A radio frequency power control method based on model-free adaptive control

Abstract

This invention discloses a model-free adaptive control method for radio frequency (RF) power control, comprising: real-time acquisition and preprocessing of raw incident and reflected power; dynamic correction of the reference incident power using a piecewise function; calculation of the raw tracking error based on the reference incident power and the incident power, and calculation of the composite error signal by introducing a reflected power penalty term; estimation of the pseudo-gradient vector reflecting the instantaneous influence of each control quantity on the incident power using a projection algorithm based on historical input and output data; calculation of the increment of each control quantity based on the composite error signal and the pseudo-gradient vector; and calculation of the control quantity at the next moment based on the increment. This RF power control method does not require a precise mathematical model of the system; it achieves multi-input adaptive cooperative control by online estimation of the pseudo-gradient vector reflecting the influence of each actuator, effectively improving the system's control accuracy and response speed; and simultaneously achieving smooth and continuous protection of RF power.

Description

Technical Field

[0001] This invention relates to the field of radio frequency power control technology, and in particular to a radio frequency power control method based on model-free adaptive control, which is applicable to applications such as communication, radar, plasma excitation, semiconductor manufacturing, and radio frequency heating that require precise control of radio frequency power and where load characteristics are time-varying. Background Technology

[0002] The radio frequency (RF) power control system is a core component of RF generation equipment. Its basic task is to efficiently and stably transmit the energy generated by the RF signal source to a varying load (such as a plasma chamber or heating chamber) through a power amplifier and matching network. In actual operation, the load impedance often fluctuates nonlinearly and time-varyingly with temperature, pressure, process formulation, or ionization state, resulting in drastic changes in the incident and reflected power on the RF transmission line.

[0003] Existing RF power control solutions mainly suffer from the following technical defects: 1. Strong model dependence and poor adaptability: Traditional PID control relies on fixed parameters. When the load impedance changes and causes the dynamic characteristics of the system to change, the control performance will drop significantly, resulting in problems such as large power tracking error and excessively long settling time. 2. Delayed and fragmented protection mechanism: Conventional reflected power protection mostly adopts the hard protection method of "threshold comparison + logic shutdown". This method is a passive protection after the fact. The moment of action can easily cause power surge. Moreover, the protection logic and power control loop are independent of each other, and it is impossible to achieve smooth power transition during the protection process. 3. Difficulty in coordinating multiple actuators: Modern RF systems typically have multiple controllable actuators (such as power amplifiers and variable gain amplifiers), which together affect the incident power and are strongly coupled with each other; traditional single-input single-output (SISO) control strategies are difficult to effectively coordinate multiple inputs, resulting in the control potential not being fully utilized. Summary of the Invention

[0004] This application provides a radio frequency power control method based on model-free adaptive control, which solves the problems of poor model adaptability, protection power abrupt change, and difficulty in coordinating multiple actuators in the prior art, and realizes multi-input adaptive cooperative control and smooth and continuous power protection effect.

[0005] This application provides a model-free adaptive control-based radio frequency power control method, applied to radio frequency power systems, comprising the following steps: S1: Real-time acquisition of the raw incident power at time k via a directional coupler. and original reflected power The original sampled values ​​are filtered and outlier removed to obtain the preprocessed incident power. and reflected power ; S2: Based on the pre-processed reflection power The reference incident power is dynamically corrected using a piecewise function. The piecewise function divides the reflected power into a normal region, a transition region, and a protection region, and corrects the power for each region. S3: Based on the corrected reference incident power With the pre-processed incident power Calculate the original tracking error The composite error signal is calculated by introducing a reflection power penalty term. ; S4: Constructing the multi-input control vector Includes the primary AM modulated signal output to the digital gain multiplier. , Preamplifier gain of output to variable gain amplifier and the power amplifier gain output to the power amplifier Three control variables are used to estimate pseudo-gradient vectors reflecting the instantaneous influence of each control variable on the incident power using a projection algorithm based on historical input and output data. ; S5: Based on the composite error signal and pseudo gradient vector The incremental control variables are calculated using a pseudo-gradient-based proportional allocation strategy. ; S6: Based on the increment of each control quantity Calculate the control quantity at the next time step The output is then subjected to amplitude limiting.

