An automatic tuning method and device for impedance matching network

Abstract

The application relates to the technical field of automatic impedance matching of radio frequency circuits, and provides an impedance matching network automatic tuning method and device, which comprises the following steps: obtaining input impedance of a matching network, the input impedance comprising a resistance component and a reactance component; constructing an error function based on the error of the input impedance and target impedance; calculating the gradient of the error function with respect to each adjustable element parameter through a numerical differentiation method; determining the learning rate weight of each adjustable element according to the error proportion of the resistance component and the error proportion of the reactance component; and updating the parameter value of each adjustable element based on the gradient and the learning rate weight. The application does not need to depend on the specific topological model of the matching network, and has strong universality; through dynamic allocation of the learning rate weight, multivariable decoupling control is realized, the slow convergence problem is effectively avoided, and the tuning speed and stability are significantly improved.

Description

Technical Field

[0001] This application relates to the field of automatic impedance matching in radio frequency circuits, and more particularly to an automatic tuning method and apparatus for impedance matching networks. Background Technology

[0002] In radio frequency (RF) systems, impedance matching is a crucial technical aspect for ensuring efficient signal power transmission. Good impedance matching prevents power reflection, optimizes RF source output efficiency, and protects RF power devices from overvoltage damage. In practical engineering applications, load impedance often dynamically shifts with changes in operating frequency, ambient temperature, and other factors; therefore, the matching network needs to possess real-time adaptive tuning capabilities.

[0003] Existing automatic impedance matching technologies typically employ lookup tables or gradient search methods based on analytical models. However, lookup tables require pre-measuring and storing a large amount of matching state data, consuming significant storage space and struggling to cover all load variations. Gradient search methods based on analytical models depend on the specific topology of the matching network (e.g., L-type, T-type, or π-type), requiring the derivation of complex impedance transformation formulas and exhibiting poor versatility. Furthermore, in multivariable tuned systems, adjustable components (e.g., adjustable capacitors) often have coupling effects on the real and imaginary parts of the input impedance. Traditional gradient descent algorithms, when dealing with such multivariable coupling problems, are prone to slow convergence, oscillations during tuning, or even getting trapped in local optima due to improper step size selection or mutual constraints on gradient directions, making it difficult to meet the demands of applications with high real-time requirements. Summary of the Invention

[0004] This application provides an automatic tuning method for impedance matching networks, which solves the technical problems of poor versatility and slow convergence speed caused by multivariable coupling in existing impedance matching tuning methods.

[0005] To achieve the above objectives, this application adopts the following technical solution: In a first aspect, an automatic tuning method for an impedance matching network is provided, comprising: acquiring the input impedance of the matching network, the input impedance including a resistive component and a reactance component; constructing an error function based on the error between the input impedance and a target impedance; calculating the gradient of the error function with respect to the parameters of each adjustable element by numerical differentiation; determining the learning rate weight of each adjustable element according to the error proportion of the resistive component and the error proportion of the reactance component; and updating the parameter values ​​of each adjustable element based on the gradient and the learning rate weight.

[0006] The gradient is calculated by numerical differentiation, which does not depend on the specific topology model of the matching network, thus improving the versatility of the algorithm. At the same time, the learning rate weights of each adjustable element are determined independently according to the error proportions of the resistance and reactance components, realizing decoupled control of the multivariable tuning process. This effectively avoids the slow convergence problem caused by multivariable coupling and significantly improves the tuning speed and stability.

[0007] As one implementation, determining the learning rate weight of each adjustable element based on the error proportion of the resistance component and the error proportion of the reactance component includes: allocating the error proportion of the resistance component to the adjustable elements sensitive to the resistance component and allocating the error proportion of the reactance component to the adjustable elements sensitive to the reactance component based on the sensitivity relationship between each adjustable element to the resistance component and the reactance component.

[0008] In one implementation, the error proportions of the resistance component and the reactance component are determined as follows: the absolute value of the error between the resistance component and the target impedance is calculated to obtain the real part error; the absolute value of the reactance component is calculated; the sum of the real part error and the absolute value of the reactance component is calculated to obtain the total error; the ratio of the real part error to the total error is determined as the error proportion of the resistance component; and the ratio of the absolute value of the reactance component to the total error is determined as the error proportion of the reactance component.

[0009] As one implementation, the step of calculating the gradient of the error function with respect to each adjustable element parameter by numerical differentiation includes: applying positive and negative bidirectional perturbations to the current adjustable element parameters; calculating the error function values ​​after applying the positive and negative perturbations respectively; and calculating the partial derivative of the error function with respect to the current adjustable element parameters based on the changes in the error function values ​​corresponding to the positive and negative perturbations.

[0010] In one implementation, the partial derivative is calculated using the central difference formula: the partial derivative is equal to the difference between the first error value and the second error value, divided by twice the perturbation step size; wherein, the first error value is the error function value calculated after applying a positive perturbation based on the current adjustable element parameters, and the second error value is the error function value calculated after applying a negative perturbation based on the current adjustable element parameters.

[0011] As one implementation method, updating the parameter values ​​of each adjustable element further includes: obtaining the historical adjustment amount of each adjustable element in the previous iteration; weighting and superimposing the historical adjustment amount with the adjustment amount of the current iteration to obtain the final adjustment amount; and updating the parameter values ​​of each adjustable element based on the final adjustment amount.

