Satellite antenna dynamic pointing calibration method and system based on on-board autonomous gradient optimization

Abstract

The application discloses a satellite antenna dynamic pointing calibration method and system based on on-board autonomous gradient optimization, relates to the technical field of satellite communication, and realizes closed-loop calibration without ground intervention by processing signal signal-to-noise ratio (SNR) data uploaded by a ground station in real time on board, combining a multi-physical field coupling model of temperature and stress sensors, and reducing pointing error in a dynamic environment to less than 0.3° from more than 2.0° in a traditional scheme.

Description

Technical Field

[0001] This invention relates to the field of satellite communication technology, specifically to a method and system for dynamic pointing calibration of satellite antennas based on on-board autonomous gradient optimization. Major applications include continuous beam alignment of geostationary orbit (GEO) communication satellites, enhanced stability of data transmission links between low Earth orbit (LEO) remote sensing satellites and the Earth, and autonomous pointing maintenance of deep space probes in long-latency communication environments. In the military field, it can improve anti-jamming communication capabilities; in the civilian field, it is suitable for multi-beam scheduling of high-throughput satellites (HTS), and is particularly suitable for compensating for pointing deviations caused by periodic thermal deformation (typically ±80°C) of the satellite platform due to solar radiation cycles. Background Technology

[0002] Existing satellite antenna pointing calibration technologies have systemic shortcomings, primarily manifested in the contradiction between static calibration dependence and dynamic environment mismatch. Taking the Extended Kalman Filter (EKF) scheme used by the Milstar satellite as an example, its calibration process relies on three dedicated ground stations (such as those in California and Ohio) located at different longitudes. It collects 136 sets of power difference data over 48 hours using a dual concentric circle scanning mode and establishes a periodic error model using Fourier harmonic decomposition. While this scheme can reduce the static error to 0.03°, it suffers from three fundamental flaws:

[0003] First, the hardware costs are high, requiring the deployment of high-precision power measurement terminals (costing over $500,000 per station) and optical auxiliary equipment such as prisms, which increases the system complexity by more than 40%.

[0004] Secondly, its dynamic adaptability is insufficient. Its Fourier model can only compensate for regular thermal deformation within a 24-hour period (achieved through 21 harmonic coefficients), and cannot handle sudden mechanical stress deformation (such as that caused by the impact of solar panel deployment). strain);

[0005] Third, it has limitations in real-time performance. The generation of calibration instructions relies entirely on calculations by the ground station, which makes it ineffective in scenarios such as deep space exploration where communication delays exceed 20 minutes.

[0006] In addition, the traditional scheme uses the dual-circle scanning power difference method ( ,in, The power difference is the core measurement quantity in the traditional dual-circle scanning scheme; This is the measured transmit power of the antenna along its outer scanning circular trajectory. While the measured transmit power (the value of the antenna's transmit power on the inner scanning circular trajectory) can eliminate common-mode interference caused by atmospheric attenuation, it requires strictly synchronized data acquisition, and the clock synchronization accuracy of the ground station needs to reach the microsecond level (error tolerance). In practice, ionospheric disturbances often lead to... The signal-to-noise ratio degrades by more than 6dB. More seriously, the existing technology lacks on-board autonomous decision-making capabilities. Each calibration requires re-uploading the entire set of Fourier coefficients (approximately 21×3=63 parameters), occupying a data transmission bandwidth of up to 150kbps. Real-time updates are difficult to complete when the LEO satellite transit window is short (usually <10 minutes). Summary of the Invention

[0007] To address the issue of decreased pointing accuracy of spaceborne parabolic antennas during on-orbit operation due to thermal deformation, mechanical stress, and attitude drift, an autonomous and highly dynamic adaptive satellite antenna dynamic pointing calibration method and system based on on-board autonomous gradient optimization is proposed. By processing signal-to-noise ratio (SNR) data uploaded from ground stations in real time on-board and combining it with a multi-physics coupling model of temperature and stress sensors, closed-loop calibration without ground intervention is achieved, reducing the pointing error in dynamic environments from over 2.0° in traditional methods to within 0.3°.

