Tuning network matching method and device based on neural network initial value guidance

Abstract

This invention discloses a tuning network matching method and apparatus based on neural network initial value guidance. The method includes connecting different standard loads to the output of the tuning network, calculating the impedance of the antenna feeder in each frequency band and the parameter values ​​of components in the tuning network, and constructing a dataset; training the neural network using the dataset to learn the mapping relationship between impedance and component parameter values, resulting in a trained neural network; using the trained neural network to obtain predicted parameter values ​​of the components based on the actual impedance of the antenna feeder in the target frequency band; determining the parameter search range of the optimization algorithm based on the predicted parameter values ​​of the components, and controlling the motor to rotate to the motor angle corresponding to the parameter search range according to a pre-stored mapping table; and performing iterative optimization with minimizing the standing wave ratio as the objective function to obtain the optimal parameter values ​​of the components.

Description

Technical Field

[0001] This invention relates to the field of radio frequency communication technology, and specifically to a tuning network matching method and apparatus based on neural network initial value guidance. Background Technology

[0002] In communications, the tuning network is a key component for impedance matching between the transmitter and the antenna feeder; its tuning speed and matching accuracy directly affect communication quality. Traditional tuning methods mainly rely on a global search of the multidimensional solution space formed by capacitor C and inductor L, but this method has the following problems:

[0003] (1) Low efficiency of multidimensional search: The multidimensional solution space composed of multiple capacitors C and inductors L has a high dimension, and traditional traversal search needs to cover a large number of parameter combinations. Since the capacitors and inductors are adjusted by motor drive, each parameter adjustment involves the rotation of the motor, which not only consumes a lot of time (a single tuning often exceeds 30 seconds), but also aggravates hardware wear and reduces the service life of the equipment due to frequent mechanical movement.

[0004] (2) Initial value blindness: In the absence of effective initial values, optimization algorithms such as genetic algorithms need to set a large population size and number of iterations to cover the possible solution space, resulting in a large degree of blindness in the optimization process. This further increases the tuning time cost, and frequent parameter adjustments will exacerbate the wear and tear on hardware such as motors, affecting the stability of the equipment.

[0005] (3) Insufficient matching accuracy and adaptability: Traditional tuning methods do not take into account the dynamic changes of load impedance. When faced with complex or sudden load conditions, they are prone to problems such as insufficient matching accuracy (high standing wave ratio) and low success rate (usually less than 90%), which makes it difficult to meet the needs of fast tuning in communication.

[0006] In related technologies, the paper "Design of Automatic Impedance Matching System Based on BP Neural Network, Master's Thesis of Tianjin University of Technology" proposes an automatic impedance matching algorithm based on fuzzy neural network. This algorithm directly utilizes the neural network for optimization rather than using it to learn and turn the tuning network into a black box to generate initial values ​​for hardware optimization. Patent application CN118586356A, through the design of the matching circuit of the optimized RF system, employs an analog search algorithm to automatically optimize the RF circuit matching; however, the reference in this scheme... The initial value is manually set, and the effect of the set initial value is not ideal. Moreover, the entire matching process is implemented in software without hardware intervention. However, the patent application document with publication number CN105846834A proposes that different frequency bands no longer correspond to different hardware filtering paths, but to pre-stored control parameters. When the wireless communication device receives radio frequency signals of different frequency bands, it only needs to change the control parameters of the filtering path, and multiple frequency band radio frequency signals can be filtered through a single filtering path. However, this solution directly uses the pre-stored control parameters to achieve the filtering of radio frequency signals, without mentioning how the control parameters are established.

[0007] Therefore, there is an urgent need for a fast matching method that can improve tuning efficiency, reduce hardware losses, and enhance matching accuracy in order to overcome the bottlenecks of traditional tuning technology. Summary of the Invention

[0008] The technical problem to be solved by this invention is how to improve antenna tuning efficiency and achieve rapid impedance matching.

[0009] The present invention solves the above-mentioned technical problems through the following technical means: A tuning network matching method based on neural network initial value guidance is proposed, the method comprising: Different standard loads are connected to the output of the tuning network, the impedance of the antenna feeder in each frequency band and the parameter values ​​of the components in the tuning network are calculated, and a dataset is constructed. The neural network is trained using a dataset to learn the mapping relationship between impedance and the parameter values ​​of the components, resulting in a well-trained neural network. The trained neural network is used to obtain the predicted values ​​of the component parameters based on the actual impedance of the antenna feeder in the target frequency band. The parameter search range of the optimization algorithm is determined based on the predicted parameter values ​​of the components. The motor is controlled to rotate to the motor angle corresponding to the parameter search range according to the pre-stored mapping table. The optimal parameter values ​​of the components are obtained by iterative optimization with the minimization of standing wave value as the objective function.

[0010] Furthermore, the components in the tuning network include inductors and capacitors. The process of connecting different standard loads to the output of the tuning network, calculating the impedance of the antenna feeder in each frequency band and the parameter values ​​of the components in the tuning network, and constructing a dataset includes: Different standard loads are connected to the output of the tuning network, and the parameter value combinations of different components are iterated. The impedance data of different component parameter combinations under different standard loads were measured sequentially in each frequency band using a vector network analyzer. A system of equations was constructed using impedance data corresponding to a portion of standard loads to solve for the theoretical values ​​of parameters corresponding to different components. The theoretical values ​​of the parameters are verified using impedance data corresponding to another set of standard loads, and the dataset is constructed based on the verified theoretical values ​​of the parameters and the corresponding impedances.

[0011] Furthermore, the standard load includes short-circuit load, XΩ, YΩ, ZΩ, and open-circuit load; Where X, Y, and Z are preset ohm values, and X, Y, and Z represent non-zero and non-infinite values.

[0012] Furthermore, the step of training the neural network using the dataset to learn the mapping relationship between impedance and the parameter values ​​of the components, thereby obtaining a trained neural network, includes: The neural network is trained using a portion of the sample data in the dataset to learn the mapping relationship between impedance and the parameter values ​​of the components, and to determine the basic network parameters. The parameters of the neural network are fine-tuned using another portion of the sample data in the dataset to obtain a trained neural network.

