An unmanned aerial vehicle double-loop beam tracking system and method based on an intelligent metasurface

Abstract

The application discloses a kind of unmanned plane double-ring beam tracking systems and methods based on intelligent metasurface, belong to wireless communication and electromagnetic beam control field, this scheme with ground end binocular depth camera as main sensor real-time collection unmanned plane image, the image sequence of unmanned plane is carried out target detection, correlation tracking and trajectory estimation, output with the reference of metasurface center Azimuth and pitch angle, to generate intelligent metasurface discrete encoding matrix to complete beam state update;Through unmanned plane end power feedback unit, received power is converted into voltage sequence and with preset sampling interval back to ground end;Double-ring coupling control module extracts amplitude and trend characteristics to back voltage sequence, uses improved Kalman filter to correct the angle state of unmanned plane, to form beam pointing closed-loop update.This scheme is applicable to low-altitude communication, monitoring and relay and other application scenarios.

Description

Technical Field

[0001] This invention belongs to the field of wireless communication and electromagnetic beam control, and more specifically, relates to a dual-ring beam tracking system and method for unmanned aerial vehicles based on intelligent metasurfaces. Background Technology

[0002] In applications such as communication, monitoring, and relay for low-altitude mobile platforms, link stability and pointing consistency typically depend on beam alignment. Reconfigurable Intelligent Surfaces (RIS), as reconfigurable electromagnetic control structures, can change the reflection / transmission phase distribution under external control to form beams in a specific direction. In practical deployments, the beam state update of the RIS is completed through control and bias voltage application, and the update cycle is affected by hardware and the control link. Meanwhile, UAV targets are constantly moving in time, and visual observation of UAVs suffers from frame rate and processing delays; single information sources may experience obstruction or short-term loss. Summary of the Invention

[0003] To address the aforementioned deficiencies or improvement needs of existing technologies, this invention provides a dual-ring beam tracking system and method for unmanned aerial vehicles (UAVs) based on intelligent metasurfaces. A closed-loop relationship is established between intelligent metasurface beam updates, visual angle estimation, and energy domain feedback to support continuous pointing updates of UAV targets.

[0004] To achieve the above objectives, according to a first aspect of the present invention, a dual-ring beam tracking method for unmanned aerial vehicles based on a smart metasurface is provided, comprising:

[0005] S1, Receive the UAV image of the current frame acquired by the binocular camera at the current sampling time n+1. The target detection algorithm is used to analyze the UAV image in the current frame. Target detection is performed to obtain detection results, and the ByteTrack algorithm is used to compare the detection results with the previous frame of the UAV image. Data association is performed to obtain the UAV pixel coordinates; the array discrete control quantity of RIS is generated based on the UAV pixel coordinates to update the beam state of RIS;

[0006] S2, calculate the previous frame of the drone image. The pixel speed of drones in The voltage signal corresponding to the power signal fed back by the UAV after updating the beam state of the RIS. According to the pixel speed With voltage signal The improved ByteTrack algorithm is used to compare the detection results with the previous frame of the drone image. Data association is performed to obtain the corrected UAV pixel coordinates; the corrected array discrete control quantity of RIS is generated based on the corrected UAV pixel coordinates to update the beam state of RIS again.

[0007] S3 updates n to n+1 and returns to S1 until the termination condition is met.

[0008] According to a second aspect of the present invention, a dual-ring beam tracking system for unmanned aerial vehicles based on a smart metasurface is provided, comprising:

[0009] RIS is used to receive and reflect communication signals sent by the base station;

[0010] The ground-based control device includes a binocular depth camera, a dual-loop coupled control module, and a phase / voltage transmission module;

[0011] The binocular depth camera is located on the same plane as the RIS and is used to acquire UAV images in real time.

[0012] The dual-ring coupling control module is used to execute the method as described in the first aspect to update the beam state of the RIS;

[0013] The phase / voltage transmission module is used to convert the discrete control quantity of the array into a voltage control signal of the RIS and transmit it to the voltage control board of the RIS so that the RIS can update the beam state.

[0014] The UAV-side power feedback unit is used to receive the power signal received by the UAV after the RIS updates the beam state, convert it into a voltage signal, perform analog-to-digital conversion to obtain a digital voltage signal, and then send it to the control module.