[0012] The above scheme introduces historical adjustment as inertial momentum, which can automatically amplify the effective step size when the gradient direction is consistent in continuous iteration, thereby accelerating the convergence speed of the algorithm in the flat region of the error surface.

[0013] In one implementation, the final adjustment is calculated using the following formula: the final adjustment equals the sum of the adjustment of the current iteration and the momentum coefficient multiplied by the historical adjustment of the previous iteration; wherein the adjustment of the current iteration is equal to the learning rate weight multiplied by the negative value of the gradient.

[0014] In one implementation, the error function includes: the square of the error between the resistance component and the target impedance; and the square of the reactance component.

[0015] Furthermore, the present invention also provides an automatic tuning device for impedance matching networks, comprising: An impedance acquisition module is configured to acquire the input impedance of a matching network, wherein the input impedance includes a resistive component and a reactance component; The error construction module is configured to construct an error function based on the error between the input impedance and the target impedance; The gradient calculation module is configured to calculate the gradient of the error function with respect to the parameters of each adjustable element by means of numerical differentiation; The weight determination module is configured to determine the learning rate weight of each adjustable element based on the error proportion of the resistive component and the error proportion of the reactance component. The parameter update module is configured to update the parameter values ​​of each adjustable element based on the gradient and the learning rate weights.

[0016] The above-mentioned device realizes the automatic tuning method of the impedance matching network through the coordinated work of its various modules, and has the beneficial effects of strong versatility, fast convergence speed and stable tuning.

[0017] In one embodiment, the device further includes a processor configured to execute a program stored in a memory to implement the method as described in any of the above embodiments; The memory is configured to store the program; An adjustable element interface is configured to connect to each adjustable element in the matching network and adjust the parameter values ​​of each adjustable element according to the instructions of the parameter update module.

[0018] The present invention provides an automatic tuning method and apparatus for impedance matching networks, which has the following significant advantages: High versatility: Gradients are calculated using numerical differentiation, which does not depend on the specific topology model or prior parameters of the matching network. It can be adapted to any topology of the adjustable matching network, which greatly reduces the customization requirements of the algorithm on the hardware platform.

[0019] Fast convergence speed: By introducing a decoupled learning rate mechanism, the learning rate weights of each adjustable element are dynamically allocated according to the proportion of the real part error and the imaginary part error of the input impedance. This allows the elements corresponding to the main error components to obtain a larger update step size, effectively avoiding the slow convergence problem caused by the mutual restraint of multiple variables in traditional gradient descent.

[0020] Smooth and controllable tuning: An inertial momentum mechanism is introduced, which uses the exponentially decaying average of historical update quantities as the inertial term. When the gradient direction is consistent in continuous iteration, the effective step size is automatically amplified. Even in the flat region of the error surface, the speed of advancement can still be maintained, while oscillation is suppressed, making the tuning process smooth and controllable.

[0021] It should be understood that the descriptions of technical features, technical solutions, beneficial effects, or similar language in this application do not imply that all features and advantages can be achieved in any single embodiment. Rather, it is understood that the description of a feature or beneficial effect means that a specific technical feature, technical solution, or beneficial effect is included in at least one embodiment. Therefore, the descriptions of technical features, technical solutions, or beneficial effects in this specification do not necessarily refer to the same embodiment. Furthermore, the technical features, technical solutions, and beneficial effects described in this embodiment can be combined in any suitable manner. Those skilled in the art will understand that embodiments can be implemented without one or more specific technical features, technical solutions, or beneficial effects of a particular embodiment. In other embodiments, additional technical features and beneficial effects may be identified in specific embodiments that do not embody all embodiments. Attached Figure Description

[0022] Figure 1 This is a schematic diagram of the L-shaped matching network circuit structure provided in the embodiments of this application; Figure 2 A flowchart of an automatic tuning method for an impedance matching network provided in an embodiment of this application; Figure 3 This is a schematic diagram of the data changes during the tuning process provided in an embodiment of this application; Figure 4 This is a schematic diagram of the impedance trajectory variation on the Smith chart provided in the embodiments of this application.

[0023] Among them, C2 is a parallel adjustable capacitor, and C3 is a series adjustable capacitor. Detailed Implementation

[0024] In the description of this application, unless otherwise stated, " / " means "or," for example, A / B can mean A or B. The "and / or" in this document is merely a description of the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent: A alone, A and B simultaneously, and B alone. Furthermore, "at least one" means one or more, and "multiple" means two or more. The terms "first," "second," etc., do not limit the quantity or order of execution, and "first," "second," etc., do not necessarily imply differences.

[0025] It should be noted that, in this application, the terms "exemplary" or "for example" are used to indicate that something is being described as an example, illustration, or illustration. Any embodiment or design described as "exemplary" or "for example" in this application should not be construed as being more preferred or advantageous than other embodiments or design solutions. Specifically, the use of terms such as "exemplary" or "for example" is intended to present the relevant concepts in a concrete manner.