[0013] Furthermore, the step of using a trained neural network to obtain the predicted parameter values ​​of the component based on the actual impedance of the antenna feeder in the target frequency band includes: The actual impedance of the antenna feeder in the target frequency band was measured from the output of the tuned network using a vector network analyzer. The actual impedance of the antenna feeder in the target frequency band is input into the trained neural network to obtain the predicted parameter values ​​of the component.

[0014] Further, the step of determining the parameter search range of the optimization algorithm based on the predicted parameter values ​​of the component, controlling the motor to rotate to the motor angle corresponding to the parameter search range according to the pre-stored mapping table, and performing iterative optimization to obtain the optimal parameter values ​​of the component with minimizing the standing wave value as the objective function includes: The parameter search range of the optimization algorithm is determined based on the predicted parameter values ​​of the components:

[0015] in, These are the predicted values ​​for the component's parameters. This represents the actual limit of the component's parameters. , , Indicates product; The corresponding motor angle is found from the mapping table based on the parameter search range, and the motor is controlled to rotate to that motor angle. The motor is connected to the component. The standing wave ratio (SWR) is measured at the input of the tuned network using a vector network analyzer, and the optimal parameter values ​​of the component are obtained by iterative optimization with the minimum SWR as the objective function.

[0016] Furthermore, the optimization algorithm employs a genetic algorithm, an ant colony algorithm, or a particle swarm optimization algorithm.

[0017] Furthermore, the process of constructing the mapping table includes: Establish the initial mapping relationship between the motor rotation angle and the component values ​​in the tuning network; The standard load is connected to the output of the tuned network theoretical model, and the initial mapping table is traversed to correct the initial mapping relationship between the motor rotation angle and the component value, thus obtaining the mapping table.

[0018] Furthermore, establishing the initial mapping relationship between the motor rotation angle and the component values ​​in the tuning network includes: Control the motor connected to the capacitor to rotate to different angles, and measure the actual capacitance value of the capacitor when the motor rotates to different angles; Linear interpolation was used to fit the discrete measured capacitance values ​​to obtain the mapping relationship between the capacitance value and the motor rotation angle. A three-dimensional electromagnetic model of the inductor L and its surrounding cavity was established, and the three-dimensional battery model was simulated to obtain the S matrix corresponding to different motor rotation angles. The mapping relationship between inductance value and motor rotation angle is obtained by inverting the S-matrix; By combining the mapping relationship between capacitor value and motor rotation angle and the mapping relationship between inductance value and motor rotation angle, an initial mapping relationship between motor rotation angle and component values ​​in the tuning network is formed.

[0019] Furthermore, the step of obtaining the mapping relationship between the inductance value and the motor rotation angle based on the S-matrix inversion includes: The input impedance of the inductor is derived by inverting the S-matrix; The inductance value is calculated based on the imaginary part of the input impedance, expressed by the following formula:

[0020] In the formula, This represents the imaginary part of the input impedance. For operating frequency, Pi Indicates product; Based on the calculated inductance value, the mapping relationship between the inductance value and the motor rotation angle is obtained.

[0021] Furthermore, a standard load is connected to the output of the tuned network theoretical model, and the initial mapping table is traversed to correct the initial mapping relationship between the motor rotation angle and the component values, resulting in the mapping table, which includes: Different standard loads are connected to the output of the tuned network theoretical model, and different component values ​​in the initial mapping relationship are traversed. The theoretical output impedance corresponding to the current traversed component value is calculated. The impedance value of the standard load is used as the actual output impedance, and the mean absolute error between the actual output impedance and the theoretical output impedance is calculated. The value of the currently traversed component is corrected based on the mean absolute error, resulting in a corrected mapping table.

[0022] Furthermore, the correction of the currently traversed component values ​​based on the mean absolute error is used to obtain a corrected mapping table, expressed by the formula:

[0023]

[0024] In the formula, This is the corrected capacitance value. This is the corrected inductance value. and These are the capacitance and inductance values ​​currently being iterated. This represents the mean absolute error.

[0025] Furthermore, this invention also proposes a tuning network matching device based on neural network initial value guidance, the device comprising: The dataset construction module is used to connect different standard loads to the output of the tuning network, calculate the impedance of the antenna feeder in each frequency band and the parameter values ​​of the components in the tuning network, and construct the dataset. The training module is used to train the neural network using the dataset to learn the mapping relationship between impedance and the parameter values ​​of the components, and to obtain a trained neural network. The prediction module is used to obtain the predicted values ​​of the component parameters based on the actual impedance of the antenna feeder in the target frequency band using a trained neural network. The optimization module is used to determine the parameter search range of the optimization algorithm based on the predicted parameter values ​​of the components, and to control the motor to rotate to the motor angle corresponding to the parameter search range according to the pre-stored mapping table. The optimal parameter values ​​of the components are obtained by iterative optimization with minimizing the standing wave value as the objective function.

[0026] Furthermore, the present invention also proposes a computer-readable storage medium storing a computer program thereon, which, when executed by a processor, implements the above-described tuning network matching method based on neural network initial value guidance.

[0027] The advantages of this invention are: (1) This invention addresses the problems of low efficiency in multidimensional search and blindness in initial values ​​during the traditional tuning process of tuning networks. It achieves rapid matching through three stages: data acquisition, neural network training, and tuning. First, different standards are connected to the output of the tuning network to calculate the impedance of the antenna feeder in each frequency band and the parameter values ​​of the components in the tuning network to construct a dataset. The dataset is used to train the neural network to learn these data and establish the mapping relationship between impedance and capacitance and inductance. During actual tuning, the actual output impedance is input into the neural network for optimization to obtain the parameter prediction value of the component. This parameter prediction value is used as the initial value of the optimization algorithm to reduce the search space and guide the algorithm to optimize the parameter value of the component in the reduced search space. The reduction of the search space reduces the number and amplitude of motor rotations. Combined with the accuracy of the initial value, it reduces invalid mechanical motion. It not only meets the requirements of rapid tuning but also significantly reduces the hardware wear rate. Therefore, this invention significantly improves the tuning speed, reduces hardware wear, and has a high matching success rate. It is suitable for rapid tuning and table building of tuning networks.