[0015] In summary, compared with the prior art, the above-described technical solutions conceived by this invention can achieve the following beneficial effects:

[0016] The present invention provides a dual-ring beam tracking system and method for unmanned aerial vehicles (UAVs) based on a smart metasurface. This system uses a ground-based binocular depth camera as the main sensor to acquire UAV images in real time. It performs target detection, correlation tracking, and trajectory estimation on the UAV image sequences, outputting the azimuth angle φ and pitch angle θ with the metasurface center as a reference. Based on the angle states, it generates a discrete encoding matrix for the smart metasurface. By encoding matrix Mapped to array bias voltage matrix The signal is then sent to the voltage control board to apply a corresponding bias to the intelligent metasurface array unit to complete the beam state update. The UAV-side power feedback unit converts the received power into a voltage sequence and transmits it back to the ground at preset sampling intervals. The dual-loop coupled control module extracts the amplitude and trend characteristics of the transmitted voltage sequence and uses an improved Kalman filter to correct the UAV's angle state, thus forming a closed-loop beam pointing update. This scheme establishes a closed-loop relationship between intelligent metasurface beam updating, visual angle estimation, and energy domain feedback to support continuous pointing updates for UAV targets, and is suitable for applications such as low-altitude communication, monitoring, and relay. Attached Figure Description

[0017] Figure 1 This is one of the schematic diagrams of the UAV dual-ring beam tracking method based on intelligent metasurface provided in the embodiments of the present invention.

[0018] Figure 2 This is the second schematic diagram of the process of the dual-ring beam tracking method for UAVs based on intelligent metasurfaces provided in this embodiment of the invention.

[0019] Figure 3 This is the third schematic diagram of the UAV dual-ring beam tracking method based on intelligent metasurface provided in the embodiments of the present invention.

[0020] Figure 4 This is a schematic diagram illustrating the process of calculating UAV pixel coordinates based on the YOLOv8 and ByteTrack algorithms, as provided in an embodiment of the present invention.

[0021] Figure 5 This is a schematic diagram illustrating the geometric relationship and angle calculation between the camera coordinate system and the intelligent metasurface coordinate system provided in an embodiment of the present invention.

[0022] Figure 6 A block diagram showing the parameter adjustment of an improved Kalman filter provided in an embodiment of the present invention.

[0023] Figure 7 This is a schematic diagram illustrating the process of calculating the corrected UAV pixel coordinates based on YOLOv8 and the improved ByteTrack algorithm, as provided in an embodiment of the present invention.

[0024] Figure 8 This is a timing diagram illustrating angle correction within a single intelligent metasurface beam state update interval, as provided in an embodiment of the present invention.

[0025] Figure 9 This is a schematic diagram showing the correspondence between 2-bit 4-state phase control and four-level bias voltages provided in an embodiment of the present invention. Detailed Implementation

[0026] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention. Furthermore, the technical features involved in the various embodiments of this invention described below can be combined with each other as long as they do not conflict with each other.

[0027] This invention provides a dual-ring beam tracking method for unmanned aerial vehicles (UAVs) based on a smart metasurface, such as... Figures 1-3 As shown, it includes:

[0028] S1, Receive the UAV image of the current frame acquired by the binocular camera at the current sampling time n+1. The target detection algorithm is used to analyze the UAV image in the current frame. Target detection is performed to obtain detection results, and the ByteTrack algorithm is used to compare the detection results with the previous frame of the UAV image. (i.e., the UAV image acquired by the binocular camera at the previous sampling time n) is correlated to obtain the UAV pixel coordinates; the array discrete control quantity of RIS is generated based on the UAV pixel coordinates to update the beam state of RIS.

[0029] S2, calculate the previous frame of the drone image. The pixel speed of drones in The voltage signal corresponding to the power signal fed back by the UAV after updating the beam state of the RIS. According to the pixel speed With voltage signal The improved ByteTrack algorithm is used to compare the detection results with the previous frame of the drone image. Data association is performed to obtain the corrected UAV pixel coordinates; the corrected array discrete control quantity of RIS is generated based on the corrected UAV pixel coordinates to update the beam state of RIS again; n=1,2,…;

[0030] S3 updates n to n+1 and returns to S1 until the termination condition is met.