[0008] The technical solution of this invention is: a dynamic pointing calibration method for satellite antennas based on on-board autonomous gradient optimization, the specific steps of which are as follows:

[0009] Step 1: Data Collection

[0010] (11) During the satellite's operation in orbit, the ground station continuously receives satellite signals and records the received signal quality parameters each time it passes over the satellite. The signal quality parameters include, but are not limited to, the signal-to-noise ratio (SNR). The ground station sends the collected SNR information to the satellite via the uplink.

[0011] (12) While receiving signal-to-noise ratio information, the satellite records antenna stress and temperature information through mechanical and thermal sensors at its own antenna.

[0012] Step 2, Error Analysis

[0013] (21) Using the signal quality parameters received by the ground station and the satellite attitude and orbit information, establish an antenna pointing error model:

[0014]

[0015] in, This is the current signal-to-noise ratio measurement. This represents the theoretical maximum signal-to-noise ratio, i.e., the optimal value when the antenna is aligned. This refers to the pitch angle pointing error; This refers to the yaw angle pointing error; To account for the overall pointing error angle; This is the pattern attenuation function;

[0016] (22) The satellite actively applies a small perturbation once in each of the four orthogonal directions along the antenna pointing axis in the elevation and azimuth planes; based on a nonlinear mapping model Calculate the gradient vector of the current pointing deviation. ;in, The signal-to-noise ratio gradient vector; This is the partial derivative of the signal-to-noise ratio with respect to the pitch angle; This is the partial derivative of the signal-to-noise ratio with respect to the yaw angle;

[0017] Achieve this through the finite difference method:

[0018]

[0019] in, For the current moment The signal-to-noise ratio measurement value; The time of the previous measurement Signal-to-noise ratio; For time intervals; This refers to the actively applied pitch angle disturbance.

[0020] Introducing a sliding time window mechanism, Dynamically adjusted to ,in, Based on the sampling interval, in seconds It is determined based on the satellite attitude control cycle; This is the window width adjustment factor, in units of ,control The sensitivity to gradient changes allows for automatic shortening of the sampling interval when the gradient is large, thereby improving the convergence speed.

[0021] (23) By analyzing the relationship between signal quality parameters and satellite attitude and antenna force and heat parameters, the magnitude and direction of the current antenna pointing error can be determined;

[0022] (24) When the satellite actively performs attitude disturbances, if the received signal quality parameters show a significant decrease, it indicates that the antenna pointing may have deviated from the direction of the ground station.

[0023] Step 3: Error Correction

[0024] (31) Based on the error analysis results, the antenna pointing error is gradually reduced by adjusting the satellite attitude, the angle of the antenna rotating arm, and the angle of the feed electric scanning.

[0025] (32) After each adjustment, collect the signal quality parameters received by the ground station again to evaluate the correction effect;

[0026] (33) Through iterative adjustments, the antenna pointing error is eventually brought to within the allowable range, achieving high-precision pointing;

[0027] Step 4: Calibration Completed

[0028] (41) When the received signal quality parameters reach the expected index and remain stable at a high level, the antenna pointing calibration is considered complete;

[0029] (42) Record the final satellite attitude and antenna rotation arm angle parameters as a reference for subsequent antenna pointing control.

[0030] Furthermore, in step three, error correction, adaptive gradient descent correction is adopted, and the iterative formula is designed as follows:

[0031]

[0032]

[0033] in, For the first The pitch angle correction for the next iteration; For the first Yaw angle correction for the next iteration; This is the learning rate coefficient, with a value ranging from 0.1 to 0.3. The attenuation coefficient;

[0034] Exponential decay term The introduction of this feature makes the iteration step size increase with the number of iterations. The method of increasing and adaptively decreasing ensures that the solution is quickly approached in the early stages, while avoiding oscillations near the extreme points in the later stages.