[0028] (2) This invention ensures the validity of training data through a physical equation verification mechanism, which solves the credibility problem of neural network black box models and improves matching accuracy and success rate.

[0029] (3) When constructing the mapping relationship between motor rotation angle and component parameter values, this invention addresses the problem that the inductance value of inductor L is difficult to measure directly due to environmental influences. It establishes an accurate mapping relationship between motor rotation angle and component parameter values ​​through "electromagnetic simulation-S-parameter inversion", solves the problem of measuring inductance value caused by environmental interference, thereby improving the accuracy of tuning parameter calculation and improving matching accuracy.

[0030] (4) The present invention uses test data of standard load to dynamically correct the mapping relationship between the established motor rotation angle and the parameter value of the component, which can reduce the error of the tuning network model and thus improve the matching accuracy.

[0031] Additional aspects and advantages of the invention will be set forth in part in the description which follows, and in part will be obvious from the description, or may be learned by practice of the invention. Attached Figure Description

[0032] The accompanying drawings, which form part of this specification, are used to provide a further understanding of the invention. The illustrative embodiments of the invention and their descriptions are used to explain the invention and do not constitute an undue limitation thereof. Figure 1 This is a flowchart illustrating a tuning network matching method based on neural network initial value guidance proposed in an embodiment of the present invention. Figure 2This is a schematic diagram illustrating the principle of a tuning network matching method based on neural network initial value guidance in one embodiment of the present invention; Figure 3 This is a schematic diagram of the tuning network and components corresponding to the motor in one embodiment of the present invention; Figure 4 This is a schematic diagram of the structure of a tuning network matching device based on neural network initial value guidance according to an embodiment of the present invention. Detailed Implementation

[0033] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below in conjunction with the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0034] like Figures 1 to 2 As shown, the first embodiment of the present invention proposes a tuning network matching method based on neural network initial value guidance, the method comprising the following steps: S10. Connect different standard loads to the output of the tuning network, calculate the impedance of the antenna feeder in each frequency band and the parameter values ​​of the components in the tuning network, and construct a dataset. S20. Use the dataset to train the neural network to learn the mapping relationship between impedance and the parameter values ​​of the components, and obtain a trained neural network. S30. Using the trained neural network, the predicted values ​​of the component parameters are obtained based on the actual impedance of the antenna feeder in the target frequency band. S40. Determine the parameter search range of the optimization algorithm based on the predicted parameter values ​​of the components, and control the motor to rotate to the motor angle corresponding to the parameter search range according to the pre-stored mapping table. Iteratively optimize the component to obtain the optimal parameter value by minimizing the standing wave value as the objective function.

[0035] It should be noted that in this embodiment, different standards are connected to the output of the tuning network to calculate the impedance of the antenna feeder in each frequency band and the parameter values ​​of the components in the tuning network to construct a dataset. The dataset is used to train the neural network so that the neural network learns from this data and establishes a mapping relationship between impedance and capacitance and inductance. During actual tuning, the actual output impedance is input into the neural network for optimization to obtain the predicted parameter values ​​of the components. These predicted parameter values ​​are used as the initial values ​​of the optimization algorithm to narrow the search space. The algorithm is then guided to optimize the parameter values ​​of the components within the narrowed search space. The reduction in the search space reduces the number and amplitude of motor rotations. Combined with the accuracy of the initial values, this reduces invalid mechanical motion, which not only meets the requirements for fast and accurate tuning but also significantly reduces the hardware wear rate. Therefore, this invention significantly improves the tuning speed, reduces hardware wear, and has a high matching success rate, making it suitable for rapid tuning and table construction of tuning networks.

[0036] As a further preferred technical solution, step S10: connecting different standard loads to the output of the tuning network, calculating the impedance of the antenna feeder in each frequency band and the parameter values ​​of the components in the tuning network, and constructing a dataset, specifically includes the following steps: S11. Connect different standard loads to the output of the tuning network respectively, and iterate through the parameter value combinations of different components. It should be noted that the tuning network in this embodiment includes inductors and capacitors. The capacitor C has X capacitance values, with uniform spacing between adjacent values; the inductor L has X inductance values, with uniform spacing between adjacent values. Combining these parameters, X is obtained. n A combination of capacitors and inductors (the number of capacitors and inductors in the tuning network is n).

[0037] Furthermore, the tuning network includes, but is not limited to, T-type tuning networks, π-type tuning networks, L-type tuning networks, and other types of tuning networks.

[0038] In this embodiment, five standard loads are sequentially connected to the output of the antenna tuning network: 0Ω, 50Ω, 100Ω, 200Ω, and an open circuit (equivalent to infinite impedance). A vector network analyzer is used to sequentially measure X within the target shortwave frequency band. n S-matrix of various combinations under 5 loads (including , , , (Parameters such as capacitance and inductance are recorded), and the correlation data between the load impedance and S-parameters for each capacitor-inductor combination is recorded. By using the correlation data, the inductance and capacitance values ​​are treated as a black box, and an approximate function for the black box is obtained through a neural network.

[0039] S12. Use a vector network analyzer to measure the impedance data of different component parameter combinations under different standard loads in each frequency band. S13. Use the impedance data corresponding to a portion of standard loads to construct a system of equations to solve for the theoretical values ​​of the parameters corresponding to different components; Specifically, this embodiment consists of two capacitors and one inductor. Taking a type network as an example, a ternary equation system is constructed using impedance data corresponding to three loads: 0Ω, 50Ω, and 100Ω (because these three sets of data are on the same order of magnitude as the actual impedances): eq1: Z_L1 = f (C,L) (0Ω load); eq2: Z_L2 = f (C,L) (50Ω load); eq3: Z_L3 = f (C,L) (100Ω load). The theoretical values ​​of two capacitors C and one inductor L are solved by the equation system, where Z_L1, Z_L2, and Z_L3 are the output impedances corresponding to different standard loads, and f (C,L) represents the constructed function, which can be obtained by fitting a neural network.