[0026] Example 1 like Figure 2 As shown, this embodiment provides an automatic tuning method for impedance matching networks. This method is applicable to RF matching networks containing tunable components and aims to address the slow convergence speed caused by the dependence of existing tuning algorithms on specific network topology models and multivariable coupling. The method includes the following steps: Step S100: Obtain the input impedance of the matching network; The input impedance includes a resistive component and a reactance component.

[0027] In some implementations, this step involves acquiring the complex impedance signal at the input of the matching network in real time using an impedance detection circuit or sensor. In practical applications, the matching network may consist of adjustable capacitors, adjustable inductors, or combinations thereof, and its topology can be L-type, T-type, π-type, or any other arbitrary form.

[0028] This embodiment does not depend on a specific network topology. It only needs to obtain the values ​​of the real part (resistance component) and the imaginary part (reactance component) of the input impedance to perform subsequent tuning calculations, which lays the foundation for the universality of the algorithm.

[0029] Step S200: Construct an error function based on the error between the input impedance and the target impedance.

[0030] In some implementations, the target impedance is typically set to the characteristic impedance of the system, such as 50Ω. The error function is used to quantify the degree of mismatch between the current input impedance and the target impedance.

[0031] By constructing a function that includes both resistive and reactive error components, the impedance matching problem can be transformed into an optimization problem of finding the minimum value of the error function.

[0032] The error function can be constructed in a flexible manner, as long as it can comprehensively reflect the error of the real and imaginary parts. For example, it can be in the form of a sum of squares or other function forms that can characterize distance.

[0033] Step S300: Calculate the gradient of the error function with respect to the parameters of each adjustable element using numerical differentiation.

[0034] In some implementations, this step is crucial for achieving universal tuning in this embodiment. Traditional analytical gradient methods require deriving complex impedance transformation formulas based on specific circuit topologies. Once the network structure changes, the formulas need to be re-derived, resulting in poor universality. In contrast, this embodiment employs a numerical differentiation approach, treating the matching network as a "black box" system. By applying small perturbations to the parameters of each adjustable component and observing the changes in the error function, the gradient can be approximated.

[0035] This approach does not require prior knowledge of the specific topology and internal parameter model of the matching network, and can adapt to any form of adjustable matching network, greatly reducing the algorithm's dependence on customized hardware platforms.

[0036] Step S400: Determine the learning rate weight of each adjustable element based on the error proportion of the resistive component and the error proportion of the reactance component.

[0037] It should be noted that in multivariable tuned systems, different adjustable components often have different sensitivities to the real and imaginary parts of the input impedance, and there are coupling effects between the components. If a uniform learning rate is used, it can easily lead to slow correction of the main error components, or oscillations due to the constraints between variables.

[0038] This embodiment dynamically assigns different learning rate weights to each adjustable element by analyzing the proportion of resistive and reactive component errors in the total error. The underlying logic is that when the error proportion of a certain component is large, it means that this component is the main cause of the current mismatch. At this time, the adjustment intensity of the adjustable element that is more sensitive to the influence of this component should be increased, that is, it should be given a larger learning rate weight; conversely, the weight should be reduced.

[0039] This decoupled control strategy ensures that the adjustment step size of each component matches its contribution to error elimination, effectively avoiding the slow convergence problem caused by multivariable coupling.

[0040] Step S500: Update the parameter values ​​of each adjustable element based on the gradient and the learning rate weights.

[0041] In some implementations, based on the calculated gradient direction and determined learning rate weights, the adjustment amount of each adjustable element is calculated, and the corresponding motor or drive circuit is controlled to adjust the parameter values ​​of the adjustable capacitor or inductor, thereby changing the impedance transformation characteristics of the matching network and bringing the input impedance closer to the target impedance. By iteratively executing the above steps, the error function gradually decreases until a matching state is achieved.

[0042] Through the above scheme, this embodiment uses numerical differentiation to get rid of the dependence on the network topology model and realizes the generalization of the algorithm; at the same time, through the decoupling learning rate mechanism based on error proportion, it solves the problem of slow convergence caused by multivariate coupling and realizes fast and stable automatic impedance matching.

[0043] Example 2 This embodiment, based on embodiment 1, provides a detailed explanation of the specific process for determining the learning rate weights of each adjustable element.

[0044] In step S400, based on the sensitivity relationship between each adjustable element to the resistance component and the reactance component, the error proportion of the resistance component is allocated to the adjustable element sensitive to the resistance component, and the error proportion of the reactance component is allocated to the adjustable element sensitive to the reactance component.

[0045] It should be noted that different adjustable components in a matching network often have different sensitivities to the real and imaginary parts of the input impedance. For example, in a common L-type matching network, the parallel adjustable capacitor mainly affects the resistive component (real part) of the input impedance, while the series adjustable capacitor mainly affects the reactive component (imaginary part) of the input impedance.

[0046] This embodiment matches the error percentage with the adjustment capability of the components. When the error percentage of the resistance component is large, it indicates that the current mismatch is mainly caused by the real part error. In this case, the adjustment force of adjustable components sensitive to the resistance component (such as parallel capacitors) should be increased; conversely, the adjustment force of adjustable components sensitive to the reactance component (such as series capacitors) should be increased.

[0047] This allocation method achieves a precise correspondence between error sources and adjustment methods, thereby effectively solving the problem of slow convergence caused by multivariate coupling.

