Ground satellite communication antenna angle controller based on convolutional neural network
By integrating a lightweight convolutional neural network model onto an STM32 chip, sensor data is automatically calibrated, solving the calibration complexity and attitude drift problems of the attitude angle control device for ground satellite communication antennas on embedded platforms, and achieving high-precision and stable attitude control.
Patent Information
- Application Number
- CN202520055633.6
- Authority / Receiving Office
- CN · China
- Patent Type
- Utility models(China)
- Current Assignee / Owner
- Filing Date
- 2025-01-10
- Publication Date
- 2026-01-23
- Estimated Expiration
- 2035-01-10
AI Technical Summary
In the existing technology, the attitude angle control device of the ground satellite communication antenna has problems such as complex sensor error calibration, significant attitude angle drift and poor adaptability to dynamic environment. In particular, the computing resources are limited on the embedded platform, making it difficult to achieve high-precision and stable control.
By combining a lightweight convolutional neural network (CNN) model with an STM32 chip, and through the fusion of data from accelerometers, gyroscopes, and magnetometers, the sensors are automatically calibrated, and quaternion rotation function values are output to achieve high-precision attitude angle control, avoiding the cumulative errors of traditional integral calculations.
It achieves efficient, automatic calibration, drift-resistant and highly dynamic antenna control on embedded platforms, simplifies the calibration process, and improves the stability and accuracy of long-term operation.
Smart Images
Figure CN223828723U_ABST
Abstract
Description
Technical Field
[0001] This utility model relates to a ground satellite communication antenna angle controller based on a convolutional neural network. Background Technology
[0002] Ground-based satellite communication antennas play a crucial role in communication, navigation, and telemetry, and their performance directly affects the stability and accuracy of communication links. Precise control of antenna pointing is key to achieving stable communication, with attitude angle control devices playing a central role.
[0003] Core technologies in attitude control
[0004] Attitude angle control devices typically rely on the following sensors and algorithms: Magnetometer: Used to measure the direction of the Earth's magnetic field, but easily affected by magnetic field interference in the environment, leading to inaccurate data. Accelerometer: Used to measure the direction of gravity to determine the device's tilt angle, but its output is easily affected by external interference such as vibration in dynamic environments. Gyroscope: Used to measure angular velocity and obtain the attitude angle through integration, but prone to drift (accumulated zero bias error) during long-term operation.
[0005] To achieve precise antenna control, the data from these sensors needs to be effectively fused and processed. However, existing technologies have the following shortcomings:
[0006] Sensor error calibration is difficult: magnetometers, accelerometers and gyroscopes need to be calibrated separately, which is a complex and time-consuming process, and the calibration results are unstable under dynamic environments.
[0007] Attitude angle drift is a significant problem: traditional methods such as Kalman filtering and complementary filtering have accumulated errors, leading to inaccurate attitude estimation.
[0008] Poor adaptability to dynamic environments: Traditional algorithms experience a significant performance drop in dynamic scenarios such as rapid motion and vibration, affecting the accurate pointing of the antenna.
[0009] Convolutional Neural Networks (CNNs) have significant advantages in processing nonlinear features and fusing multidimensional data. In recent years, CNNs have been applied to multi-sensor data processing, demonstrating excellent performance, particularly in attitude estimation and control. However, the high computational resource requirements of deep learning models limit their application on embedded platforms. Currently, the mainstream implementation schemes for attitude angle control devices on the market mainly include traditional methods and some intelligent solutions. The following is a detailed introduction to the most similar implementation schemes:
[0010] Sensor calibration is complex and difficult to adapt dynamically.
[0011] Current technologies for calibrating magnetometers, accelerometers, and gyroscopes rely on manual intervention or calibration equipment. The calibration process is not only complex and time-consuming, but also requires specific experimental environments and additional equipment. Under dynamic conditions, the zero-bias error and noise of the sensors vary significantly, leading to unstable calibration results.