[0035] Furthermore, in step three, error correction, dynamic error compensation is adopted by fusing temperature sensor data in real time. and strain gauge output Establish a physical-driven compensation quantity:

[0036]

[0037] in, For structural feature length, , These are the elastic modulus and the shear modulus, respectively. , These are the coefficient of thermal expansion and the strain sensitivity coefficient of the material. This represents the total number of temperature sensors. This represents the number of measurement points for the strain gauge. For the first The equivalent area of ​​each strain point.

[0038] Furthermore, in adaptive gradient descent correction, the gradient vector is updated at a frequency of 10Hz; in dynamic error compensation, the physical field parameters are refreshed at a frequency of 1Hz; and a dual-rate mechanism is used to balance the computational load and response speed.

[0039] Furthermore, the temperature change of the antenna structure Mechanical stress variation measured using a spaceborne PT1000 platinum resistance sensor. Data was collected by a Wheatstone bridge strain gauge.

[0040] The present invention also provides a satellite antenna dynamic pointing calibration system based on on-board autonomous gradient optimization for implementing the aforementioned satellite antenna dynamic pointing calibration method based on on-board autonomous gradient optimization. It constructs a hierarchical data processing architecture, and its hardware carrier is a multi-core processor module integrated into the satellite integrated electronic system. The multi-core processor module includes four parts: a data interface unit, a gradient calculation unit, a physical field compensation unit, and an instruction synthesis unit.

[0041] The data interface unit is connected to the satellite management system via the SpaceWire bus, and receives downlink signal-to-noise ratio (SNR) telemetry data packets from the ground station in real time. At the same time, it collects analog signals from temperature sensors and strain gauges in key locations via the CAN bus, and converts them into digital sequences after conversion by a 24-bit ADC. The satellite actively applies small perturbations in four directions in the elevation and azimuth planes, that is, perturbations are applied to the four orthogonal directions of the satellite antenna axis, and calculates the gradient based on the changes in the received SNR values.

[0042] The physical field compensation unit has a built-in satellite structural parameter database that stores the thermal expansion coefficients of each component. Elastic modulus When the temperature change is received With strain change At that time, the directional shift caused by deformation is calculated in real time based on the cantilever beam deformation model.

[0043] The beneficial effects of this invention are: First, addressing the high cost of ground calibration equipment, a spatial geometric mapping model based on downlink signal-to-noise ratio (SNR) inversion is proposed, utilizing... The nonlinear relationship replaces traditional power difference measurement, allowing ground stations to provide only raw SNR data without undertaking complex calculations, reducing ground facility costs by over 90%; among other things, Signal-to-noise ratio (SNR) of the signal received by the ground station. The antenna pointing error angle (unit: degrees) includes elevation and azimuth components; The antenna structure temperature change (unit: °C) was measured using a spaceborne PT1000 platinum resistance sensor. Mechanical stress change (unit: microstrain) The data was collected by a Wheatstone bridge strain gauge.

[0044] Secondly, a multi-physics joint compensation model is established to address the coupling interference between diurnal periodic thermal deformation and random mechanical stress. By fusing real-time data from onboard temperature and strain sensors, frequency-time domain coordinated suppression of dynamic errors is achieved, enabling... The pointing deviation within the temperature range remained stable within 0.3°; among which, Multiphysics coupling compensation angle (unit: degrees); Thermal deformation compensation coefficient (unit: ), and the coefficient of thermal expansion of the material Related; The rate of change of temperature over time (unit: ); Stress compensation coefficient (unit: (), determined by structural modal parameters; The rate of change of mechanical stress over time (unit: ).