[0048] Furthermore, the error proportions of the resistance component and the reaction component are determined through normalization calculations.

[0049] Specifically, this embodiment uses the following steps to calculate the proportion of each error component: Step S401: Calculate the absolute value of the error between the resistance component and the target impedance to obtain the real part of the error.

[0050] For example, suppose the resistive component of the current input impedance is The target impedance is (Typically 50Ω), then the real part error is .

[0051] Step S402: Calculate the absolute value of the reactance component.

[0052] Assume the reactance component of the current input impedance is If the target value of the reactance component is 0 under ideal matching conditions, then the deviation of the current reactance component from the target value is... Since it is itself, its absolute value is taken. As a measure of the error of the imaginary part.

[0053] Step S403: Calculate the sum of the absolute values ​​of the real part error and the reactance component to obtain the total error. That is, the total error. .

[0054] Step S404: The ratio of the real part error to the total error is determined as the error proportion of the resistance component, i.e., the weighting coefficient. .

[0055] Step S405: The ratio of the absolute value of the reactance component to the total error is determined as the error proportion of the reactance component, i.e., the weighting coefficient. .

[0056] Through the above normalization process, regardless of the absolute values ​​of the real and imaginary errors, the sum of their weight coefficients remains 1, ensuring that the learning rate weights are allocated at a reasonable level and preventing the adjustment step size from becoming uncontrolled due to an excessively large error component. In practical applications, to prevent calculation errors caused by a zero denominator, a very small smoothing factor ε can be added to the calculation of the total error, i.e. .

[0057] At this point, a base learning rate is set. Then the learning rates of capacitors C2 and C3 are... , .

[0058] Combination Figure 3 The diagram shows the data changes during the tuning process. In the initial tuning phase, assuming a large real part error, the calculated percentage of the resistance component error is shown below. Approaching 1. Based on the sensitivity relationship, this weight is allocated to the adjustable element sensitive to the real part, significantly increasing the learning rate weight of this element, thus rapidly correcting the real part error with a larger step size in the early stages of iteration. As iteration progresses, the real part error decreases rapidly, while the imaginary part error may become the main problem, at which point the error proportion of the reactance component increases. As the value gradually increases, the system automatically shifts the adjustment focus to the adjustable element that is sensitive to the imaginary part.

[0059] This dynamic weight allocation mechanism enables the tuning process to quickly correct the main contradictions, significantly reducing the number of iterations and achieving rapid convergence.

[0060] Example 3 This embodiment, based on Embodiment 1, provides a detailed explanation of the specific process of calculating the gradient using numerical differentiation.

[0061] In step S300, the gradient of the error function with respect to each adjustable element parameter is calculated using numerical differentiation, specifically including: Apply positive and negative bidirectional perturbations to the current adjustable component parameters; Calculate the error function values ​​after applying positive and negative perturbations respectively; Based on the changes in the error function values ​​corresponding to the positive and negative disturbances, the partial derivatives of the error function with respect to the current adjustable element parameters are calculated.

[0062] This embodiment uses the differential perturbation method to approximate the gradient calculation. Since the internal topology of the matching network may be unknown or too complex, making it difficult to derive the analytical derivative expression, this embodiment uses the active change of the parameter values ​​of the adjustable element and the observation of the response of the error function to infer the gradient direction.

[0063] For any adjustable component parameter, such as the capacitance value of an adjustable capacitor, a small increment (positive perturbation) and a small decrement (negative perturbation) are added to its current value. By measuring the error function values ​​at these two perturbation points, the trend of the error function changing with the parameter can be obtained. If a positive perturbation increases the error function value, while a negative perturbation decreases it, then the gradient direction points in the direction of parameter decrease, and vice versa. This method of calculating the gradient using measured data completely eliminates the dependence on the specific circuit model of the matching network, enabling the algorithm to adapt to matching networks with various topologies and greatly improving its versatility.

[0064] Furthermore, the partial derivative is calculated using the central difference formula: the partial derivative is equal to the difference between the first error value and the second error value, divided by twice the perturbation step size; wherein, the first error value is the error function value calculated after applying a positive perturbation based on the current adjustable element parameters, and the second error value is the error function value calculated after applying a negative perturbation based on the current adjustable element parameters.

[0065] Specifically, for The partial derivatives are calculated as follows: ;for The partial derivatives are calculated as follows: In the formula, , For the k-th iteration, the error function J corresponds to the capacitance. ,capacitance The partial derivatives, This represents the error function value at the k-th iteration. , The capacitance at the k-th iteration is respectively , The current parameter value, , For capacitors respectively , Compensation for applied small perturbations, , They are respectively in Add or remove a disturbance The error function value was then calculated. , They are respectively in Add or remove a disturbance The error function value was then calculated.

[0066] In this embodiment, compared to forward or backward differencing methods that only apply unidirectional perturbations, the central difference formula used in this embodiment has higher numerical calculation accuracy. Mathematically, the truncation error of forward differencing is O(δ), while the truncation error of central differencing is O(δ²), where δ is the perturbation step size. This means that with the same perturbation step size, central differencing can more accurately approximate the true derivative value, thus providing more precise directional guidance for gradient descent and reducing tuning oscillations or convergence errors caused by gradient calculation errors.