[0040] S14. Verify the theoretical values ​​of the parameters using the impedance data corresponding to another part of the standard load, and construct the dataset based on the verified theoretical values ​​of the parameters and the corresponding impedances.

[0041] Specifically, this embodiment uses measurement data of a 200Ω load and an open-circuit load to verify the rationality of the solution (any two sets of measurement data can be used to verify the rationality of the solution). The error between the theoretical prediction value and the actual measurement value is calculated: error = |Z_predicted - Z_measured|, where Z_predicted represents the predicted output impedance value and Z_measured represents the actual load impedance value used. If the error is <5%, the data is considered valid and the sample is retained; otherwise, invalid data is discarded to ensure the physical rationality of the training set.

[0042] As a further preferred technical solution, step S20: training the neural network using the dataset to learn the mapping relationship between impedance and component parameter values, thereby obtaining a trained neural network, specifically includes the following steps: S21. Use partial sample data from the dataset to train the neural network to learn the mapping relationship between impedance and component parameter values, and determine the basic network parameters. S22. Use another part of the sample data in the dataset to fine-tune the parameters of the neural network to obtain a trained neural network.

[0043] It should be noted that this embodiment employs a transfer learning strategy. First, the neural network is trained using effective data with a 200Ω load to determine the basic network parameters. Then, effective data with 0Ω, 50Ω, 100Ω, and open-circuit loads are used to fine-tune the network, improving its adaptability to different load scenarios. The data used for neural network training and the data used for neural network fine-tuning can be arbitrarily selected; this embodiment does not impose specific limitations.

[0044] It should be noted that this embodiment ensures the validity of training data through a physical equation verification mechanism, solving the credibility problem of neural network black-box models and improving matching accuracy and success rate. Experimental data show that the matching success rate is increased from 88% in traditional methods to 99.5%, and the initial value accuracy can reach VSWR < 2, laying the foundation for the rapid convergence of the optimization algorithm.

[0045] Furthermore, by employing transfer learning strategies, the neural network can adapt to different load scenarios. Combined with the global optimization capabilities of the optimization algorithm, it can cope with complex and ever-changing antenna and feeder impedance environments, making it widely applicable.

[0046] As a further preferred technical solution, step S30: using a trained neural network to obtain the predicted parameter values ​​of the component based on the actual impedance of the antenna feeder in the target frequency band, specifically includes the following steps: S31. Use a vector network analyzer to measure the actual impedance of the antenna feeder in the target frequency band from the output of the tuning network. S32. Input the actual impedance of the antenna feeder in the target frequency band into the trained neural network to obtain the predicted parameter values ​​of the component.

[0047] It should be noted that in this embodiment, a vector network analyzer is used to measure the actual impedance (including real, imaginary, and frequency information) of the antenna feeder at the target frequency band from the output of the tuning network, and this measurement is then input into the trained neural network. Based on the learned mapping relationship, the neural network outputs predicted values ​​of the capacitor C and inductor L as the initial values ​​for the optimization algorithm.

[0048] As a further preferred technical solution, step S40: determining the parameter search range of the optimization algorithm based on the predicted parameter values ​​of the component, and controlling the motor to rotate to the motor angle corresponding to the parameter search range according to the pre-stored mapping table, and performing iterative optimization with minimizing the standing wave value as the objective function to obtain the optimal parameter value of the component, specifically includes the following steps: S41. Based on the predicted parameter values ​​of the components, the parameter search range of the optimization algorithm is determined as follows:

[0049] in, These are the predicted values ​​for the component's parameters. This represents the actual limit of the component's parameters. , , Indicates product; It should be noted that, in this embodiment, by using the parameter prediction value output by the neural network as a benchmark, the multidimensional search space of the optimization algorithm is compressed from [the actual lower limit of the component, the actual upper limit of the component] to [initial value × A (A=[0,1], min(the actual upper limit of the component, the initial value × B (B=[1, positive infinity))], which greatly reduces the search range. In the test, the total tuning time was reduced from 70 minutes in the traditional method to 10 minutes, meeting the requirements for fast tuning.

[0050] S42. Based on the parameter search range, find the corresponding motor angle from the mapping table, and control the motor to rotate to that motor angle. The motor is connected to the component. Specifically, in this embodiment, the values ​​of capacitor C and inductor L are converted into the corresponding rotation angle of the motor by looking up a pre-calibrated "angle-parameter value" mapping table, and the motor is controlled by computer to rotate to the position corresponding to the initial value.

[0051] S43. Measure the standing wave value from the input of the tuned network using a vector network analyzer, and perform iterative optimization with minimizing the standing wave value as the objective function to obtain the optimal parameter values ​​of the component.

[0052] It should be noted that the standing wave ratio (VSWR) is measured at the input of the tuned network using a vector network analyzer. With minimizing VSWR as the objective function, the optimization algorithm iteratively optimizes the capacitor and inductor values ​​within the search space until VSWR < S (achieving a good matching standard), thus obtaining the optimal parameter values ​​of the components.

[0053] Furthermore, the optimization algorithm used in this embodiment includes, but is not limited to, genetic algorithms, ant colony algorithms, or particle swarm optimization algorithms. This embodiment does not specifically limit the optimization algorithm used.

[0054] As a further preferred technical solution, before step S10: connecting different standard loads to the output of the tuning network, calculating the impedance of the antenna feeder in each frequency band and the parameter values ​​of the components in the tuning network, and constructing a dataset, the method further includes pre-calibrating an angle-parameter value mapping table, specifically including: S1. Establish the initial mapping relationship between the motor rotation angle and the component values ​​in the tuning network; S2. Connect the standard load to the output of the tuned network theoretical model and traverse the initial mapping table to correct the initial mapping relationship between the motor rotation angle and the component value, and obtain the mapping table.