[0045] Finally, to address the control instability problem in long-delay scenarios such as deep space exploration, a progressive gradient descent algorithm with historical data correlation capabilities is designed. By introducing an exponential decay factor Balancing convergence speed with overshoot risk, a convergence accuracy of 0.5° is maintained even with a 20-minute communication delay. The angle correction amount for the kth iteration (unit: degrees); This is the gradient descent step size coefficient (dimensionless), and its value is related to the antenna beamwidth. The signal-to-noise ratio gradient vector. , Let be the partial derivative of the signal-to-noise ratio with respect to the pitch angle. This is the partial derivative of the signal-to-noise ratio with respect to the yaw angle; It is an exponential decay factor; It is a time-varying decay term, which enables adaptive step size control.

[0046] Compared to the Milstar solution, this technology has three main advantages: First, by replacing the ground-based computing center with an onboard embedded processing unit, the latency for generating calibration commands is reduced from hours to milliseconds; second, a lightweight parameter update mechanism is adopted, requiring only a few data transfers per iteration. The incremental data (approximately 10 bytes) reduces bandwidth requirements by 99% compared to traditional Fourier coefficient uploads.

[0047] At the structural design level, this solution reconstructs the hardware architecture of traditional calibration systems through deep integration of onboard processing units. Traditional solutions rely on a measurement network composed of three dedicated ground stations distributed at different longitudes, each requiring a high-precision power meter and clock synchronization device, leading to an exponential increase in system complexity. This solution migrates the computational core to an onboard multi-core processor, utilizing the existing interface resources of the Attitude Determination System (ADS) to directly acquire gyroscope and star-sensor data via the SpaceWire bus. Simultaneously, it reuses the existing temperature and strain sensor network on the satellite platform, reducing the cost of new hardware by 92%. The key to structural optimization lies in designing a lightweight data pipeline. After aligning the raw SNR data stream (50kbps) and sensor data stream (10kbps) with onboard timestamps, they are input into a shared memory buffer, where a dual-core processor performs gradient calculations and physics compensation respectively. The computational latency is compressed from minutes in traditional ground processing to milliseconds.

[0048] In terms of functionality and performance, this solution achieves two breakthroughs through the synergistic effect of gradient descent algorithm and multiphysics model. Firstly, the calibration accuracy reaches 0.3° in dynamic environments, a six-fold improvement over traditional methods, thanks to the nonlinear mapping model. High-fidelity approximation of the antenna pattern, and temperature compensation term. First, it provides accurate prediction of thermal deformation. Second, it significantly enhances resistance to communication delays. In deep space exploration scenarios, even with a 20-minute SNR data delay, it can still achieve accurate predictions through a historical data weighting algorithm. It can still maintain a convergence accuracy of 0.5°, where the weighting function Decays exponentially based on temporal proximity. Attached Figure Description

[0049] Figure 1 Flowchart of the dynamic pointing calibration method for satellite antennas;

[0050] Figure 2 This is a diagram showing the system connection relationships. Detailed Implementation

[0051] The present invention will now be further described with reference to the accompanying drawings.

[0052] The satellite antenna dynamic pointing calibration method based on on-board autonomous gradient optimization includes three main stages: satellite-ground collaborative data acquisition, on-board autonomous error analysis, and multi-physics field coupling compensation. It achieves continuous optimization of antenna pointing by constructing a closed-loop control architecture.

[0053] Step 1: Data Collection

[0054] During the satellite's operation in orbit, the ground station continuously receives satellite signals and records the received signal quality parameters at each pass, including but not limited to the signal-to-noise ratio (SNR). The ground station then transmits the collected SNR values ​​to the satellite via the uplink.

[0055] While receiving signal-to-noise ratio information, the satellite also records antenna stress and temperature information through mechanical and thermal sensors located at its antenna.

[0056] Step 2: Error Analysis

[0057] An antenna pointing error model is established using the signal quality parameters received by the ground station and the satellite attitude and orbit information.

[0058] The satellite actively applies a small perturbation once in each of the four orthogonal directions along the antenna pointing axis in the pitch and azimuth planes.

[0059] By analyzing the relationship between signal quality parameters and satellite attitude and antenna force and heat parameters, the magnitude and direction of the current antenna pointing error can be determined.