[0006] The beneficial effects of the above embodiments are as follows: the radio frequency power control method does not require a precise mathematical model of the system. By estimating the pseudo gradient vector that reflects the influence of each actuator online, it can achieve adaptive and coordinated control of multiple inputs, which can effectively improve the control accuracy and response speed of the system. At the same time, the radio frequency power control method integrates power closed-loop control with reflection protection. When the reflected power changes abruptly, it can achieve instantaneous, continuous, and proportional suppression of power, avoiding power surges and system oscillations in traditional protection methods, and achieving smooth and continuous protection of radio frequency power.

[0007] Based on the above embodiments, this application can be further improved as follows: In one embodiment of this application, the filtering and outlier removal process in S1 includes: S1.1: Moving average filtering: The original sampled values ​​are processed using a moving average filter, and the calculation formula is as follows: ; ; in, Discrete-time index; : No. The original sampled value of the incident power at that moment; : No. Original sampled value of reflected power at any given time; : No. The incident power filter value at any given time; : No. Filter value of reflected power at any time; : Length of the sliding window; : Summation index variable; S1.2: Outlier Detection and Removal: If the relative deviation between the current raw sample value and the filter value from the previous time step exceeds the threshold, it is determined to be an outlier, and the filter value from the previous time step is used to replace the current filter value. ; ; in, : No. The incident power filter value at any given time; : No. Filter value of reflected power at any time; : Threshold coefficient for abnormal incident power detection; : Threshold coefficient for abnormal reflection power detection.

[0008] Technical benefits: By suppressing measurement noise through moving average filtering and avoiding control malfunctions caused by sampling interference through outlier removal, the reliability of input data is improved, providing accurate state feedback for subsequent control calculations.

[0009] In one embodiment of this application, the reference incident power in S2 The corrected formula is: ; in, The desired incident power set by the user; The safe threshold for reflected power; For the transition zone bandwidth; The coefficient for the square transition region; For deep protection linear coefficients; This is the depth protection bias coefficient.

[0010] Technical effect: Through three-level segmented dynamic correction of reference power, a graded protection effect of "precise control under normal operating conditions, smooth reduction when approaching the threshold, and rapid reduction when exceeding the threshold" is achieved, which not only ensures equipment safety but also maintains the continuity of power control and avoids power abrupt changes caused by traditional hard switching.

[0011] In one embodiment of this application, the composite error signal in S3 The calculation formula is: ; ; in, For the first Incident power tracking error at time t, The reference incident power determined for S2, This represents the incident power after preprocessing by S1. The reflection power suppression coefficient; This is the reflected power value after S1 preprocessing; This is the safe threshold for reflected power.

[0012] Technical effect: By directly embedding the reflected power penalty term into the error signal, the impact of excessive reflected power can be applied to the control calculation instantaneously without waiting for reference power correction. This complements the steady-state correction of S2, achieving dual protection of "steady-state planning + transient response" and improving the response speed to sudden changes in reflected power.

[0013] In one embodiment of this application, the pseudo gradient vector in S4 The estimation algorithm is as follows: First, define the control input vector and its changes: ; ; ; ; in, For the first Control the input vector at all times; , , The first The control quantity is constantly output to the digital gain multiplier, variable gain amplifier, and power amplifier; For the first The control input change vector changes at time -1; Step size factor; As a weighting factor; For the first The change in incident power at any given time.

[0014] Technical benefits: By estimating the pseudo gradient vector online, the dynamic influence of each control variable on the incident power can be grasped in real time without the need to establish a prior mathematical model of the system. It can automatically adapt to the drift of system characteristics caused by load changes, device aging, etc., and provide allocation basis for multi-actuator collaborative control.

[0015] In one embodiment of this application, the pseudo gradient vector The estimation process also includes a reset mechanism: when the magnitude of the pseudo-gradient vector is less than a threshold, or the sign of any pseudo-gradient changes, the reset mechanism is activated. Reset to initial value .

[0016] Technical effect: By resetting the mechanism, pseudo-gradient estimates caused by noise or coupling interference are avoided from deviating from their physical meaning, ensuring that the estimates are always within a reasonable range, thus improving the stability and reliability of the control algorithm.