[0055] As a further preferred technical solution, step S1: establishing the initial mapping relationship between the motor rotation angle and the component values ​​in the tuning network, specifically includes the following steps: S11' Control the motor connected to the capacitor to rotate to different angles, and measure the actual capacitance value of the capacitor when the motor rotates to different angles; S12': Linear interpolation is used to fit the discrete measured capacitance values ​​to obtain the mapping relationship between the capacitance value and the motor rotation angle; Specifically, the capacitance value of capacitor C can be directly measured by controlling the rotation of the motor connected to the capacitor to different angles. Then, the capacitance value corresponding to capacitor C was measured using a capacitance meter; linear interpolation was used to fit the discrete measured data to obtain the functional mapping relationship between the capacitance value and the motor rotation angle.

[0056] in, , The fitting coefficients for capacitor C are calculated using the least squares method (fitting error < 2%).

[0057] It should be noted that, as Figure 3 As shown, the tuning network can be equipped with multiple capacitors, each of which is connected to a corresponding motor, and a mapping relationship between the capacitance value and the corresponding motor rotation angle is established.

[0058] S13' Establish a three-dimensional electromagnetic model of the inductor L and its surrounding cavity, and simulate the three-dimensional battery model to obtain the S matrix corresponding to different motor rotation angles; It should be noted that this embodiment uses electromagnetic simulation software such as HFSS or CST to establish a three-dimensional electromagnetic model of the inductor L and its surrounding cavity (including factors such as the metal support, grounding structure, and inductor short-circuit bar) to simulate different motor rotation angles. The corresponding S matrix ( , , , ).

[0059] S14': The mapping relationship between inductance value and motor rotation angle is obtained by inverting the S matrix; S15': Combining the mapping relationship between capacitor value and motor rotation angle and the mapping relationship between inductance value and motor rotation angle, an initial mapping relationship is formed between motor rotation angle and component values ​​in the tuning network.

[0060] It should be noted that the inductance value of L in the tuning network is significantly affected by environmental factors such as eddy currents in the cavity and electromagnetic coupling of adjacent metal components, and the direct measurement error often exceeds 20%. This embodiment, to improve the accuracy of inductance measurement, indirectly establishes a mapping relationship between the motor rotation angle and the inductance value through "electromagnetic simulation-S-parameter inversion," solving the measurement problem caused by environmental interference and reducing the inductance value error from >20% to <5%. This avoids inaccurate mapping between the inductance value and the motor rotation angle, which could lead to deviations in the tuning parameter calculation and affect the matching accuracy.

[0061] As a further preferred technical solution, step S14': obtaining the mapping relationship between inductance value and motor rotation angle based on S matrix inversion, specifically includes the following steps: S141', The input impedance of the inductor is derived from the S-matrix; Specifically, the input impedance Z of the inductor L is derived from the S matrix. 11 The formula is as follows:

[0062] In the formula, The reflection coefficient of port 1, For reverse transmission (gain or loss). Forward transmission (gain or loss). is the reflection coefficient of port 2.

[0063] S142' Calculate the inductance value based on the imaginary part of the input impedance. The formula is as follows:

[0064] In the formula, This represents the imaginary part of the input impedance. For operating frequency, Pi Indicates product; S143'. Based on the calculated inductance value, obtain the relationship between the inductance value L and the motor rotation angle. A mapping table between them.

[0065] It should be noted that, as Figure 3 As shown, the tuning network can be configured with multiple inductors, each of which is connected to a corresponding motor, and a mapping relationship between the inductance value and the corresponding motor rotation angle is established.

[0066] Furthermore, the tuning network includes, but is not limited to, T-type tuning networks, π-type tuning networks, double L-type tuning networks, and other types of tuning networks.

[0067] As a further preferred technical solution, step S2: connecting the standard load to the output of the tuning network theoretical model and traversing the initial mapping relationship table to correct the initial mapping relationship between the motor rotation angle and the component value, thereby obtaining the mapping relationship table, specifically includes the following steps: S21' Connect different standard loads to the output of the tuned network theoretical model, traverse different component values ​​in the initial mapping relationship, and calculate the theoretical output impedance corresponding to the current traversed component value; Specifically, in this embodiment, five standard loads are connected sequentially at the output of the tuning network: short circuit (equivalent to 0Ω impedance), XΩ, YΩ, ZΩ, and open circuit (equivalent to infinite impedance), where X, Y, and Z are preset ohmic values ​​representing non-zero and non-infinite values; the capacitance values ​​C are iterated through. Xc The capacitance range and the inductance L ( ) Xl (sensory value level) Xc × Xl ( ) combinations, using a vector network analyzer to measure the S-matrix of each combination in the target frequency band, and then based on the tuned network theoretical model, using the known input impedance Calculate the theoretical output impedance using the component parameter values ​​(C, L). :

[0068] In the formula, The calculation function representing the circuit model used is based on the characteristics of the circuit model itself. and There is a corresponding function.

[0069] S22'. Use the impedance value of the standard load as the actual output impedance, and calculate the average absolute error between the actual output impedance and the theoretical output impedance. Specifically, in this embodiment, the impedance value of the standard load is used as the actual output impedance. The mean absolute error between the actual output impedance and the theoretical output impedance is calculated as follows:

[0070] In the formula, This represents the mean absolute error.

[0071] S23' Correct the values ​​of the currently traversed components based on the mean absolute error to obtain the corrected mapping table.

[0072] Specifically, the parameter values ​​of the currently traversed components are corrected based on the mean absolute error, as expressed by the formula:

[0073]

[0074] In the formula, This is the corrected capacitance value. This is the corrected inductance value. and These are the capacitance and inductance values ​​currently being iterated.

[0075] It should be noted that, in order to reduce model mismatch error, this embodiment uses a dynamic correction mechanism for the initial mapping relationship based on test data of five standard loads, which reduces the tuning network model error from 12.5% ​​to 3.8%, thereby improving the matching accuracy.

[0076] A specific application example of the antenna dual-stage optimized tuning method proposed in this embodiment is as follows: (1) Experimental equipment Shortwave antenna RF system, operating frequency band Fmin-Fmax; Tuning network: π-type network, including C1 (80pF~900pF, 5 levels), C2 (80pF~900pF, 5 levels), and L (0μH~3.5μH, 5 levels). Each component is driven by a stepper motor (resolution 0.1°). Vector network analyzer: Measurement frequencies include the operating frequency band, and it supports real-time measurement of the S-matrix; Standard loads: 0Ω, 50Ω, 100Ω, 200Ω (accuracy ±1%), open circuit load (insulation resistance > 100MΩ); Simulation software: Creates an inductor cavity model; Control computer: Can run MATLAB (genetic algorithm implementation, generator rotation commands).