[0060] If the received signal quality parameters decrease significantly when the satellite actively performs attitude disturbances, it indicates that the antenna pointing may have deviated from the direction of the ground station.

[0061] Step 3: Error Correction

[0062] Based on the error analysis results, the antenna pointing error was gradually reduced by adjusting the satellite attitude, the angle of the antenna rotating arm, and the angle of the feed electric scanning.

[0063] After each adjustment, the signal quality parameters received by the ground station are collected again to evaluate the correction effect.

[0064] Through iterative adjustments, the antenna pointing error was eventually brought within an acceptable range, achieving high-precision pointing.

[0065] Step 4: Calibration complete

[0066] When the received signal quality parameters reach the expected level and remain stable at a high level, the antenna pointing calibration is considered complete.

[0067] Record the final satellite attitude and antenna rotation arm angle parameters as a reference for subsequent antenna pointing control.

[0068] The specific implementation process is as follows: Figure 1 , 2 As shown, the system first receives the satellite downlink signal from the ground station and calculates the signal-to-noise ratio (SNR) parameter. After encapsulation using the CCSDS telemetry protocol, the signal is uploaded to the onboard processing unit, which simultaneously records the current attitude angle (pitch angle). Yaw angle Data from temperature / stress sensors, along with other sensors, forms a spatiotemporally correlated raw dataset. This is followed by the error direction identification stage, where the onboard embedded processor uses a nonlinear mapping model. Calculate the gradient vector of the current pointing deviation. ,in, This is the current signal-to-noise ratio measurement (dimensionless, usually expressed in dB); This represents the theoretical maximum signal-to-noise ratio (the optimal value when the antenna is aligned). This refers to the pitch angle pointing error (unit: degrees, in satellite coordinate system). Yaw angle pointing error (unit: degrees, in satellite coordinate system); To synthesize the pointing error angle (equivalent space vector magnitude); This is the pattern attenuation function (derived from the cosine square approximation of the antenna pattern); Signal-to-noise ratio gradient vector (unit: ); This is the partial derivative of the signal-to-noise ratio with respect to the pitch angle; This is the partial derivative of the signal-to-noise ratio with respect to the yaw angle.

[0069] This process is achieved using the finite difference method:

[0070]

[0071] in, For the current moment The signal-to-noise ratio measurement value; The time of the previous measurement Signal-to-noise ratio; The time interval is in seconds. The pitch angle disturbance is actively applied (in degrees), typically 0.1°.

[0072] The improvement lies in introducing a sliding time window mechanism, which... Dynamically adjusted to (in, The basic sampling interval, in seconds (s), is determined based on the satellite attitude control cycle. This is the window width adjustment factor, in units of ,control (Sensitivity to gradient changes) enables the automatic shortening of the sampling interval to improve convergence speed when the gradient is large.

[0073] The core step is adaptive gradient descent correction, and its iterative formula is designed as follows:

[0074]

[0075]

[0076] in, The pitch angle correction for the k-th iteration (unit: degrees); The yaw angle correction for the k-th iteration (in degrees); The learning rate coefficient (dimensionless) ranges from 0.1 to 0.3 (matching the antenna beamwidth). This is the attenuation coefficient (unit: 1 / time).

[0077] The essential difference from the traditional gradient method lies in the exponential decay term. The introduction of this factor allows the iteration step size to adaptively decrease as the number of iterations k increases, ensuring rapid approximation of the optimal solution in the early stages while avoiding oscillations near extreme points in the later stages. Experimental data show that when The number of convergence steps was reduced by 40% and the overshoot was reduced to below 0.02°.

[0078] Another innovation is the dynamic error compensation module, which integrates temperature sensor data in real time. and strain gauge output Establish a physical-driven compensation quantity:

[0079]

[0080] in, For structural feature length, , These are the elastic modulus and the shear modulus, respectively. , These are the coefficient of thermal expansion and the strain sensitivity coefficient of the material. This represents the total number of temperature sensors. This represents the number of measurement points for the strain gauge. For the first The model calculates the deformation effect in real time using a pre-stored finite element parameter library of the satellite structure, and then superimposes this value with the gradient descent output to form the final attitude correction command.