[0017] In one embodiment of this application, the increment of the control quantity in S5 The calculation formula is: ; in: For the first The time calculated at time 1 The increment of a control quantity; To control the step size factor; For the first Time of the first The estimated values ​​of the pseudo gradients; As a weighting factor; It is the square norm of the pseudo-gradient vector; For the first The composite error signal at time t.

[0018] Technical effect: By adopting a pseudo-gradient-based proportional allocation strategy, the control variables with a large impact on the output undertake the main adjustment task, while the control variables with a small impact undertake the auxiliary task. The allocation ratio is determined entirely by the real-time estimated value under the current operating conditions, without the need for manual preset weights. This enables adaptive collaborative control of multiple actuators, effectively improving the control accuracy and response speed of the system.

[0019] In one embodiment of this application, the limiting process in S6 specifically includes: ; in: For the first The moment to be output One control quantity; For the first The maximum permissible output value of each control variable; For the first The minimum permissible output value of a control variable.

[0020] Technical effect: By limiting the amplitude, the output control quantity is ensured to meet the physical constraints of the actuator, avoiding equipment damage or control mismatch caused by the control quantity exceeding the actuator's range, thus improving the safety of system operation. Attached Figure Description

[0021] To more clearly illustrate the specific embodiments of the present invention or the technical solutions in the prior art, the accompanying drawings used in the description of the specific embodiments or the prior art will be briefly introduced below. In all the drawings, similar elements or parts are generally identified by similar reference numerals. In the drawings, the elements or parts are not necessarily drawn to scale.

[0022] Figure 1 This is a structural block diagram of the radio frequency power system in the embodiments of this application; Figure 2 This is a flowchart illustrating the steps of a model-free adaptive control-based radio frequency power control method in an embodiment of this application. Figure 3 This is a schematic diagram illustrating the dynamic correction of the reference incident power with respect to the change in reflected power in an embodiment of this application. Detailed Implementation

[0023] The present invention will be further illustrated below with reference to specific embodiments. It should be understood that these embodiments are for illustrative purposes only and are not intended to limit the scope of the invention. After reading the present invention, any modifications of the present invention in various equivalent forms by those skilled in the art will fall within the scope defined by the appended claims.

[0024] Example: A model-free adaptive control-based RF power control method is proposed and applied to RF power systems, such as... Figure 1 As shown, the system includes a digital signal source, a digital gain multiplier, a digital-to-analog converter, a variable gain amplifier, a power amplifier, a directional coupler, an RF matching network, a load, and a controller, all connected in sequence. The controller acquires the incident power and reflected power in real time and outputs three control quantities—the primary AM modulation signal, the preamplifier gain, and the power amplifier gain—to the digital gain multiplier, the variable gain amplifier, and the power amplifier, respectively.

[0025] like Figure 2 As shown, the radio frequency power control method includes the following steps: S1: Data Acquisition and Preprocessing: Real-time acquisition of the raw incident power at time k. and original reflected power The original sampled values ​​are filtered and outlier removed to obtain the preprocessed incident power. and reflected power .

[0026] This step involves real-time acquisition of incident power using a directional coupler. and reflected power To ensure data reliability, the raw sampled values ​​are filtered to suppress measurement noise and preprocessed. The sampling frequency should be at least 10 times the control cycle. Preprocessing specifically includes outlier detection and removal: S1.1: Moving average filtering; The original sampled values ​​are processed using a moving average filter, and its calculation formula is as follows: (1); (2); in: Discrete-time index; : No. The original sampled value of the incident power at that moment; : No. Original sampled value of reflected power at any given time; : No. The incident power filter value at any given time; : No. Filter value of reflected power at any time; : Length of the sliding window, ranging from 4 to 10; : Sum index variable.

[0027] S1.2: Outlier detection and removal; If the relative deviation between the current raw sample value and the filter value from the previous time step exceeds the threshold, it is determined to be an outlier, and the filter value from the previous time step is used to replace the current filter value. (3); (4); in: : No. At any given moment, the incident power filter value (W); : No. Time-reflected power filter value (W); Incident power anomaly detection threshold coefficient, ranging from 0.1 to 0.3; : The threshold coefficient for abnormal reflection power detection, which is 0.1 to 0.3.

