CN-115481591-B - Microwave filter coupling matrix extraction method based on neural network

CN115481591BCN 115481591 BCN115481591 BCN 115481591BCN-115481591-B

Abstract

The invention belongs to the fields of microwave technology and artificial intelligence application, and particularly provides a method for extracting a coupling matrix of a microwave filter based on a neural network, which is used for solving the problems of complex calculation, insufficient accuracy and the like of the traditional Cauchy method, the optimization method and the like; after training, the coupling matrix of the microwave filter can be quickly extracted according to S parameters by using the trained convolutional neural network model, so that the real-time performance is high.

Inventors

ZHANG HAOTIAN
DING SHUAI
FAN DONGDONG

Assignees

电子科技大学

Dates

Publication Date: 20260508
Application Date: 20220921

Claims (7)

1. The microwave filter coupling matrix extraction method based on the neural network is characterized by comprising the following steps of: s1, establishing a forward data set extraction program for calculating S parameters by a microwave filter coupling matrix, wherein the specific process is as follows: S1.1, setting the coupling matrix to be solved as The corresponding microwave filter has the order of The bandwidth is The center frequency is The quality factor is Solving the initial frequency as Solving the termination frequency as Solving the frequency interval as ; S1.2 calculating the order : Number of frequency solutions : ; S1.3. construction Order matrix : , ; S1.4. construction Order matrix : ; S1.5 for each solution frequency Normalization is carried out: , Is that Is a normalized result of (2); S1.6 calculating the solving frequency Impedance matrix at : , wherein, In imaginary units, then the impedance matrix Is the inverse of (1) ; S1.7 setting matrix 、 Is a matrix Is a sub-matrix of: , ; S1.8 calculating the solving frequency Where (a) Is that ; S1.9 calculating the solution frequency Where (a) Is that ; S1.10 calculating the solving frequency Group delay at ; S2, extracting a training data set containing S parameters and a microwave filter coupling matrix through a program; S3, constructing a convolutional neural network model; s4, training the convolutional neural network model by using a training data set; and S5, performing reverse extraction of the microwave filter coupling matrix by using the trained convolutional neural network model.
2. The method for extracting the coupling matrix of the microwave filter based on the neural network as claimed in claim 1, wherein the specific process of the step S2 is as follows: S2.1 according to Design objective of order filter, which is synthesized by generalized Chebyshev method Order ideal coupling matrix ; S2.2 in its ideal coupling matrix A kind of electronic device Within the range, randomly generate A group coupling matrix, wherein, Is a possible detuning range of the filter; s2.3 calculation Using Forward dataset extraction procedure Corresponding to the group coupling matrix 、、 Parameters and couple each group of the matrix 、、 The parameter combination is used as input, the coupling matrix is used as a label vector to form a training sample, and a training data set is further constructed and obtained.
3. The method for extracting a coupling matrix of a microwave filter based on a neural network as claimed in claim 2, wherein the training samples include two kinds of: first training sample is to The real part of (2), Is used to determine the imaginary part of (c), The real part of (2) The imaginary parts of the four channels are respectively used as a channel to form four-channel input data; second training sample will 、 And AND Respectively as a channel to adopt In units, three channels of input data are formed.
4. The method for extracting a coupling matrix of a microwave filter based on a neural network as claimed in claim 1, wherein in the step S3, the convolutional neural network model includes an input layer, a first convolutional layer, a first pooling layer, a second convolutional layer, a second pooling layer, a third pooling layer, a first full-connection layer, a second full-connection layer, a third full-connection layer, and an output layer sequentially connected in order.
5. The method for extracting the coupling matrix of the microwave filter based on the neural network is characterized in that the input layer is four-channel input data or three-channel input data, the first convolution layer and the second convolution layer adopt ELU functions as activation functions, the convolution kernel sizes are 3 multiplied by 1, the first pooling layer, the second pooling layer and the third pooling layer adopt maximum pooling, the filter size is 2 multiplied by 1, the number of neurons of the first full-connection layer is 1024, the number of neurons of the second full-connection layer is 256, the number of neurons of the third full-connection layer is 64, the ELU functions are adopted as activation functions, the output layer is a coupling matrix, and the output layer is output in a one-dimensional vector mode.
6. The method for extracting the coupling matrix of the microwave filter based on the neural network as claimed in claim 1, wherein the specific training process in the step S4 is as follows: s4.1, taking input data in the training sample as input of a neural network model, and outputting a prediction vector by the neural network model; s4.2, setting SmoothL a1 loss function as a loss function of the neural network: ; Wherein, the An average value of absolute values of element differences between the prediction vector and the label vector; and S4.3, setting a neural network training termination target to finish model training, namely stopping training when the training round number reaches a preset threshold value or the loss function reaches the preset threshold value, repeating the step S2 if the training round number reaches the preset threshold value and the loss function still does not reach the preset threshold value, collecting more sample data, and retraining until the loss function reaches the preset threshold value.
7. The method for extracting the coupling matrix of the microwave filter based on the neural network as claimed in claim 1, wherein the specific process of the step S5 is that S parameters and S parameters of the microwave filter to be extracted are obtained through testing or simulation And parameters and input data forming a neural network model, and outputting a corresponding coupling matrix by the trained neural network model.