[0081] Data exchange between steps is achieved via the onboard bus. The error identification module updates the gradient vector at a frequency of 10Hz, while the compensation module refreshes the physical field parameters at a frequency of 1Hz. A dual-rate mechanism balances computational load and response speed. Compared to the single-batch processing mode of the Milstar scheme, the streaming processing architecture of this design reduces computational latency from minutes to sub-seconds, meeting the real-time requirements of the LEO satellite transit window.

[0082] The principle of this scheme is as follows: Feedback substitution: After the satellite signal is received by the ground station, the ground station measures the signal-to-noise ratio (SNR) and transmits it back to the satellite. The satellite uses the SNR value as the sensor value and optimizes the algorithm to maximize the SNR, thereby indirectly correcting the pointing error; Dynamic compensation: Real-time fusion of thermal, mechanical, and inertial sensor data to predict and offset the multi-physics coupling effect.

[0083] The core idea of ​​this scheme is to construct an on-board autonomous closed-loop control system, which dynamically eliminates antenna pointing errors through multi-source data fusion and a compensation model driven by physical mechanisms. Its theoretical framework includes three core modules: error modeling, gradient optimization, and multi-physics coupling, forming a complete error suppression chain.

[0084] The total pointing error can be decomposed into:

[0085] in:

[0086] Static assembly error: It follows a Gaussian distribution with zero mean;

[0087] Thermally induced deformation error: Modeling based on thermoelastic theory In the formula For structural feature length, Let the moment of inertia of the cross section be... For the first Temperature rise at each temperature measurement point.

[0088] Mechanical stress error: expressed by modal superposition method ,in For the first First natural frequency, This refers to the modal participation factor.

[0089] Time-varying drift error: , It is a random walk process.

[0090] The error elimination mechanism is as follows:

[0091] 1. Gradient optimization module, employing an improved gradient descent algorithm:

[0092]

[0093] Where: exponential decay term To achieve variable step size control while satisfying the Lyapunov stability condition. (H is the Hessian matrix).

[0094] Symbolic function terms : Suppress local extremum traps and improve global convergence.

[0095] 2. Multiphysics compensation module, constructing a deformation predictor:

[0096]

[0097] This model achieves real-time deformation prediction by preloading finite element parameters of the satellite structure, and the compensation accuracy is improved compared with the traditional Fourier method.

[0098] This solution constructs a hierarchical data processing architecture on-board, with its hardware carrier being a multi-core processor module integrated into the satellite's integrated electronic system. This module comprises four parts: a data interface unit, a gradient calculation unit, a physical field compensation unit, and an instruction synthesis unit. The data interface unit connects to the satellite management system via the SpaceWire bus, receiving real-time downlink signal-to-noise ratio (SNR) telemetry data packets (format conforming to CCSDS 132.0-B-2 standard) from ground stations. Simultaneously, it acquires data via the CAN bus from temperature sensors (PT1000 platinum resistance thermometers, high accuracy) distributed in key locations such as the antenna support and feedhorn. ) and strain gauges (Wheatstone bridge, range) The analog signal is converted into a digital sequence by a 24-bit ADC. The satellite actively applies small perturbations in four directions in the elevation and azimuth planes, that is, perturbations in the four orthogonal directions of the satellite antenna axis, and calculates the gradient based on the changes in the received SNR value.