[0028] S2: Dynamic correction of reference incident power based on reflection power constraint: based on the preprocessed reflection power The reference incident power is dynamically corrected using a piecewise function. The piecewise function divides the reflected power into a normal region, a transition region, and a protection region, and corrects the power for each region.

[0029] In RF power control systems, reflected power is a key indicator for measuring load matching and protecting equipment safety. Excessive reflected power can damage power amplifiers or matching networks due to excessive return energy. Existing solutions often employ hard threshold protection, directly shutting down the output or rapidly reducing power when the reflected power exceeds a set value. While this protects the equipment, it can easily lead to sudden power fluctuations, causing process disturbances or even system oscillations. To address this issue, this invention proposes a dynamic reference incident power correction method based on reflected power constraints. The core idea of ​​this method is: based on the current reflected power... The size of the reference value for dynamically adjusting the incident power. This allows the power control target to decrease smoothly as the reflected power increases, thus achieving a natural integration of protection actions and power control.

[0030] In this invention, a piecewise function design is specifically employed. The reflected power is divided into three intervals, and the reference incident power is calculated using the following formula. : (5); in: The desired incident power set by the user, in watts (W). The reflected power safety threshold is expressed in watts (W). The transition bandwidth, measured in watts (W), is used to define the width of the transition region. This is the square transition zone coefficient, in units of To control the curvature of power reduction within the transition region; The linear coefficient for deep protection is dimensionless and controls the rate of decrease of power within the protected area as reflected power increases. This is the deep protection bias factor, measured in watts (W), used to ensure the function operates within a certain range. The continuity at a point is usually taken as .

[0031] like Figure 3 As shown, this piecewise function divides the reflected power into three intervals: Normal area ( At this point, the reflected power is within a safe range, maintaining the user-set desired power. It remains unchanged and does not affect the control accuracy under normal operating conditions.

[0032] Transition zone ( When the reflected power is close to but has not yet exceeded the safety threshold, a square function is used to reduce the reference power from... The descent begins smoothly in the form of a quadratic curve. The advantage of this design is that the derivative is 0 at the beginning of the transition zone, ensuring smooth intervention of the protection action. At the same time, as the reflected power increases, the descent slope gradually increases, achieving progressive protection that "reduces speed as it gets closer to danger".

[0033] Nature Reserve ( When the reflected power exceeds the safety threshold, a linear function is used to make the reference power decrease rapidly with a fixed slope as the reflected power increases, thus achieving forced protection. (Coefficient) Control the descent rate, Guarantee the function in Continuity at that point.

[0034] Through the above segmented design, a graded protection effect of "precise control under normal operating conditions, smooth reduction when approaching the threshold, and rapid reduction when exceeding the threshold" is achieved, which not only ensures equipment safety but also maintains the continuity and smoothness of power control.

[0035] S3: Error signal calculation: based on the corrected reference incident power. Compared with the pre-processed actual incident power Calculate the original tracking error The composite error signal is calculated by introducing a reflection power penalty term. .

[0036] After obtaining the reference incident power Next, the error signal for the controller needs to be calculated. Traditional power control methods only use the raw tracking error. As a control basis, this means that the control loop can only sense the change in reflected power after the target value is changed through step S2, resulting in a certain response delay. To solve this problem, this invention introduces a reflected power penalty term based on traditional tracking error, constructing a composite error signal. This allows the influence of reflected power to directly affect control calculations without waiting for target value correction.

[0037] In this invention, the original power tracking error is first calculated: (6); in, For the first The incident power tracking error (W) at time t. The reference incident power (W) is determined in step S2. The actual incident power (W) after preprocessing in step S1.

[0038] Based on this, a reflection power penalty term is introduced to construct a composite error signal: (7); in For the first The composite error signal at each time step is used to replace the original tracking error for subsequent control calculations; The reflection power suppression coefficient ( This is used to adjust the suppression intensity of the control quantity when the reflected power exceeds the limit; The reflected power value after preprocessing in step S1; The safe threshold for reflected power; This means taking the larger of 0 and the reflected power exceeding the limit, meaning that a penalty is only applied when the reflected power exceeds the threshold.