[0012] Attitude angle drift problem is significant
[0013] In traditional methods, gyroscope data needs to be integrated to calculate the attitude angle, but due to the accumulation of zero-bias error, this leads to significant attitude angle drift. Furthermore, while Kalman filtering can fuse data from multiple sensors, its response to environmental changes is slow, making it difficult to completely eliminate drift.
[0014] Poor adaptability to dynamic environments
[0015] Traditional methods perform poorly in dynamic scenarios (such as rapid rotation, vibration, or strong disturbances), making it difficult to accurately estimate attitude angles. While some existing deep learning-based solutions offer greater adaptability, they often rely on high-performance computing hardware and are not suitable for resource-constrained embedded platforms.
[0016] Insufficient integration of hardware and algorithms
[0017] Existing intelligent solutions are mostly based on academic research, and the model implementation relies on high-performance hardware such as GPUs. They lack deep integration with embedded hardware, resulting in high energy consumption and high cost, making it difficult to meet the needs of practical applications. Utility Model Content
[0018] This invention provides a ground satellite communication antenna angle controller based on a convolutional neural network.
[0019] The technical solution adopted by this utility model to solve the above-mentioned technical problems is as follows:
[0020] A ground satellite communication antenna angle controller based on a convolutional neural network includes a microcontroller, a connector connected to the microcontroller, an accelerometer, a gyroscope, a magnetometer, a temperature sensor, a global positioning system, flash memory, an optical tube, a power supply, and a serial port. The microcontroller is an STM32 chip, which integrates a lightweight convolutional neural network (CNN) model. This model takes signals collected by the accelerometer, gyroscope, and magnetometer as inputs. The input signals are fused to form a matrix, and the model outputs quaternion rotation function values. The quaternion rotation function values are used to accurately represent the current attitude angle, thereby achieving high-precision attitude angle control of the antenna. The lightweight convolutional neural network (CNN) model includes an input layer, convolutional and pooling layers, a fully connected layer, and a normalized output layer.
[0021] The input layer is input in the form of a time-series matrix via accelerometers, gyroscopes, and magnetometers;
[0022] The convolutional and pooling layers extract temporal features through multiple one-dimensional convolutional operations with weights of J×1×K×1. The ReLU (Rectified Linear Unit) non-linear activation function is used to enhance feature representation capabilities, and the max pooling layer reduces feature dimensions and improves computational efficiency.
[0023] The fully connected layer maps the features extracted by the convolutional layer (with weights of J×1×K×144) to the quaternion space;
[0024] The normalized output layer ensures the validity of the physical meaning of the quaternion values;
[0025] Preferably, the connector provides debugging and download functions for the microcontroller and uses serial communication via VCP_TX, VCP_RX, TMS, TCK, SWDIO, SWO, and NRST for program download and real-time debugging.
[0026] Preferably, the accelerometer's built-in chip uses I... 2 The C bus communicates with the STM32 chip, responsible for acquiring and transmitting high-precision triaxial acceleration data. The circuit design is equipped with filter capacitors and pull-up resistors.
[0027] Preferably, the gyroscope is connected to the STM32 chip via an I2C interface through a built-in chip, and the circuit design includes a bias calibration circuit and a filtering circuit.
[0028] Preferably, the magnetometer's built-in chip uses I... 2 The C interface connects to the STM32 chip for data transmission and enables reliable data acquisition through high-precision measurement. The circuit design optimizes the matching impedance and adds a filter capacitor.
[0029] Preferably, the temperature sensor monitors the temperature of the circuit board or the external environment, and transmits it via I... 2 Temperature data is transmitted via the C interface.
[0030] Preferably, the GPS provides accurate location information and timestamps for the system's geographic reference and calibration.
[0031] Preferably, the flash memory stores convolutional neural network model parameters, system firmware, and intermediate running data.
[0032] Preferably, the optical tube is a status indicator that displays the system operating status through a light-emitting diode, and is connected to a microcontroller through D1_R, D1_G, D2_R, D2_G interfaces to control the emission color.
[0033] Preferably, the power supply and serial port not only provide power to the entire system, but also transmit data and instructions through the serial port. They are also connected to an external pin (HUB_DETECT) to detect whether the antenna is correctly installed on the angle adjustment platform.