[0077] (2) Experimental steps 2-1) Data Acquisition a. Connect the output of the tuning network to five standard loads in sequence, and set 125 combinations of C1, C2 and L through the control computer.

[0078] b. Measure the S-matrix every 5 kHz in the Fmin-Fmax frequency band using a vector network analyzer and record the impedance data corresponding to each combination.

[0079] c. Valid data were screened according to the data validation mechanism, and finally 120 valid samples were obtained (5 sets of data with excessive error were removed).

[0080] 2-2) Neural Network Training a. Construct the neural network model as described in 3.2.2, and implement it using MATLAB.

[0081] b. First, train the network using effective data with a 200Ω load to determine the basic network parameters; then, fine-tune the network using effective data with 0Ω, 50Ω, 100Ω and open-circuit loads to improve the network's adaptability to different load scenarios.

[0082] 2-3) Tuning Test a. Connect the output of the tuning network to the actual antenna feeder and measure its impedance at the C-frequency point using a vector network analyzer. The impedance is (30+j20)Ω.

[0083] b. Input the impedance data into the neural network to obtain the initial values: C1=350pF, C2=400pF, L=1.8μH.

[0084] c. Set the search range for the genetic algorithm: C1∈[210pF, 525pF], C2∈[240pF, 600pF], L∈[1.08μH, 2.7μH], population size 20, iterations 20 generations.

[0085] d. After tuning, the VSWR dropped to 1.1, with a total time of 10 minutes.

[0086] (3) Comparison of experimental results The test results are compared in Table 1: Table 1 Comparison of Experimental Results

[0087] In addition, such as Figure 4 As shown, the second embodiment of the present invention also proposes a tuning network matching device based on neural network initial value guidance, the device comprising: The dataset construction module 10 is used to connect different standard loads to the output of the tuning network, calculate the impedance of the antenna feeder in each frequency band and the parameter values ​​of the components in the tuning network, and construct the dataset. Training module 20 is used to train the neural network using the dataset to learn the mapping relationship between impedance and the parameter values ​​of the components, so as to obtain a trained neural network. The prediction module 30 is used to obtain the predicted values ​​of the component parameters based on the actual impedance of the antenna feeder in the target frequency band using a trained neural network. The optimization module 40 is used to determine the parameter search range of the optimization algorithm based on the predicted parameter values ​​of the component, and to control the motor to rotate to the motor angle corresponding to the parameter search range according to the pre-stored mapping table, and to perform iterative optimization with minimizing the standing wave value as the objective function to obtain the optimal parameter value of the component.

[0088] As a further preferred technical solution, the dataset construction module 10 specifically includes: The parameter combination traversal unit is used to connect different standard loads to the output of the tuning network and traverse the parameter value combinations of different components. Impedance data calculation unit, used to measure the impedance data of different components under different standard loads by sequentially measuring the parameter values ​​of different components in each frequency band using a vector network analyzer; The parameter solving unit is used to construct a system of equations using impedance data corresponding to a portion of standard loads to solve for the theoretical values ​​of parameters corresponding to different components. The verification unit is used to verify the theoretical values ​​of the parameters using impedance data corresponding to another part of the standard load, and to construct the dataset based on the verified theoretical values ​​of the parameters and the corresponding impedances.

[0089] As a further preferred technical solution, the training module 20 specifically includes: The training unit is used to train the neural network using a portion of the sample data in the dataset to learn the mapping relationship between impedance and the parameter values ​​of the components, and to determine the basic network parameters. The fine-tuning unit is used to fine-tune the parameters of the neural network using another portion of the sample data in the dataset, so as to obtain a trained neural network.

[0090] As a further preferred technical solution, the prediction module 30 specifically includes: The actual impedance calculation unit is used to measure the actual impedance of the antenna feeder in the target frequency band from the output of the tuned network using a vector network analyzer. The prediction unit is used to input the actual impedance of the antenna feeder in the target frequency band into the trained neural network to obtain the predicted parameter values ​​of the component.

[0091] As a further preferred technical solution, the optimization module 40 specifically includes: The parameter search range determination unit is used to determine the parameter search range of the optimization algorithm based on the predicted parameter values ​​of the components.

[0092] in, These are the predicted values ​​for the component's parameters. This represents the actual limit of the component's parameters. , , Indicates product; A motor position lookup unit is used to find the corresponding motor angle from the mapping table based on the parameter search range, and control the motor to rotate to that motor angle. The motor is connected to the component. The parameter optimization unit is used to measure the standing wave value from the input of the tuned network using a vector network analyzer, and to perform iterative optimization with minimizing the standing wave value as the objective function to obtain the optimal parameter values ​​of the component.

[0093] As a further preferred technical solution, the device further includes: The mapping relationship establishment module is used to establish the initial mapping relationship between the motor rotation angle and the component values ​​in the tuning network; The mapping relationship calibration module is used to connect the standard load to the output of the tuning network theoretical model and traverse the initial mapping relationship table to correct the initial mapping relationship between the motor rotation angle and the component value, thus obtaining the mapping relationship table.

[0094] As a further preferred technical solution, the mapping relationship establishment module specifically includes: The first mapping relationship establishment unit is used to control the motor connected to the capacitor to rotate to different angles and measure the measured value of the capacitor when the motor rotates to different angles; linear interpolation fitting is used for the discrete measured values ​​of the capacitance to obtain the mapping relationship between the capacitance value and the motor rotation angle; The second mapping relationship establishment unit is used to establish a three-dimensional electromagnetic model of the inductor L and its surrounding cavity, and to simulate the three-dimensional battery model to obtain the S matrix corresponding to different motor rotation angles; the mapping relationship between the inductance value and the motor rotation angle is obtained by inverting the S matrix; and the initial mapping relationship between the motor rotation angle and the component values ​​in the tuning network is formed by combining the mapping relationship between the capacitance value and the motor rotation angle and the mapping relationship between the inductance value and the motor rotation angle.