[0067] During implementation, the choice of perturbation step size requires a trade-off between computational accuracy and system noise immunity. If the step size is too small, although theoretically more accurate, it is easily affected by measurement noise, leading to drastic fluctuations in the calculated gradient value. If the step size is too large, it may exceed the linear approximation range, making the difference results unable to accurately reflect the local gradient. Typically, the perturbation step size can be set between 0.1% and 1% of the adjustable component parameter adjustment range, or adaptively adjusted according to the system noise level. Through the above central difference calculation, the influence of each adjustable component parameter on the error function can be accurately obtained, providing a reliable basis for subsequent parameter updates.

[0068] Example 4 Based on Example 1, this embodiment further optimizes the parameter update process by introducing a momentum mechanism to accelerate convergence and suppress oscillations.

[0069] Specifically, in step S500, when updating the parameter values ​​of each adjustable element, the following steps are also included: Obtain the historical adjustment amount of each adjustable element in the previous iteration; The historical adjustment amount is weighted and summed with the adjustment amount of the current iteration to obtain the final adjustment amount; The parameter values ​​of each adjustable element are updated based on the final adjustment amount.

[0070] In traditional gradient descent algorithms, parameter updates depend only on the current gradient direction and step size. However, on the error surface of impedance matching, there are often flat regions or narrow valleys where the gradient value is small, resulting in an excessively small iteration step size and extremely slow convergence. Alternatively, near the optimal solution, the gradient direction may oscillate due to measurement noise or surface characteristics, making precise convergence difficult. This embodiment introduces a momentum mechanism, utilizing the inertia of historical adjustments to correct the current update direction.

[0071] Specifically, the final adjustment is equal to the sum of the adjustment in the current iteration and the momentum coefficient multiplied by the historical adjustment in the previous iteration; wherein the adjustment in the current iteration is equal to the learning rate weight multiplied by the negative value of the gradient.

[0072] For example, assuming the current iteration number is k, for a certain adjustable element, its final adjustment amount is... The calculation formula is: .in, The adjustment amount is calculated based on the current gradient. The learning rate weights determined in the foregoing embodiments, The current gradient; This is the momentum coefficient, which typically ranges from 0 to 1. This is the historical adjustment amount from the previous iteration.

[0073] It should be noted that when β is large, the historical adjustment amount accounts for a large proportion of the final adjustment amount, meaning that the parameter update direction has strong inertia and is not easily affected by the current gradient noise. It can quickly pass through flat regions, but it may also miss the optimal solution due to excessive inertia. When β is small, the parameter update mainly depends on the current gradient, the inertia effect is weakened, the algorithm is more flexible, but it may converge more slowly. In practical applications, the specific value of β can be set or adaptively adjusted according to the dynamic characteristics of the matching network. For example, in the early stage of tuning, in order to quickly approach the target region, a large β value (such as 0.8 or 0.9) can be set; when approaching the matching state, in order to improve the convergence accuracy, the β value can be appropriately reduced (such as 0.5 or smaller).

[0074] By introducing a momentum mechanism, this embodiment achieves accelerated convergence. When the gradient directions are consistent across multiple iterations, the momentum term accumulates, gradually increasing the actual adjustment step size. This accelerates the algorithm's progress in the flat region of the error surface, solving the problem of slow convergence in flat regions in traditional gradient descent. Furthermore, when the gradient direction oscillates around the optimal solution, the positive and negative momentum terms cancel each other out, reducing the actual adjustment step size and enabling the algorithm to converge smoothly to the optimal solution, avoiding overshoot and oscillations caused by excessively large step sizes.

[0075] Example 5 This embodiment, based on Embodiment 1, provides a detailed explanation of the specific construction form of the error function.

[0076] In step S200, the error function includes: the square term of the error between the resistance component and the target impedance; and the square term of the reactance component.

[0077] Specifically, the error function is a quantitative indicator that measures the distance between the current input impedance and the target impedance. The error function J constructed in this embodiment can be expressed as: ;in, The resistive component of the current input impedance. The target impedance is (typically 50Ω). This represents the reactance component of the current input impedance.

[0078] By employing the squared term form, the error function is ensured to always be non-negative, avoiding the problem of positive and negative errors canceling each other out. This makes the optimization objective clearly defined as finding the minimum value of the function. Secondly, the squared term amplifies large errors and suppresses small errors. This allows the tuning algorithm to generate a large gradient signal when the mismatch is severe, resulting in a fast response. Conversely, as the algorithm approaches a matching state, the gradient signal automatically decreases, which helps improve the final matching accuracy and prevents overshoot oscillations.

[0079] It should be noted that the formula uses [a specific method / approach] for the resistance component. This is because the goal of matching is to make the input resistance equal to the target impedance. As for the reactance component, since the reactance should be 0 under ideal matching conditions, it is directly adopted. This combination method can reflect the matching degree of the real and imaginary parts in a balanced way, avoiding the dominance of the optimization process by a single component with an excessively large value, and ensuring that the algorithm can converge to the target value of both the real and imaginary parts simultaneously. At the same time, the derivative of the error function with respect to each parameter is easy to calculate, which facilitates the rapid acquisition of gradient information through numerical differentiation, reduces computational complexity, and is beneficial for real-time implementation on embedded processing platforms (such as FPGA and DSP).