[0034] Compared with existing technologies, this utility model, a ground satellite communication antenna angle controller based on a convolutional neural network, forms a highly integrated self-calibration control device for ground satellite communication antennas through optimized circuit design and CNN model deployment. It features automatic calibration, anti-drift, high dynamic adaptability, and efficient operation on embedded platforms. Suitable for high-precision antenna control in complex environments, it designs and optimizes a lightweight convolutional neural network model, enabling efficient operation on embedded platforms such as STM32 chips. This overcomes the dependence of traditional deep learning models on high-performance computing hardware, achieving high-precision attitude estimation and control in resource-constrained embedded systems. By processing accelerometer, gyroscope, and magnetometer data through the CNN model, it automatically corrects the zero-bias error and noise of the sensors, eliminating the need for manual calibration, significantly simplifying the calibration process, and improving the device's usability and dynamic environment adaptability. The CNN model directly outputs quaternion rotation function values, avoiding the accumulated errors in traditional integral calculations, completely solving the attitude angle drift problem, and improving the stability and accuracy during long-term operation. Attached Figure Description
[0035] Figure 1 This is a block diagram illustrating the circuit principle of the ground satellite communication antenna angle controller based on a convolutional neural network according to this invention.
[0036] Figure 2 This is a logical relationship diagram of the lightweight convolutional neural network (CNN) model for the ground satellite communication antenna angle controller based on the present invention. Detailed Implementation
[0037] The present invention will be further described in detail below with reference to the accompanying drawings and embodiments.
[0038] like Figures 1 to 2 As shown, the ground satellite communication antenna angle controller based on convolutional neural network includes a microcontroller 1, a connector 2, an accelerometer 3, a gyroscope 4, a magnetometer 5, a temperature sensor 6, a global positioning system 7, a flash memory 8, an optical tube 9, a power supply, and a serial port 10 connected to the microcontroller 1.
[0039] Microcontroller 1 is an STM32 chip, which runs a lightweight convolutional neural network (CNN) model. This model takes signals collected by accelerometer 3, gyroscope 4 and magnetometer 5 as inputs. The input signals are fused to form a matrix, and the output is a quaternion rotation function value. The quaternion rotation function value is used to accurately represent the current attitude angle, thereby realizing high-precision attitude angle control of the antenna.
[0040] Lightweight convolutional neural network (CNN) models consist of an input layer, convolutional and pooling layers, fully connected layers, and a normalized output layer.
[0041] The input layer is fed in the form of a time series matrix via accelerometer 3, gyroscope 4 and magnetometer 5.
[0042] Convolutional and pooling layers extract temporal features through multiple one-dimensional convolution operations with weights of J×1×K×1. The ReLU non-linear activation function is used to enhance feature representation capabilities, while max pooling layers reduce feature dimensions and improve computational efficiency.
[0043] The fully connected layer maps the features extracted by the convolutional layer (with weights of J×1×K×144) to the quaternion space.
[0044] The normalized output layer ensures the validity of the physical meaning of the quaternion values.
[0045] As the core of the entire system, microcontroller 1 is responsible for acquiring sensor data, running a lightweight convolutional neural network (CNN) model, and outputting control signals.
[0046] Connector 2 provides debugging and download functions for microcontroller 1, and uses serial communication via VCP_TX, VCP_RX, TMS, TCK, SWDIO, SWO, and NRST for program download and real-time debugging.
[0047] Accelerometer 3 measures linear acceleration in real time for attitude calculation and via I 2 Data transmission is performed via the SCL and SDA ports of the C interface. The accelerometer 3 built-in chip transmits data through I... 2 The C-bus communicates with the STM32 chip, responsible for acquiring and transmitting high-precision triaxial acceleration data. Meanwhile, the circuit design is equipped with filter capacitors and pull-up resistors to improve signal stability and anti-interference ability, thereby ensuring the reliability of data transmission.