[0095] As a further preferred technical solution, the mapping relationship calibration module specifically includes: The mapping relationship traversal unit is used to connect different standard loads to the output of the tuned network theoretical model, traverse different component values ​​in the initial mapping relationship, and calculate the theoretical output impedance corresponding to the currently traversed component value. The error calculation unit is used to take the impedance value of the standard load as the actual output impedance and calculate the average absolute error between the actual output impedance and the theoretical output impedance. The relation correction unit is used to correct the value of the currently traversed element based on the mean absolute error, and obtain the corrected mapping relation table.

[0096] Furthermore, the third embodiment of the present invention also proposes a computer-readable storage medium storing a computer program thereon, which, when executed by a processor, implements the tuning network matching method based on neural network initial value guidance as described in the first embodiment above.

[0097] It should be noted that other embodiments or specific implementation methods of the tuning network matching device and computer-readable storage medium based on neural network initial value guidance described in this invention can refer to the above-described method embodiments, and will not be repeated here.

[0098] It should be noted that the computer-readable medium disclosed in this embodiment may be a computer-readable signal medium or a computer-readable storage medium, or any combination thereof. A computer-readable storage medium may be, for example,—but not limited to—an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any combination thereof. More specific examples of a computer-readable storage medium may include, but are not limited to: an electrical connection having one or more wires, a portable computer disk, a hard disk, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, and portable compact disk read-only memory (CD-ROM). ROM, optical storage devices, magnetic storage devices, or any suitable combination thereof. In this disclosure, a computer-readable storage medium can be any tangible medium containing or storing a program that can be used by or in connection with an instruction execution system, apparatus, or device. In this disclosure, a computer-readable signal medium can include a data signal propagated in baseband or as part of a carrier wave, carrying computer-readable program code. Such propagated data signals can take various forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination thereof. A computer-readable signal medium can also be any computer-readable medium other than a computer-readable storage medium, which can send, propagate, or transmit a program for use by or in connection with an instruction execution system, apparatus, or device. The program code contained on the computer-readable medium can be transmitted using any suitable medium, including but not limited to: wires, optical fibers, RF (radio frequency), etc., or any suitable combination thereof.

[0099] The aforementioned computer-readable medium may be included in the aforementioned electronic device; or it may exist independently and not assembled into the electronic device. The aforementioned computer-readable medium carries one or more programs, which, when executed by the electronic device, cause the electronic device to perform a zero-sample image anomaly detection method according to the above embodiments.

[0100] Computer program code for performing the operations of this disclosure can be written in one or more programming languages ​​or a combination thereof, including object-oriented programming languages ​​such as Java, Smalltalk, and C++, and conventional procedural programming languages ​​such as the "C" language or similar programming languages. The program code can be executed entirely on the user's computer, partially on the user's computer, as a standalone software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server.

[0101] In cases involving remote computers, the remote computer can be connected to the user's computer via any type of network—including a local area network (LAN) or a wide area network (WAN), or it can be connected to an external computer (e.g., via the Internet using an Internet service provider).

[0102] It should be understood that various parts of the present invention can be implemented in hardware, software, firmware, or a combination thereof. In the above embodiments, multiple steps or methods can be implemented in software or firmware stored in memory and executed by a suitable instruction execution system. For example, if implemented in hardware, as in another embodiment, it can be implemented using any one or a combination of the following techniques known in the art: discrete logic circuits having logic gates for implementing logical functions on data signals, application-specific integrated circuits (ASICs) having suitable combinational logic gates, programmable gate arrays (PGAs), field-programmable gate arrays (FPGAs), etc.

[0103] In the description of this specification, references to terms such as "one embodiment," "some embodiments," "example," "specific example," or "some examples," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of the invention. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples.

[0104] Furthermore, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of indicated technical features. Thus, a feature defined as "first" or "second" may explicitly or implicitly include at least one of that feature. In the description of this invention, "a plurality of" or "several" means at least two, such as two, three, etc., unless otherwise explicitly specified.

[0105] Although embodiments of the present invention have been shown and described above, it is understood that the above embodiments are exemplary and should not be construed as limiting the present invention. Those skilled in the art can make changes, modifications, substitutions and variations to the above embodiments within the scope of the present invention.

Claims

1. A tuning network matching method based on neural network initial value guidance, characterized in that, include: Different standard loads are connected to the output of the tuning network, the impedance of the antenna feeder in each frequency band and the parameter values ​​of the components in the tuning network are calculated, and a dataset is constructed. The neural network is trained using a dataset to learn the mapping relationship between impedance and the parameter values ​​of the components, resulting in a well-trained neural network. The trained neural network is used to obtain the predicted values ​​of the component parameters based on the actual impedance of the antenna feeder in the target frequency band. The parameter search range of the optimization algorithm is determined based on the predicted parameter values ​​of the components. The motor is controlled to rotate to the motor angle corresponding to the parameter search range according to the pre-stored mapping table. The optimal parameter values ​​of the components are obtained by iterative optimization with the minimization of standing wave value as the objective function.

2. The tuning network matching method based on neural network initial value guidance as described in claim 1, characterized in that, The components in the tuning network include inductors and capacitors. The process of connecting different standard loads to the output of the tuning network, calculating the impedance of the antenna feeder in each frequency band and the parameter values ​​of the components in the tuning network, and constructing a dataset includes: Different standard loads are connected to the output of the tuning network, and the parameter value combinations of different components are iterated. The impedance data of different component parameter combinations under different standard loads were measured sequentially in each frequency band using a vector network analyzer. A system of equations was constructed using impedance data corresponding to a portion of standard loads to solve for the theoretical values ​​of parameters corresponding to different components. The theoretical values ​​of the parameters are verified using impedance data corresponding to another set of standard loads, and the dataset is constructed based on the verified theoretical values ​​of the parameters and the corresponding impedances.