[0099] The physics compensation unit has a built-in satellite structural parameter database that stores the thermal expansion coefficients of each component. Elastic modulus Material properties, when receiving temperature changes With strain change At that time, the directional shift caused by deformation is calculated in real time based on the cantilever beam deformation model:

[0100]

[0101] This formula originates from the theory of beam bending deformation in mechanics of materials, where... For characteristic length, Let be the moment of inertia of the cross section. For the dynamic stress disturbance caused by solar panel vibration, frequency domain decomposition is performed using the modal superposition method:

[0102]

[0103] In the formula The natural frequency of order k is... This represents the mode shape participation factor. The outputs of the gradient calculation unit and the physics compensation unit are vector-superimposed in the command synthesis unit to generate the total correction. The signal is sent to the attitude control computer via the 1553B bus, which drives the reaction flywheel or magnetic torque converter to perform pointing adjustment.

[0104] The fundamental breakthrough in the core steps involved in the improvement is reflected in two aspects: first, the exponential decay factor introduced in the gradient descent algorithm. Its mathematical essence is to embed time-varying gain in the iteration step size, so that the algorithm in the early stage ( When the value is relatively small, maintain a large step size to quickly approach the extreme point, while in the later stages ( (As the value increases) the step size is automatically reduced to suppress overshoot. This design is based on Lyapunov stability theory, and is achieved by constructing an energy function. It can be proven that when ( When the Lyapunov coefficients are used, the system is globally asymptotically stable. Secondly, the multiphysics compensation model transforms the pure mathematical fitting of traditional Fourier series into prediction driven by physical mechanisms. It directly correlates temperature / stress changes with mechanical deformation through the material constitutive equation, making it possible to predict non-periodic disturbances (such as those caused by micrometeoroid impacts). Even under conditions of sudden disturbances, the model can still maintain compensation accuracy. Experiments show that the model can reduce pointing error by 67% under sudden disturbances.

[0105] The above description is only a preferred embodiment of the present invention. It should be noted that for those skilled in the art, several improvements and modifications can be made without departing from the principle of the present invention, and these improvements and modifications should also be considered within the scope of protection of the present invention.

Claims

1. A satellite antenna dynamic pointing calibration method based on on-board autonomous gradient optimization, characterized in that, The specific steps are as follows: Step 1: Data Collection (1.1) During the satellite's operation in orbit, the ground station continuously receives satellite signals and records the received signal quality parameters at each transit, including the signal-to-noise ratio (SNR); the ground station transmits the collected SNR information to the satellite via the uplink. (1.2) While receiving signal-to-noise ratio information, the satellite records antenna stress and temperature information through mechanical and thermal sensors at its own antenna. Step 2: Error Analysis (2.1) Using the signal quality parameters received by the ground station and the satellite attitude and orbit information, establish an antenna pointing error model: Wherein, "SNR" represents the current signal-to-noise ratio measurement; max Δψ represents the theoretical maximum signal-to-noise ratio, i.e., the optimal value when the antenna is aligned; Δψ represents the pitch pointing error; Δψ represents the yaw pointing error. For the overall pointing error angle; cos 2 (·) represents the pattern attenuation function; (2.2) The satellite actively applies a small perturbation once in each of the four orthogonal directions along the antenna pointing axis in the elevation and azimuth planes; based on a nonlinear mapping model. Calculate the gradient vector of the current pointing deviation. in, The signal-to-noise ratio gradient vector; This is the partial derivative of the signal-to-noise ratio with respect to the pitch angle; This is the partial derivative of the signal-to-noise ratio with respect to the yaw angle; Achieve this through the finite difference method: Where SNR(t) is the signal-to-noise ratio measurement at the current time t; SNR(t-Δt) is the signal-to-noise ratio at the previous measurement time (t-Δt); Δt is the time interval; and Δψ is the actively applied pitch angle disturbance. Introducing a sliding time window mechanism to dynamically adjust Δt to Where τ is the basic sampling interval, in seconds (s), determined according to the satellite attitude control cycle; γ is the window width adjustment coefficient, in 1 / (dB), which controls the sensitivity of Δt to gradient changes, so that the sampling interval is automatically shortened when the gradient is large to improve the convergence speed. (2.3) By analyzing the relationship between signal quality parameters and satellite attitude and antenna force and heat parameters, the magnitude and direction of the current antenna pointing error can be determined; (2.4) When the satellite actively performs attitude disturbances, if the received signal quality parameters show a significant decrease, it indicates that the antenna pointing may have deviated from the direction of the ground station. Step 3: Error Correction (3.1) Based on the error analysis results, the antenna pointing error is gradually reduced by adjusting the satellite attitude, the angle of the antenna rotating arm, and the feed electric scanning angle; (3.2) After each adjustment, the signal quality parameters received by the ground station are collected again to evaluate the correction effect; (3.3) Through iterative adjustments, the antenna pointing error is eventually brought to within the allowable range, achieving high-precision pointing; Step 4: Calibration Completed (4.1) When the received signal quality parameters reach the expected index and remain stable at a high level, the antenna pointing calibration is considered complete. (4.2) Record the final satellite attitude and antenna rotation arm angle parameters as the reference for subsequent antenna pointing control; In the error correction of step three, adaptive gradient descent correction is adopted, and the iterative formula is designed as follows: Where, δφ k δψ is the pitch angle correction value for the k-th iteration. k α is the yaw angle correction for the k-th iteration; α is the learning rate coefficient, ranging from 0.1 to 0.3; β is the decay coefficient. Exponential decay term e -βk The introduction of this feature allows the iteration step size to adaptively decrease as the number of iterations k increases, ensuring rapid approximation of the optimal solution in the early stages while avoiding oscillations near the extreme points in the later stages. In the error correction of step three, dynamic error compensation is adopted, which is achieved by real-time fusion of temperature sensor data T. i and strain gauge output σ j Establish a physical-driven compensation quantity: Among them, L i Let E be the structural characteristic length, and G be the elastic modulus and shear modulus, respectively. α i β j Here, m represents the coefficient of thermal expansion and the strain sensitivity coefficient of the material, n represents the total number of temperature sensors, and n represents the number of measurement points of the strain gauge. j Let be the equivalent area of ​​the j-th strain point.