[0080] Example 6 To verify the effectiveness of the automatic tuning method for impedance matching networks provided by this invention, a specific L-type matching network is used as an example for detailed explanation below. It should be understood that the L-type matching network is merely a preferred application scenario of the technical solution of this invention. Those skilled in the art can apply this method to T-type, π-type, or other arbitrary topological matching networks, all of which fall within the protection scope of this invention.

[0081] like Figure 1 As shown, the L-type matching network circuit structure used in this embodiment includes: a parallel adjustable capacitor C2 and a series adjustable capacitor C3.

[0082] One end of the parallel adjustable capacitor C2 is connected between the signal source and the series adjustable capacitor C3, and the other end is grounded; the series adjustable capacitor C3 is connected between the parallel adjustable capacitor C2 and the load.

[0083] It should be noted that, according to circuit theory, in this topology, the parallel adjustable capacitor C2 has a greater impact on the resistive component (real part) of the input impedance, while the series adjustable capacitor C3 has a greater impact on the reactive component (imaginary part) of the input impedance. This forms the basis for determining the sensitivity relationship in this embodiment.

[0084] The simulation parameters are set as follows: operating frequency is 13.56MHz, and the target impedance of the system is... The impedance is 50Ω. The load impedance is set to... Ω. In the initial state, the input impedance of the matching network is far from the target impedance, and the standing wave ratio (VSWR) is high. The target VSWR is set to VSWR < 1.1, which is considered to be the matching state.

[0085] The method described in the above embodiments performs automatic tuning, and the specific process is as follows: First, obtain the input impedance of the current matching network and calculate the resistance component error and reactance component error. In the initial stage of tuning, due to the large error between the real part (25Ω) and the target impedance (50Ω) of the load impedance, and the large error in the imaginary part (60Ω), the proportion of real part error and imaginary part error calculated by the system are both quite significant.

[0086] Next, the learning rate weights are determined based on the error proportions. Since C2 is sensitive to the real part and C3 is sensitive to the imaginary part, the system automatically assigns the larger proportion of real part error to the learning rate weight of C2 and the larger proportion of imaginary part error to the learning rate weight of C3. This decoupling mechanism allows the two capacitors to independently and quickly correct their respective dominant error components, avoiding the mutual constraints caused by dependent variable coupling in traditional methods.

[0087] Simultaneously, the gradient is calculated using numerical differentiation, and the capacitance parameters are updated in conjunction with a momentum mechanism. During the iteration process, the momentum term leverages the inertia of historical adjustments to accelerate progress in regions with relatively flat error surfaces and suppress oscillations near the optimal solution.

[0088] Combination Figure 3 and Figure 4 As shown, this illustrates the data changes and impedance trajectory during the tuning process of numerical gradient descent (independent of network topology). From Figure 3 As can be seen from the iterative curves, the error function value decreases rapidly with the number of iterations, and the VSWR curve stabilizes quickly. Specifically, under the simulation conditions of this embodiment, only about 14 iterations are needed to reduce the VSWR to below 1.1, satisfying the preset matching conditions. The parameter values ​​of capacitors C2 and C3 rapidly approach the optimal solution during the iteration process, and the change process is stable without any violent oscillations.

[0089] from Figure 4 The Smith chart trajectory visually demonstrates that the impedance trajectory moves rapidly from the initial mismatch region (lower left) towards the center of the chart (matching point). Due to the decoupled learning rate, the adjustment step size of the impedance trajectory in the real and imaginary parts is reasonably allocated, resulting in a near-straight, fast-converging path, rather than the spiral-like wandering path common in traditional methods. This invention achieves fast, stable, and high-precision automatic impedance matching through a decoupled learning rate and momentum mechanism.

[0090] Example 7 This embodiment provides an automatic impedance matching network tuning device, which implements the steps of the above method through hardware or a combination of hardware and software, thereby achieving a fast, stable, and universal automatic impedance matching effect. The device includes an impedance acquisition module, an error construction module, a gradient calculation module, a weight determination module, and a parameter update module.

[0091] The impedance acquisition module is configured to acquire the input impedance of the matching network, wherein the input impedance includes a resistive component and a reactance component.

[0092] In some implementations, the impedance acquisition module serves as the interface unit between the device and the RF front-end circuitry. In a practical hardware architecture, this module may include a directional coupler, a detector circuit, and an analog-to-digital converter (ADC). The directional coupler acquires the reflected signal from the input of the matching network, processes it through the detector circuit to obtain an analog voltage signal, and then converts it into a digital signal via the ADC before sending it to the processor. The impedance acquisition module inside the processor calculates the current complex impedance value, i.e., the resistance component R, based on these digital signals. in and reactance component X inIt should be understood that the specific implementation of the impedance acquisition module is not limited to this. For example, an integrated impedance analyzer chip or network analyzer module can also be used, as long as it can output the real and imaginary parts of the input impedance in real time.

[0093] The error construction module is configured to construct an error function based on the error between the input impedance and the target impedance.

[0094] In some implementations, the error construction module receives the current impedance data output by the impedance acquisition module and reads the pre-stored target impedance value (typically 50Ω). This module internally includes the calculation logic for the error function, in the form of a sum of squares as described in Example 5. The error construction module squares the error between the resistance component and the target impedance, squares the reactance component, and then superimposes the two to construct an error function value J that comprehensively reflects the current degree of mismatch. This error function value will serve as the benchmark for subsequent gradient calculations.