[0039] The physical meaning of this composite error is: when the reflected power is within the safe range ( )hour, The composite error equals the original tracking error, and the system normally tracks the power target set by the user; when the reflected power exceeds the limit ( When the composite error is reduced by a penalty proportional to the over-limit, it is equivalent to the control target being "reduced in real time", causing the controller to actively reduce the output power within the current cycle.

[0040] Through this design, step S3 and step S2 complement each other: step S2 modifies the reference power. To achieve steady-state power planning, step S3 directly corrects the error signal through a penalty term to achieve rapid transient response. These two processes work together to ensure both steady-state power suppression when reflected power continuously exceeds limits and to provide instantaneous response capability to sudden changes in reflected power.

[0041] S4: Online estimation of pseudo-gradient vectors for multi-input systems: constructing multi-input control vectors Includes the primary AM modulated signal output to the digital gain multiplier. , Preamplifier gain of output to variable gain amplifier and the power amplifier gain output to the power amplifier Three control variables are used to estimate pseudo-gradient vectors reflecting the instantaneous influence of each control variable on the incident power using a projection algorithm based on historical input and output data. .

[0042] In multiple-input single-output (MISO) systems, multiple control variables jointly affect the same output, and complex coupling relationships exist between them. For RF power control systems, the parameters of the digital gain multiplier, variable gain amplifier, and power amplifier all affect the final incident power. However, traditional control methods often struggle to quantitatively describe the specific impact of each control variable on the output, let alone dynamically allocate control tasks based on this impact. Without accurately grasping the "influence" of each control variable, multi-input collaborative control is impossible. To address this problem, this invention introduces a pseudo-gradient vector to describe the real-time impact of each control variable on the output. Pseudo-gradient vector Used to characterize each control variable at the current moment. Incident power The instantaneous effect gain, i.e., "the first" "How much will the incident power change if the control variable is changed by one unit?" Due to the nonlinear time-varying characteristics of the RF system, the pseudo gradient changes with time and needs to be obtained in real time through online estimation.

[0043] In this invention, the control input vector and its changes are first defined: (8); (9); Simultaneously define the change in incident power: (10); in , , The first The control quantities output to the digital gain multiplier, variable gain amplifier, and power amplifier at all times are the primary AM modulation signal, preamplifier gain, and power amplifier gain.

[0044] Online estimation of pseudo-gradients employs a projection algorithm, the core idea of ​​which is to continuously update the estimated value using historical input-output data and error correction. The algorithm is as follows: (11); in: The step size factor is dimensionless and ranges from (0,1], controlling the convergence rate of parameter estimation. The weighting factor is dimensionless, takes a positive value, and is used to prevent the denominator from being zero and to suppress... Parameter drift caused by excessively small values; , which controls the square norm of the input change vector; , is the predicted value of the current power change based on the pseudo gradient of the previous time step, in watts (W).

[0045] The physical meaning of this formula is: (The part in parentheses is missing) This represents the prediction error between the "actually observed power change" and the "power change predicted based on the pseudo-gradient estimate from the previous time step." If the actual change is greater than the predicted value, it indicates that the pseudo-gradient estimate is too low and needs to be increased; conversely, it should be decreased. This determines how the correction amount is distributed across the three pseudo-gradients: the greater the change of a certain control variable in the previous step, the more correction amount it receives, reflecting the attribution logic that "whoever changes more contributes more to the error".

[0046] To prevent the estimated value from deviating from its physical meaning due to noise or coupling interference, this invention further introduces a reset mechanism: when the magnitude of the pseudo-gradient vector is too small ( Or the sign of a pseudo-gradient changes abruptly. When ), Reset to initial value This ensures that the estimated values ​​always remain within a reasonable range. To reset the threshold, For symbolic functions, For the first The initial values ​​of the pseudo gradients.

[0047] Using the online estimation method described above, the system can monitor the impact of each control variable on the incident power in real time. This dynamic allocation mechanism based on real-time data effectively solves the problem of coordinated control under multi-input coupling, and relies entirely on the system's own input and output data without the need to establish complex prior mathematical models.

[0048] S5: Calculation of multi-input control law based on composite error: Based on the composite error signal and pseudo gradient vector The incremental control variables are calculated using a pseudo-gradient-based proportional allocation strategy. .

