 Effect
A simulation debugging platform is built based on three-dimensional electromagnetic software HFSS and MATLAB. A six-order non-cross-coupled dielectric filter simulation model shown in figure 5 is established in three-dimensional electromagnetic software HFSS, the output response of the simulation model is solved, and parameters, a training model and actual indexes are set in MATLAB. In the experiment, the output response of the dielectric filter is changed by changing the depth of the resonant hole of the dielectric filter. The designed dielectric filter has 6 resonance holes and is of a symmetrical structure, so that x is ═ x1,x2,x3]。
Setting a target center frequency fc *2.610GHz, target bandwidth W*0.193GHz, target Return loss ζ*-18 dB. The allowable error of the center frequency is
Figure RE-GDA0003091636330000075
δW0.005 GHz. The learning rate of the training model was set to 0.006, the decay rate to 0.5, the decay period to 10, and the maximum number of training times to 60.
In order to verify the effectiveness of the method, four groups of comparison experiments are set, each group of experiments is tested for 52 times, and except that the types of the characteristics utilized in the data preprocessing stage are different, other conditions are consistent. Specifically, the experiment one uses amplitude-frequency response, which is one of the existing methods; experiment two utilizes amplitude-frequency response and target index; experiment three utilizes amplitude frequency response and phase frequency response; the experiment four utilizes three characteristics, namely the method provided by the invention. In addition, if three performance indexes simultaneously meet the requirements, debugging is successful, and the ratio of the number of successful debugging times to the total number of debugging times is the debugging success rate. The three performance indexes of the experimental results are shown in graphs (a), (b) and (c) of fig. 6, respectively, graph (a) of fig. 6 is a central frequency graph of the four sets of experimental results, graph (b) of fig. 6 is a bandwidth graph of the four sets of experimental results, graph (c) of fig. 6 is a return loss graph of the four sets of experimental results, and the overall results are shown in table 1.
The experimental results show that the center frequencies of the four groups of experiments are all in the range of 2.6-2.62GHz, the broadband indexes are also all in the range of 0.188-0.198GHz, and the two frequency indexes meet the requirements. However, the return loss cannot meet the requirement, and the times of the return loss meeting the index requirement in the first to fourth experiments are gradually increased. From the aspect of debugging success rate, the method provided by the invention has the highest debugging success rate, which shows that the three characteristics contain information valuable for debugging, can assist each other to jointly guide debugging, and simultaneously proves that the method provided by the invention has feasibility and can improve the accuracy of the model.
TABLE 1 four groups of comparative experiment debugging results
Figure RE-GDA0003091636330000081
The above description is only for the purpose of illustrating the preferred embodiments of the present invention and is not to be construed as limiting the invention, and any modifications, equivalents, improvements and the like that fall within the spirit and principle of the present invention are intended to be included therein.

Claims (6)