[0095] The gradient calculation module is configured to calculate the gradient of the error function with respect to the parameters of each adjustable element by means of numerical differentiation.

[0096] In some implementations, the gradient calculation module is the core hardware logic unit for realizing the "black box" tuning characteristics of this invention. This module integrates a numerical differentiation algorithm, as described in embodiments 3 and 4 above. In actual operation, the gradient calculation module generates a perturbation control signal, which, through the parameter update module, controls the adjustable elements to perform small positive and negative bidirectional perturbations. After each perturbation, the gradient calculation module reads the new error function value fed back by the error construction module and uses the central difference formula to calculate the partial derivatives of the error function with respect to the parameters of each adjustable element, i.e., the gradient. Since this process only requires measuring the error change before and after the perturbation, without needing to know the specific topology of the matching network, this device can adapt to various types of matching networks.

[0097] The weight determination module is configured to determine the learning rate weight of each adjustable element based on the error proportion of the resistive component and the error proportion of the reactance component.

[0098] In some implementations, a weight determination module is used to decouple the learning rate calculation logic, as described in Embodiment 2 above. This module first calculates the total error based on the absolute values ​​of the resistance component error and the reactance component error, and then normalizes the results to obtain the error proportions of the resistance component and the reactance component. Subsequently, the weight determination module allocates these error proportions to the corresponding adjustable components based on pre-stored sensitivity relationships between each adjustable component and the resistance and reactance components, thereby calculating an independent learning rate weight for each adjustable component. For example, for components sensitive to the real part, when the real part error proportion is large, this module will assign a larger learning rate weight to it, thereby accelerating real part correction. This hardware logic design effectively solves the problem of slow convergence caused by multivariable coupling.

[0099] The parameter update module is configured to update the parameter values ​​of each adjustable element based on the gradient and the learning rate weights.

[0100] In some implementations, the parameter update module is the execution output unit of the device. This module receives gradient data from the gradient calculation module and learning rate weight data from the weight determination module, and calculates the final parameter adjustment amount using a momentum mechanism (as described in Example 4). The parameter update module outputs control commands to adjust the physical parameter values ​​of the adjustable capacitor or adjustable inductor in the matching network through a drive circuit (such as a motor driver or voltage control circuit). During the iteration process, the parameter update module continuously adjusts the adjustable components until the error function value output by the error construction module is less than a preset threshold or the standing wave ratio reaches the target value.

[0101] Through the collaborative work of the above modules, the impedance matching network auto-tuning device provided in this embodiment integrates the numerical differential gradient calculation and the decoupled learning rate mechanism into the hardware logic, realizing universal tuning independent of the network topology model, and significantly improving the convergence speed and stability.

[0102] Example 8 This embodiment, based on Embodiment 7, provides a detailed description of the specific hardware implementation of an automatic tuning device for impedance matching networks.

[0103] In some implementations, the device includes a processor, a memory, and an adjustable element interface. The processor, acting as the core control unit, is connected to the memory via a data bus and executes the computer program stored in the memory to implement an automatic tuning method for an impedance matching network as described in any of Embodiments 1 to 6. The memory stores the aforementioned computer program and intermediate data generated during the tuning process, such as historical impedance values, historical adjustment amounts, and gradient data. The adjustable element interface is configured to connect to each adjustable element in the matching network and adjust the parameter values ​​of each adjustable element according to instructions from the parameter update module.

[0104] At the hardware architecture level, processors can be embedded chips of various types. For example, in applications with extremely high real-time requirements, such as radio frequency communication base stations, processors can be field-programmable gate arrays (FPGAs) or digital signal processors (DSPs). FPGAs can leverage their parallel processing advantages to complete numerical differentiation calculations and gradient update logic within extremely short clock cycles, thereby compressing tuning time to the microsecond level. In cost-sensitive or computationally-constrained scenarios (such as laboratory instruments or low-frequency radio frequency power supplies), processors can also be general-purpose microcontrollers (MCUs) or system-on-a-chip (SoCs), implementing the above tuning process through software algorithms. Memory can be random access memory (RAM), flash memory, or electrically erasable programmable read-only memory (EEPROM) used to store algorithm code and configuration parameters.

[0105] Adjustable component interfaces are crucial for converting digital control signals into analog circuit parameters. Specifically, an adjustable component interface typically includes a digital-to-analog converter (DAC) and a driver circuit. After the processor calculates the target capacitance value or adjustment amount of the adjustable capacitor, it transmits the digital value to the DAC, which converts it into a corresponding analog voltage signal. This analog voltage signal is then output to the adjustable component in the matching network via a driver circuit (such as an operational amplifier or a high-voltage driver chip). For example, if the matching network uses a voltage-controlled variable capacitor diode, the analog voltage output from the interface directly changes the junction capacitance of the diode; if the matching network uses a motor-driven variable capacitor or inductor, the control signal output from the interface drives a stepper motor to rotate, thereby changing the overlap area of ​​the capacitor plates or the position of the inductor core. Through this hardware architecture, the device can translate algorithm-level parameter update instructions into physical-level impedance characteristic changes, thus forming a closed-loop control circuit of "impedance sampling - error calculation - parameter update - physical adjustment," ensuring the feasibility of the technical solution in practical engineering.