[0048] The gyroscope 4 is used to measure angular velocity, thereby assisting in the estimation of angle changes and attitude. The gyroscope is connected to the STM32 chip via an I2C interface through a built-in chip. The circuit design includes a bias calibration circuit and a filtering circuit, which effectively reduces noise interference and improves measurement accuracy.
[0049] Magnetometer 5 is used to measure the direction of the Earth's magnetic field, providing absolute direction reference and compensation for attitude calculation. Its built-in chip connects to I... 2 The C interface connects to the STM32 chip for data transmission and achieves reliable data acquisition through high-precision measurement. The circuit design optimizes the matching impedance and adds filter capacitors to effectively improve signal quality and reduce noise interference, thereby ensuring the accuracy of data acquisition and the stability of system operation.
[0050] Temperature sensor 6 monitors the temperature of the circuit board and the external environment, and transmits it via I... 2 Temperature data is transmitted via the C interface.
[0051] The GPS 7 provides precise location information and timestamps for system georeferencing and calibration.
[0052] Flash 8 stores convolutional neural network model parameters, system firmware, and intermediate runtime data.
[0053] The light channel tube 9 is a status indicator that displays the system's operating status via LEDs. It is connected to a microcontroller via D1_R, D1_G, D2_R, and D2_G interfaces to control the emitted color.
[0054] The power supply and serial port 10 not only provide power to the entire system, but also enable data and command communication via the serial port. At the same time, an external terminal (HUB_DETECT) is connected to a pin of the STM32 chip to detect whether the antenna is correctly installed on the angle adjustment platform.
[0055] An accelerometer 3, a gyroscope 4, and a magnetometer 5 form a sensor module. This sensor module is used to collect attitude data in three-dimensional space and is connected to the STM32 main control chip via an I2C interface. The accelerometer 3 measures gravitational acceleration, reflecting the tilt angle; the gyroscope 4 measures angular velocity data; and the magnetometer 5 collects geomagnetic direction data.
[0056] The STM32 chip also includes a control signal output module, a communication module, and a power management module. The control signal output module outputs control signals to the servo motor of the antenna adjustment platform to adjust the antenna direction. The signal output uses PWM, combined with antenna angle feedback to achieve closed-loop control. The communication module provides RS422 or UART communication interfaces for debugging, data transmission, and external command input. The power management module provides 3.3V to ensure the normal operation of the sensor and the STM32 chip. The power management circuit uses a voltage regulator chip to provide 3.3V to the STM32 chip and sensor, and incorporates electromagnetic shielding and decoupling capacitors to reduce power supply noise.
[0057] The STM32 chip communicates with the accelerometer 3, gyroscope 4, and magnetometer 5 via the I2C interface to receive raw data. The STM32 chip controls the antenna servo motor to adjust the angle via the PWM interface. The STM32 chip is used for debugging and external communication via the UART interface.
[0058] Once the power is turned on, the device initializes, and the accelerometer 3, gyroscope 4, and magnetometer 5 begin collecting data. The STM32 chip runs a lightweight convolutional neural network (CNN) model to complete attitude estimation. The antenna adjusts its direction according to the attitude angle to complete the communication target alignment. The UART interface is used to monitor the system status in real time or upload task instructions.
[0059] This invention relates to a ground satellite communication antenna angle controller based on a convolutional neural network (CNN). Through optimized circuit design and CNN model deployment, it forms a highly integrated self-calibration control device for ground satellite communication antennas. It features automatic calibration, anti-drift, high dynamic adaptability, and efficient operation on embedded platforms. Suitable for high-precision antenna control in complex environments, it utilizes a lightweight convolutional neural network model designed and optimized for efficient operation on embedded platforms such as STM32 chips. This overcomes the dependence of traditional deep learning models on high-performance computing hardware, achieving high-precision attitude estimation and control in resource-constrained embedded systems. The CNN model processes data from the accelerometer (3), gyroscope (4), and magnetometer (5), automatically correcting the zero-bias error and noise of the sensors without manual calibration, significantly simplifying the calibration process and improving the device's usability and dynamic environment adaptability. The CNN model directly outputs quaternion rotation function values, avoiding the accumulated errors in traditional integral calculations, completely solving the attitude angle drift problem and improving long-term stability and accuracy.