1. A multi-feature fusion modeling debugging method for a microwave filter is characterized by comprising the following steps: the method comprises the following steps:
s1: changing the state x of the adjustable component on the microwave filter for a plurality of times, sampling and measuring the S parameter S, and constructing an original data set D containing x, S and a sampling frequency fraw；
S2: for the original data set DrawCarrying out data preprocessing to obtain an original characteristic RrawConstructing a model containing the tunable part state x and the original feature RrawThe training set of (2);
s3: constructing a feature fusion part of a multi-feature debugging decision model by using a convolutional layer, a pooling layer and an activation function layer in a neural network, and processing original features into fusion features;
s4: using a full connection layer to construct a feature mapping part of a multi-feature debugging decision model, wherein the feature mapping part is used for mapping the fusion features to an adjustable component state x;
s5: training the feature fusion part and the feature mapping part by using a training set to obtain a trained high-precision multi-feature debugging decision model;
s6: and preprocessing data of the S parameter S meeting the index requirement and the sampling frequency f, inputting the data into the trained high-precision multi-feature debugging decision model to obtain a corresponding adjustable component state x, and adjusting the adjustable component state of the microwave filter to be debugged to x, thereby realizing high-precision debugging of the microwave filter.
2. The microwave filter multi-feature fusion modeling debugging method of claim 1, characterized in that: in step S2, the original data set D is usedrawS parameter S and sampling frequency f inObtaining original characteristics R by combining amplitude frequency response and phase frequency response and target indexesrawThe process mainly comprises the following three steps:
firstly, obtaining amplitude-frequency response and phase-frequency response; the S parameter is a complex matrix, S11＝a11+b11i denotes the reflectivity of the input signal energy, s21＝a21+b21i denotes the transmission rate of the input signal energy, S11And S21Are all elements of the S parameter, a11And a21Are respectively S11And S21Real part of (a), b11And b21Are respectively S11And S21The imaginary part of (a); calculating according to the formula (1) and the formula (2) to obtain the amplitude and the phase, and further combining the sampling frequency f to obtain the amplitude-frequency response
Figure FDA0002924287620000011
Sum phase frequency response
Figure FDA0002924287620000012
Figure FDA0002924287620000013
Figure FDA0002924287620000014
Then, setting a target index according to the performance index requirement of the microwave filter; the performance index includes a center frequency fcThe bandwidth W and the return loss zeta are obtained through S parameter calculation;
Figure FDA0002924287620000015
the sampling frequency corresponding to the time is the upper and lower cut-off frequency f1、f2Then the center frequency fcComprises the following steps:
fc＝(f1+f2)/2 (3)
the bandwidth is as follows:
W＝f1-f2 (4)
the performance index requirement is such that the actual performance index satisfies the following relationship:
Figure FDA0002924287620000023
|W-W*|≤δW (6)
ζ≤ζ* (7)
wherein f isc *、W*And ζ*Respectively target center frequency, target bandwidth and target return loss,
Figure FDA0002924287620000021
and
Figure FDA0002924287620000022
allowable errors for center frequency and bandwidth, respectively;
finally, the original characteristic R is obtainedraw(ii) a By utilizing the spatial relationship of the three characteristics, namely the corresponding relationship of the amplitude-frequency response and the phase-frequency response at the same frequency, the target index can be visually reflected in the amplitude-frequency response, and the amplitude-frequency response, the phase-frequency response and the target index are combined.
3. The microwave filter multi-feature fusion modeling debugging method of claim 1, characterized in that: in step S2, the original data set D is setrawIs processed into a raw feature RrawTogether with the corresponding adjustable component state x as a training sample, a plurality of training samples form a training set Dtrain。
4. The microwave filter multi-feature fusion modeling debugging method of claim 1, characterized in that: in step S3, the feature fusion part includes four convolution layers, and each convolution layer is sequentially connected to the activation function layer and the pooling layer; the size of the first convolution layer convolution kernel is 5 x 5, the number of channels is 16, and the first convolution layer convolution kernel is used for extracting line and contour information; the last three convolution layers gradually reduce the convolution kernel size and increase the number of channels, namely 32@3 x 3, 64@2 x 2 and 64@2 x 2, respectively, for extracting echo information; the activation function layer adopts a Relu function, the pooling layer adopts maximum pooling, the pooling size is 2 x 2, and the step size is 2 x 2.
5. The microwave filter multi-feature fusion modeling debugging method of claim 1, characterized in that: in step S3, the feature mapping portion includes three full-connected layers, where the first full-connected layer has 128 channels, the second full-connected layer has 64 channels, and the number of channels of the third full-connected layer is equal to the dimension of the adjustable component state x.
6. The microwave filter multi-feature fusion modeling debugging method of claim 1, characterized in that: in step S5, the loss function used for training the multi-feature debugging decision model is:
Figure FDA0002924287620000031
wherein N is the number of samples in the training set, m is the dimension of the adjustable component state x of the microwave filter, t represents a specific state in m states, N represents the nth sample, m, N and t are positive integers which are more than or equal to 1,
Figure FDA0002924287620000032
error of the t adjustable component state and the model output for the n sample:
Figure FDA0002924287620000033
wherein
Figure FDA0002924287620000034
Is the t-th state of the n-th sampleThe result of the model is output as a result,
Figure FDA0002924287620000035
is the sample value of the t-th state of the nth sample.
CN202110128530.4A 2021-01-29 2021-01-29 Multi-feature fusion modeling debugging method for microwave filter Active CN113158541B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202110128530.4A CN113158541B (en) 2021-01-29 2021-01-29 Multi-feature fusion modeling debugging method for microwave filter

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202110128530.4A CN113158541B (en) 2021-01-29 2021-01-29 Multi-feature fusion modeling debugging method for microwave filter

Publications (2)

Publication Number Publication Date
CN113158541A true CN113158541A (en) 2021-07-23
CN113158541B CN113158541B (en) 2022-07-26

Family

ID=76879029

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202110128530.4A Active CN113158541B (en) 2021-01-29 2021-01-29 Multi-feature fusion modeling debugging method for microwave filter

Country Status (1)

Country Link
CN (1) CN113158541B (en)

Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20130271233A1 (en) * 2012-04-16 2013-10-17 Electronics And Telecommunications Research Institue Apparatus and method of tuning microwave filter
CN108170922A (en) * 2017-12-21 2018-06-15 中国地质大学（武汉） A kind of aided debugging method of microwave filter, equipment and storage device
CN108879047A (en) * 2018-07-17 2018-11-23 中国地质大学（武汉） A kind of method for debugging Microwave Cavity Filter, equipment and storage equipment
CN109783905A (en) * 2018-12-28 2019-05-21 中国地质大学（武汉） Microwave Cavity Filter intelligent regulator method based on particle swarm optimization algorithm

Patent Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20130271233A1 (en) * 2012-04-16 2013-10-17 Electronics And Telecommunications Research Institue Apparatus and method of tuning microwave filter
CN108170922A (en) * 2017-12-21 2018-06-15 中国地质大学（武汉） A kind of aided debugging method of microwave filter, equipment and storage device
CN108879047A (en) * 2018-07-17 2018-11-23 中国地质大学（武汉） A kind of method for debugging Microwave Cavity Filter, equipment and storage equipment
CN109783905A (en) * 2018-12-28 2019-05-21 中国地质大学（武汉） Microwave Cavity Filter intelligent regulator method based on particle swarm optimization algorithm

Non-Patent Citations (5)

* Cited by examiner, † Cited by third party
Title
SUN,JJ: "A deep neural network based", 《MICROWAVE AND OPTICAL TECHNOLOGY LETTERS》, vol. 61, no. 9, 11 September 2019 (2019-09-11), pages 2169 - 2173 *
WU,SB;CAO,WH: "Parametric model for microwave filter by using multiple hidden layer output matrix extreme learning machine", 《IET MICROWAVES ANTENNAS & PROPAGATION》, vol. 13, no. 11, 23 September 2019 (2019-09-23), pages 1889 - 1896, XP006082398, DOI: 10.1049/iet-map.2018.5823 *
周金柱等: "基于核机器学习的腔体滤波器辅助调试", 《电子学报》, no. 06, 15 June 2010 (2010-06-15), pages 1274 - 1279 *
张永亮: "一种改进的微波滤波器调试方法", 《强激光与粒子束》 *
张永亮: "一种改进的微波滤波器调试方法", 《强激光与粒子束》, vol. 27, no. 12, 15 December 2015 (2015-12-15), pages 125 - 129 *

Also Published As

Publication number Publication date
CN113158541B (en) 2022-07-26

Similar Documents

Publication Publication Date Title
CN104122540B (en) Radar range profile's target identification method based on coupling dictionary and compressed sensing
CN109783905B (en) Intelligent microwave cavity filter debugging method based on particle swarm optimization algorithm
CN110543860A (en) Mechanical fault diagnosis method and system based on TJM (machine learning model) transfer learning
CN112580588B (en) Intelligent flutter signal identification method based on empirical mode decomposition
CN105787558A (en) Knowledge-based neural network micro-strip filter design method based on ADS
CN113128119B (en) Filter reverse design and optimization method based on deep learning
CN105137373B (en) A kind of denoising method of exponential signal
CN115081316A (en) DC/DC converter fault diagnosis method and system based on improved sparrow search algorithm
CN112989508B (en) Filter optimization design method based on deep learning algorithm
CN108875788A (en) A kind of SVM classifier parameter optimization method based on modified particle swarm optiziation
CN114611443A (en) On-chip filter reverse design method based on equivalent circuit space mapping
CN111191361A (en) Random forest based spiral inductance modeling method
CN109448039B (en) Monocular vision depth estimation method based on deep convolutional neural network
CN113158541B (en) Multi-feature fusion modeling debugging method for microwave filter
CN107659290B (en) Bandwidth extension filter and design method thereof
CN111241700B (en) Intelligent design method of microwave broadband super-surface absorber
CN112669216B (en) Super-resolution reconstruction network of parallel cavity new structure based on federal learning
CN113342734B (en) Plasma density distribution real-time computing device based on microwave reflectometer
CN110112571B (en) Antenna design method based on neural network inverse modeling
CN108491671A (en) A kind of generation method of six node triangle random field unit sample of Physical And Mechanical Indexes of The Typical
Cao et al. Fuzzy Analytic Hierarchy Process Evaluation Method of Gas Turbine Based on Health Degree
Goay et al. Progress in neural network based techniques for signal integrity analysis–a survey
CN110907912A (en) Pole characteristic solving method in target narrow-band mode of resonance region
CN112952315B (en) Medium filter migration modeling debugging method based on mechanism guidance data acquisition
CN109783875A (en) A kind of self iteration update optimization algorithm of marine gravity data

Legal Events

Date Code Title Description
PB01 Publication
PB01 Publication
SE01 Entry into force of request for substantive examination
SE01 Entry into force of request for substantive examination
GR01 Patent grant
GR01 Patent grant