[0049] In obtaining the composite error signal Then, it needs to be converted into control increments for the three actuators to achieve coordinated adjustment of multiple inputs. Traditional methods usually adopt a fixed weight allocation strategy, that is, manually setting the adjustment ratio of each control quantity. However, this method cannot adapt to the drift of system characteristics caused by load changes, and is prone to control mismatch or actuator saturation.

[0050] To address this problem, the present invention utilizes the pseudo gradient vector estimated online in step S4. The incremental control variables are calculated using a pseudo-gradient-based proportional allocation strategy. The core idea of ​​this strategy is that control variables with a large impact on the output are assigned the main adjustment task, while control variables with a small impact are assigned the auxiliary task. The allocation ratio is determined entirely by the real-time estimate under the current operating conditions, without the need for manual weight preset.

[0051] In this invention, the specific calculation method is as follows: (12); in: For the first The time calculated at time 1 The increment of a control quantity; To control the step size factor, dimensionless, , used to adjust the overall scaling of the control intensity; For the first Time of the first The estimated values ​​of the pseudo gradients; The weighting factor is dimensionless. This is used to limit the range of change in the control quantity and prevent drastic changes in the control quantity due to an excessively small pseudo-gradient estimate. , is the squared norm of the pseudo-gradient estimation vector; For the first The composite error signal at time t is expressed in watts (W).

[0052] This control law shows that the pseudo-gradient... They appear simultaneously in both the numerator and denominator: the numerator determines the relative allocation ratio of each control variable, while the denominator plays a role in overall gain adjustment. This structure allows the allocation of control variables to be entirely determined by the current input-output characteristics of the system, achieving true adaptive coordination.

[0053] It should be noted that the composite error in this control law The reflection power penalty term has been incorporated in step S3, so when the reflection power exceeds the limit, The power will decrease instantaneously, and the increments of all control variables will decrease synchronously, allowing the system to begin reducing output power within the current control cycle. This mechanism complements the target value correction in step S2: step S2 achieves steady-state power planning by modifying the reference power, while step S5 achieves rapid transient response by penalizing errors. Together, they achieve all-time reflection protection.

[0054] Through the above method, the present invention realizes adaptive cooperative control and real-time reflection protection of multi-input systems. Without the need for manual weight tuning, each actuator dynamically divides its work according to its real-time influence, while ensuring the continuity and smoothness of the control process.

[0055] S6: Control Variable Update and Output Limiting: Based on the increment of each control variable... Calculate the control quantity at the next time step The output is then subjected to amplitude limiting.

[0056] Calculate the control input at the next time step: (13); in: For the first The moment to be output One control quantity; For the first The actual output at time 1 One control quantity; For the first The time is calculated by step S5. Increment of control quantity.

[0057] Considering the physical constraints of the actuator, the final output control quantity is limited: (14); in: For the first The maximum permissible output value of each control variable; For the first The minimum permissible output value of each control variable; and These represent operations to retrieve the minimum and maximum values, respectively.

[0058] The technical solutions described in the embodiments of this application have at least the following technical effects or advantages: 1. Model-free adaptive and robust: It does not rely on the precise mathematical model of the RF system. By estimating pseudo-gradients online, it automatically adapts to changes in system characteristics caused by load changes, device aging, etc., and the control parameters do not need to be frequently tuned manually.

[0059] 2. Multi-input collaborative control, fully utilize potential: It solves the coupling problem between multiple actuators and realizes dynamic task allocation based on real-time influence, which can effectively improve the control accuracy and response speed of the system.

[0060] 3. Deep integration of protection and control, safe and smooth: The reflection protection mechanism is creatively embedded in the MFAC control law, forming a dual protection of "steady-state planning + transient response" with the traditional target value correction. When the reflected power changes abruptly, it can achieve instantaneous, continuous, and proportional suppression of power, avoiding power surges and system oscillations of traditional protection methods.

[0061] 4. Simple algorithm and easy to implement in engineering: The entire control algorithm is based on input and output data, with a clear structure, low computational load, and is easy to implement in real time on embedded platforms or FPGA / DSP.