2. The satellite antenna dynamic pointing calibration method based on on-board autonomous gradient optimization according to claim 1, characterized in that: In adaptive gradient descent correction, the gradient vector is updated at a frequency of 10Hz; in dynamic error compensation, the physical field parameters are refreshed at a frequency of 1Hz; and a dual-rate mechanism is used to balance the computational load and response speed.

3. The satellite antenna dynamic pointing calibration method based on on-board autonomous gradient optimization according to claim 1, characterized in that: The temperature change ΔT of the antenna structure is measured by a spaceborne PT1000 platinum resistance sensor; the mechanical stress change Δσ is acquired by a Wheatstone bridge strain gauge.

4. A satellite antenna dynamic pointing calibration system based on on-board autonomous gradient optimization, implementing the satellite antenna dynamic pointing calibration method based on on-board autonomous gradient optimization as described in any one of claims 1-3, characterized in that: A hierarchical data processing architecture was constructed, with its hardware carrier being a multi-core processor module integrated into the satellite's integrated electronic system. The multi-core processor module includes four parts: a data interface unit, a gradient calculation unit, a physical field compensation unit, and an instruction synthesis unit. The data interface unit is connected to the space management system via the SpaceWire bus, and receives downlink signal-to-noise ratio (SNR) telemetry data packets from the ground station in real time. At the same time, it collects analog signals from temperature sensors and strain gauges in key locations via the CAN bus, and converts them into digital sequences after conversion by a 24-bit ADC. The satellite actively applies small perturbations in four directions in the elevation and azimuth planes, that is, perturbations are applied in the four orthogonal directions of the satellite antenna axis, and the gradient is calculated based on the changes in the received SNR value. The physical field compensation unit has a built-in satellite structural parameter database that stores the thermal expansion coefficient α of each component. i Elastic modulus E j When the temperature change ΔT is received i With strain change Δσ j At that time, the directional shift caused by deformation is calculated in real time based on the cantilever beam deformation model.

Images