Description

Microwave filter coupling matrix extraction method based on neural network Technical Field The invention belongs to the fields of microwave technology and artificial intelligence application, in particular relates to a microwave filter coupling matrix extraction method, and particularly provides a neural network-based microwave filter coupling matrix extraction method which can be used for designing and tuning a microwave filter. Background The development of modern wireless communication technology has higher and higher requirements on the design and debugging efficiency of a microwave filter, and a comprehensive method of a complete system is already existing in the fields of filter synthesis and parameter extraction, but the comprehensive filter initial full-wave simulation result often shows poor frequency response performance and unpredictability, continuous optimization is needed, long time is needed to be consumed for obtaining a desired result, and in the optimization design process, the coupling matrix of the filter is extracted by utilizing the S parameter, and the next tuning direction can be determined according to the difference value between the extracted coupling matrix and the designed ideal coupling matrix. In addition, aiming at the designed filter real object, when in tuning, the tuning direction of the filter real object can be defined by comparing the difference value between the coupling matrix extracted by the measured S parameter and the ideal coupling matrix, thereby providing assistance for manual tuning and full-automatic tuning. In the aspect of filter coupling matrix extraction, the conventional methods mainly comprise a cauchy method and an optimization method, wherein the cauchy method and the optimization method have the advantages and disadvantages of high extraction speed Ke Xifa but lower out-of-band precision, and the optimization method has higher precision but low extraction speed and needs a lot of time for optimization. In contrast, the neural network technology has the characteristics of high extraction speed and relatively consistent overall regression accuracy, so that the neural network is applied to filter coupling matrix extraction, and the problems of increasing complexity, high time consumption and the like in the field are expected to be solved through learning of a large amount of data with the aid of the computer technology. Disclosure of Invention The invention aims to provide a microwave filter coupling matrix extraction method based on a neural network, which is used for solving the problems of complex calculation and insufficient accuracy of the existing Cauchy method, optimization method and other methods. In order to achieve the above purpose, the invention adopts the following technical scheme: the microwave filter coupling matrix extraction method based on the neural network is characterized by comprising the following steps of: S1, establishing a forward data set extraction program for calculating S parameters by a microwave filter coupling matrix; s2, extracting a training data set containing S parameters and a microwave filter coupling matrix through a program; S3, constructing a convolutional neural network model; s4, training the convolutional neural network model by using a training data set; and S5, performing reverse extraction of the microwave filter coupling matrix by using the trained convolutional neural network model. Further, the specific process of step S1 is as follows: s1.1, setting a coupling matrix to be solved as M, wherein the coupling matrix corresponds to a microwave filter, and has the steps of N, the bandwidth of BW, the center frequency of F 0, the quality factor of Q, the solving starting frequency of F s, the solving ending frequency of F e and the solving frequency interval of Step; S1.2, calculating to obtain an order N 1：N1 =N+2, and solving the frequency quantity P: s1.3, constructing an N 1 -order matrix R: S1.4, constructing an N 1 -order matrix U: S1.5, normalization is carried out for each solving frequency F i: f i' is the normalized result of F i; S1.6, calculating an impedance matrix Z i at a solution frequency F i: Wherein j is an imaginary unit, and the inverse of the impedance matrix Z i is Z i-1; s1.7, setting a matrix A, B as a submatrix of a matrix Z i-1: s1.8, calculating and solving S 11 at the frequency F i to be S 11(Fi)＝1+2j[Zi-1]1,1; s1.9, calculating and solving S 21 at the frequency F i to be S 21(Fi)＝-2j[Zi-1]N1,1; S1.10 calculating group delay at solving frequency F i Further, the specific process of step S2 is as follows: s2.1, according to the design target of the N-order filter, synthesizing an N 1 -order ideal coupling matrix M ideal by using a generalized Chebyshev method; S2.2, randomly generating N sample groups of coupling matrixes within a range of +/-error of an ideal coupling matrix M ideal, wherein error is a possible detuning range of the filter; S2.3, calculating S 11、S21 and TD