[0062] 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 radio frequency power control method based on model-free adaptive control, characterized in that: For application in radio frequency power systems, the following steps are included: S1: Real-time acquisition of raw incident power and original reflected power The original sampled values ​​are filtered and outlier removed to obtain the preprocessed incident power. and reflected power ; S2: Based on the pre-processed reflection power The reference incident power is dynamically corrected using a piecewise function. The piecewise function divides the reflected power into a normal region, a transition region, and a protection region, and corrects the power for each region. S3: Based on the corrected reference incident power With the pre-processed incident power Calculate the original tracking error The composite error signal is calculated by introducing a reflection power penalty term. ; S4: Constructing the multi-input control vector Includes the primary AM modulated signal output to the digital gain multiplier. , Preamplifier gain of output to variable gain amplifier and the power amplifier gain output to the power amplifier Three control variables are used to estimate pseudo-gradient vectors reflecting the instantaneous influence of each control variable on the incident power using a projection algorithm based on historical input and output data. ; S5: Based on the composite error signal and pseudo gradient vector The incremental control variables are calculated using a pseudo-gradient-based proportional allocation strategy. ; S6: Based on the increment of each control quantity Calculate the control quantity at the next time step The output is then subjected to amplitude limiting.

2. The radio frequency power control method according to claim 1, characterized in that: The filtering and outlier removal process in S1 includes: S1.1: Moving average filtering: The original sampled values ​​are processed using a moving average filter, and the calculation formula is as follows: ； ； in, Discrete-time index; : No. The original sampled value of the incident power at that moment; : No. Original sampled value of reflected power at any given time; : No. The incident power filter value at any given time; : No. Filter value of reflected power at any time; : Length of the sliding window; : Summation index variable; S1.2: Outlier Detection and Removal: If the relative deviation between the current raw sample value and the filter value from the previous time step exceeds the threshold, it is determined to be an outlier, and the filter value from the previous time step is used to replace the current filter value. ； ； in, : No. The incident power filter value at any given time; : No. Filter value of reflected power at any time; : Threshold coefficient for abnormal incident power detection; : Threshold coefficient for abnormal reflection power detection.

3. The radio frequency power control method according to claim 1, characterized in that: The reference incident power in S2 The corrected formula is: ； in, The desired incident power set by the user; The safe threshold for reflected power; For the transition zone bandwidth; The coefficient for the square transition region; For deep protection linear coefficients; This is the depth protection bias coefficient.

4. The radio frequency power control method according to claim 1, characterized in that: The composite error signal in S3 The calculation formula is: ； ； in, For the first Incident power tracking error at time t, The reference incident power determined for S2, This represents the incident power after preprocessing by S1. The reflection power suppression coefficient; This is the reflected power value after S1 preprocessing; This is the safe threshold for reflected power.

5. The radio frequency power control method according to claim 1, characterized in that: The pseudo gradient vector in S4 The estimation algorithm is as follows: First, define the control input vector and its changes: ； ； （10）； ； in, For the first Control the input vector at all times; For the first The control input change vector changes at time -1; Step size factor; As a weighting factor; For the first The change in incident power at any given time.

6. The radio frequency power control method according to claim 1, characterized in that: The pseudo gradient vector The estimation process also includes a reset mechanism: when the magnitude of the pseudo-gradient vector is less than a threshold, or the sign of any pseudo-gradient changes, the reset mechanism is activated. Reset to initial value .

7. The radio frequency power control method according to claim 1, characterized in that: The increment of the control quantity in S5 The calculation formula is: ； in: For the first The first time calculated at time 1 The increment of a control quantity; To control the step size factor; For the first Time of the first The estimated values ​​of the pseudo gradients; As a weighting factor; It is the square norm of the pseudo-gradient vector; For the first The composite error signal at time t.

8. The radio frequency power control method according to claim 1, characterized in that: The specific clipping process in S6 is as follows: ； in: For the first The moment to be output One control quantity; For the first The maximum permissible output value of each control variable; For the first The minimum permissible output value of a control variable.

9. The radio frequency power control method according to claim 1, characterized in that: The radio frequency power system includes a digital signal source, a digital gain multiplier, a digital-to-analog converter, a variable gain amplifier, a power amplifier, a directional coupler, an radio frequency matching network, a load, and a controller connected in sequence. The controller collects the incident power and reflected power in real time and outputs three control quantities—primary AM modulation signal, preamplifier gain, and power amplifier gain—to the digital gain multiplier, the variable gain amplifier, and the power amplifier, respectively.

Images