3. The tuning network matching method based on neural network initial value guidance as described in claim 2, characterized in that, The standard loads include short-circuit loads, XΩ, YΩ, ZΩ, and open-circuit loads; Where X, Y, and Z are pre-defined ohm values ​​and all represent non-zero, non-infinite values.

4. The tuning network matching method based on neural network initial value guidance as described in claim 1, characterized in that, The process of training a neural network using a dataset to learn the mapping relationship between impedance and component parameter values, resulting in a trained neural network, includes: The neural network is trained using a portion of the sample data in the dataset to learn the mapping relationship between impedance and the parameter values ​​of the components, and to determine the basic network parameters. The parameters of the neural network are fine-tuned using another portion of the sample data in the dataset to obtain a trained neural network.

5. The tuning network matching method based on neural network initial value guidance as described in claim 1, characterized in that, The method of using a trained neural network to obtain the predicted parameter values ​​of the components based on the actual impedance of the antenna feeder in the target frequency band includes: The actual impedance of the antenna feeder in the target frequency band was measured from the output of the tuned network using a vector network analyzer. The actual impedance of the antenna feeder in the target frequency band is input into the trained neural network to obtain the predicted parameter values ​​of the component.

6. The tuning network matching method based on neural network initial value guidance as described in claim 1, characterized in that, The process of determining the parameter search range of the optimization algorithm based on the predicted parameter values ​​of the component, controlling the motor to rotate to the motor angle corresponding to the parameter search range according to the pre-stored mapping table, and performing iterative optimization with minimizing the standing wave value as the objective function to obtain the optimal parameter values ​​of the component includes: The parameter search range of the optimization algorithm is determined based on the predicted parameter values ​​of the components: in, These are the predicted values ​​for the component's parameters. This represents the actual limit of the component's parameters. , , Indicates product; The corresponding motor angle is found from the mapping table based on the parameter search range, and the motor is controlled to rotate to that motor angle. The motor is connected to the component. The standing wave ratio (SWR) is measured at the input of the tuned network using a vector network analyzer, and the optimal parameter values ​​of the component are obtained by iterative optimization with the minimum SWR as the objective function.

7. The tuning network matching method based on neural network initial value guidance as described in claim 1, characterized in that, The optimization algorithm used is a genetic algorithm, an ant colony algorithm, or a particle swarm optimization algorithm.

8. The tuning network matching method based on neural network initial value guidance as described in claim 1, characterized in that, The process of constructing the mapping table includes: Establish the initial mapping relationship between the motor rotation angle and the component values ​​in the tuning network; The standard load is connected to the output of the tuned network theoretical model, and the initial mapping table is traversed to correct the initial mapping relationship between the motor rotation angle and the component value, thus obtaining the mapping table.

9. The tuning network matching method based on neural network initial value guidance as described in claim 8, characterized in that, The establishment of the initial mapping relationship between the motor rotation angle and the component values ​​in the tuning network includes: Control the motor connected to the capacitor to rotate to different angles, and measure the actual capacitance value of the capacitor when the motor rotates to different angles; Linear interpolation was used to fit the discrete measured capacitance values ​​to obtain the mapping relationship between the capacitance value and the motor rotation angle. A three-dimensional electromagnetic model of the inductor L and its surrounding cavity was established, and the three-dimensional battery model was simulated to obtain the S matrix corresponding to different motor rotation angles. The mapping relationship between inductance value and motor rotation angle is obtained by inverting the S-matrix; By combining the mapping relationship between capacitor value and motor rotation angle and the mapping relationship between inductance value and motor rotation angle, an initial mapping relationship between motor rotation angle and component values ​​in the tuning network is formed.

10. The tuning network matching method based on neural network initial value guidance as described in claim 9, characterized in that, The mapping relationship between inductance value and motor rotation angle obtained by inverting the S-matrix includes: The input impedance of the inductor is derived by inverting the S-matrix; The inductance value is calculated based on the imaginary part of the input impedance, expressed by the following formula: In the formula, This represents the imaginary part of the input impedance. For operating frequency, Pi Indicates product; Based on the calculated inductance value, the mapping relationship between the inductance value and the motor rotation angle is obtained.

11. The tuning network matching method based on neural network initial value guidance as described in claim 8, characterized in that, A standard load is connected to the output of the tuned network theoretical model, and the initial mapping table is traversed to correct the initial mapping relationship between the motor rotation angle and the component value, resulting in the mapping table, which includes: Different standard loads are connected to the output of the tuned network theoretical model, and different component values ​​in the initial mapping relationship are traversed. The theoretical output impedance corresponding to the current traversed component value is calculated. The impedance value of the standard load is used as the actual output impedance, and the mean absolute error between the actual output impedance and the theoretical output impedance is calculated. The value of the currently traversed component is corrected based on the mean absolute error, resulting in a corrected mapping table.

12. The tuning network matching method based on neural network initial value guidance as described in claim 11, characterized in that, The correction of the currently traversed component values ​​based on the mean absolute error is used to obtain a correction mapping table, expressed by the formula: In the formula, This is the corrected capacitance value. This is the corrected inductance value. and These are the capacitance and inductance values ​​currently being iterated. This represents the mean absolute error.

13. A tuning network matching device based on neural network initial value guidance, characterized in that, include: The dataset construction module is used to connect different standard loads to the output of the tuning network, calculate the impedance of the antenna feeder in each frequency band and the parameter values ​​of the components in the tuning network, and construct the dataset. The training module is used to train the neural network using the dataset to learn the mapping relationship between impedance and the parameter values ​​of the components, and to obtain a trained neural network. The prediction module is used to obtain the predicted values ​​of the component parameters based on the actual impedance of the antenna feeder in the target frequency band using a trained neural network. The optimization module is used to determine the parameter search range of the optimization algorithm based on the predicted parameter values ​​of the components, and to control the motor to rotate to the motor angle corresponding to the parameter search range according to the pre-stored mapping table. The optimal parameter values ​​of the components are obtained by iterative optimization with minimizing the standing wave value as the objective function.

14. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the tuning network matching method based on neural network initial value guidance as described in any one of claims 1-12.

Images