[0060] Finally, it should be noted that the above embodiments only illustrate the technical solutions of this utility model, and are not intended to limit it. Although this utility model has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of this utility model.
Claims
1. A ground satellite communication antenna angle controller based on a convolutional neural network, characterized in that, Includes a microcontroller (1), a connector (2) connected to the microcontroller (1), an accelerometer (3), a gyroscope (4), a magnetometer (5), a temperature sensor (6), a global positioning system (7), a flash memory (8), an optical tube (9), a power supply, and a serial port (10). The microcontroller (1) is an STM32 chip. A lightweight convolutional neural network model runs on the STM32 chip. The lightweight convolutional neural network model takes input signals through the accelerometer (3), the gyroscope (4), and the magnetometer (5). The input signals are a matrix after data fusion, and the output is a quaternion rotation function value. The quaternion rotation value represents the current attitude angle to achieve high-precision attitude angle control of the antenna. The lightweight convolutional neural network model includes an input layer, a convolutional layer and a pooling layer, a fully connected layer, and a normalized output layer. The input layer is input in the form of a time series matrix via an accelerometer (3), a gyroscope (4), and a magnetometer (5); The convolutional and pooling layers extract temporal features through multiple one-dimensional convolution operations with weights of J×1×K×1. The ReLU non-linear activation function is used to enhance feature representation capabilities, and the max pooling layer reduces feature dimensions and improves computational efficiency. The fully connected layer maps the features extracted by the convolutional layer with weights of J×1×K×144 to the quaternion space. The normalized output layer ensures the validity of the physical meaning of the quaternion values.
2. The ground satellite communication antenna angle controller based on a convolutional neural network according to claim 1, characterized in that, The connector (2) provides debugging and download functions for the microcontroller (1) and uses serial communication via VCP_TX, VCP_RX, TMS, TCK, SWDIO, SWO, NRST for program download and real-time debugging.
3. The ground satellite communication antenna angle controller based on a convolutional neural network according to claim 1, characterized in that, The accelerometer (3) has a built-in chip that communicates with I 2 The C bus communicates with the STM32 chip, responsible for acquiring and transmitting high-precision triaxial acceleration data. The circuit design is equipped with filter capacitors and pull-up resistors.
4. The ground satellite communication antenna angle controller based on a convolutional neural network according to claim 1, characterized in that, The gyroscope (4) is connected to the STM32 chip via an I2C interface through a built-in chip. The circuit design includes a bias calibration circuit and a filter circuit.
5. The ground satellite communication antenna angle controller based on a convolutional neural network according to claim 1, characterized in that, The magnetometer (5) has a built-in chip that communicates with I 2 The C interface connects to the STM32 chip for data transmission and enables reliable data acquisition through high-precision measurement. The circuit design optimizes the matching impedance and adds a filter capacitor.
6. The ground satellite communication antenna angle controller based on a convolutional neural network according to claim 1, characterized in that, The temperature sensor (6) monitors the temperature of the circuit board and the external environment, and through I 2 Temperature data is transmitted via the C interface.
7. The ground satellite communication antenna angle controller based on a convolutional neural network according to claim 1, characterized in that, The GPS (7) provides accurate location information and timestamps for the system’s geographic reference and calibration.
8. The ground satellite communication antenna angle controller based on a convolutional neural network according to claim 1, characterized in that, The flash memory (8) stores convolutional neural network model parameters, system firmware, and intermediate running data.
9. The ground satellite communication antenna angle controller based on a convolutional neural network according to claim 1, characterized in that, The light channel tube (9) is a status indicator and displays the system operating status through a light-emitting diode. It is connected to a microcontroller through the D1_R, D1_G, D2_R, D2_G interfaces to control the light emission color.
10. The ground satellite communication antenna angle controller based on a convolutional neural network according to claim 1, characterized in that, The power supply and serial port (10) provide power to the entire system and enable data and command communication through the serial port.