[0106] Although this application has been described herein in conjunction with various embodiments, those skilled in the art, by reviewing the accompanying drawings, disclosure, and appended claims, will understand and implement other variations of the disclosed embodiments in carrying out the claimed application. In the claims, the word "comprising" does not exclude other components or steps, and "a" or "an" does not exclude multiple instances. A single processor or other unit can implement several functions listed in the claims. While different dependent claims may recite certain measures, this does not mean that these measures cannot be combined to produce good results.

[0107] Although this application has been described in conjunction with specific features and embodiments, it is obvious that various modifications and combinations can be made thereto without departing from the spirit and scope of this application. Accordingly, this specification and drawings are merely exemplary illustrations of this application as defined by the appended claims, and are considered to cover any and all modifications, variations, combinations, or equivalents within the scope of this application. Clearly, those skilled in the art can make various alterations and modifications to this application without departing from the spirit and scope of this application. Thus, if such modifications and modifications of this application fall within the scope of the claims of this application and their equivalents, this application is also intended to include such modifications and modifications.

Claims

1. An automatic tuning method for impedance matching networks, characterized in that, include: Obtain the input impedance of the matching network, wherein the input impedance includes a resistive component and a reactance component; Based on the error between the input impedance and the target impedance, an error function is constructed; The gradient of the error function with respect to the parameters of each adjustable element is calculated by numerical differentiation. The learning rate weights of each adjustable element are determined based on the error proportions of the resistive component and the reactive component. Based on the gradient and the learning rate weights, update the parameter values ​​of each adjustable element; The calculation of the gradient of the error function with respect to each adjustable element parameter using numerical differentiation includes: Apply positive and negative bidirectional perturbations to the current adjustable component parameters; Calculate the error function values ​​after applying positive and negative perturbations respectively; Based on the changes in the error function values ​​corresponding to the positive and negative disturbances, the partial derivatives of the error function with respect to the current adjustable element parameters are calculated.

2. The automatic tuning method for an impedance matching network according to claim 1, characterized in that, The step of determining the learning rate weight of each adjustable element based on the error proportions of the resistive component and the reactance component includes: Based on the sensitivity relationship between each adjustable element to the resistance component and the reactance component, the error proportion of the resistance component is allocated to the adjustable elements sensitive to the resistance component, and the error proportion of the reactance component is allocated to the adjustable elements sensitive to the reactance component.

3. The automatic tuning method for an impedance matching network according to claim 2, characterized in that, The error proportions of the resistance component and the reaction component are determined in the following way: Calculate the absolute value of the error between the resistance component and the target impedance to obtain the real part of the error; Calculate the absolute value of the reactance component; The total error is obtained by summing the absolute values ​​of the real part error and the reactance component. The ratio of the real part error to the total error is denoted as the error proportion of the resistance component; The ratio of the absolute value of the reactance component to the total error is denoted as the error percentage of the reactance component.

4. The automatic tuning method for an impedance matching network according to claim 1, characterized in that, The partial derivatives are calculated using the central difference formula: The partial derivative is equal to the difference between the first error value and the second error value, divided by twice the perturbation step size; wherein, the first error value is the error function value calculated after applying a positive perturbation based on the current adjustable element parameters, and the second error value is the error function value calculated after applying a negative perturbation based on the current adjustable element parameters.

5. The automatic tuning method for an impedance matching network according to claim 1, characterized in that, The updating of the parameter values ​​of each adjustable element also includes: Obtain the historical adjustment amount of each adjustable element in the previous iteration; The historical adjustment amount is weighted and summed with the adjustment amount of the current iteration to obtain the final adjustment amount; The parameter values ​​of each adjustable element are updated based on the final adjustment amount.

6. The automatic tuning method for an impedance matching network according to claim 5, characterized in that, The final adjustment amount is calculated using the following formula: The final adjustment is equal to the sum of the adjustment in the current iteration and the momentum coefficient multiplied by the historical adjustment in the previous iteration; The adjustment amount in the current iteration is equal to the learning rate weight multiplied by the negative value of the gradient.

7. The automatic tuning method for an impedance matching network according to claim 1, characterized in that, The error function includes: The square of the error between the resistance component and the target impedance; The square term of the reactance component.

8. An automatic tuning device for an impedance matching network, applied to the method as described in any one of claims 1-7, characterized in that, include: An impedance acquisition module is configured to acquire the input impedance of a matching network, wherein the input impedance includes a resistive component and a reactance component; The error construction module is configured to construct an error function based on the error between the input impedance and the target impedance; The gradient calculation module is configured to calculate the gradient of the error function with respect to the parameters of each adjustable element by means of numerical differentiation; The weight determination module is configured to determine the learning rate weight of each adjustable element based on the error proportion of the resistive component and the error proportion of the reactance component. The parameter update module is configured to update the parameter values ​​of each adjustable element based on the gradient and the learning rate weights.

9. A processing device, characterized in that, include: A processor configured to execute a program stored in memory to implement the method as described in any one of claims 1 to 7; The memory is configured to store the program.

Images