[0031] Specifically, in step S1, target detection and data association are performed on the current frame UAV image captured by the camera to obtain the pitch angle and azimuth angle of the UAV, thereby further obtaining the array discrete control quantity of RIS.

[0032] The target detection algorithm can be any existing target detection algorithm, and the embodiments of the present invention do not limit it to a unique one.

[0033] As an example, the object detection algorithm used is YOLOv8.Figure 4 As shown, in step S1, the YOLOv8 target detector is used to analyze the UAV image in the current frame. Target detection is performed to obtain detection results, namely high-confidence detection boxes and low-confidence detection boxes; then, the ByteTrack algorithm is used with a Kalman filter to predict the previous frame of the drone image. The location of the drone in the image is used to calculate the intersection-over-union (IoU) ratio between the predicted bounding box in the previous frame and the detected bounding box in the current frame, and then set it to 1. An IoU (Interval of Value) is used to construct a matching cost matrix; when the IoU is below a threshold, the corresponding matching relationship is discarded. Then, the Hungarian algorithm is used to minimize the overall matching cost under one-to-one constraints, achieving the optimal association between the predicted trajectory and the detection box, and obtaining the UAV pixel coordinates.

[0034] It is understandable that the drone's pixel coordinates are the same as the pixel coordinates of the drone's center.

[0035] Preferably, in step S1, generating the array discrete control quantity of RIS based on the UAV pixel coordinates includes:

[0036] S101, based on camera intrinsic parameters, determines the drone's pixel coordinates. and depth Convert to 3D points in camera coordinate system ;

[0037] S102, based on the calibration bias of the camera and RIS center for 3D points Coordinate correction is performed to obtain three-dimensional points with RIS as the reference. ;

[0038] in, , , = , Calibration bias between the binocular camera and the RIS center;

[0039] S103, according to Calculate the azimuth angle of the UAV relative to the RIS With pitch angle ;

[0040] S104, based on the azimuth angle of the UAV relative to RIS With pitch angle Calculate the array discrete control variables of the RIS.

[0041] Specifically, the camera intrinsic parameters are first calibrated to obtain the back projection relationship from the pixels to the camera coordinates; then the relative position of the camera and the center of the metasurface is calibrated to obtain the mounting offset. Calibration can be achieved by placing the target at several known locations near the geometric center of the metasurface and fitting the translation relationship between the camera coordinate system and the metasurface reference point. To reduce errors, The average of multiple measurements can be taken and kept constant during system operation or periodically retested and updated.

[0042] The following is an example of coordinate restoration and angle calculation. The camera coordinate system can be represented as:

[0043]

[0044] If we only consider the main installation offset as a translation along the y-axis, such as Figure 5 As shown, the following extrinsic translation vector can be used. Correction method:

[0045]

[0046]

[0047] The azimuth and elevation angles can be calculated using the geometric relationship of "with the center of the metasurface as the pole".

[0048]

[0049]

[0050] The aforementioned angle output, with the center of the intelligent metasurface as a reference, can convert the fixed angular deviation caused by the camera mounting offset into a calibrable translation. This improves repeatability across different installation locations. If pitch / yaw rotation errors exist during on-site installation, a rotation matrix can be further introduced. Perform rotational correction.

[0051] In step S104, based on the azimuth angle of the UAV relative to the RIS... With pitch angle According to the offline beamcodebook The array discrete control values ​​of the RIS can be obtained by looking up the table, which can be used to achieve faster and more stable control outputs in common angle ranges.

[0052] It is understood that the intelligent metasurface is an a×a two-dimensional metasurface array, where the value of 'a' can be arbitrarily set. Each array unit adopts a varactor diode loading structure to achieve k-bit 22 k Phase control for a given state. That is, the discrete control quantity of the array is a k-bit encoded matrix. .

[0053] Encoding matrix Mapped to array bias voltage matrix The data is then sent to the voltage control board of the RIS, which enables the beam status update of the RIS.

[0054] Taking a=16 and k=2 as an example, the discrete control quantity of the array is a 16×16 2-bit encoded matrix. , mix the 2-bit encoded state with 2 2 =The four bias voltages are matched one-to-one, thus encoding the matrix Mapped to array bias voltage matrix ;in, , { }, for The first in i Line number j Column elements, These are four bias voltages corresponding to the 2-bit encoded state.

[0055] To improve the encoding efficiency of the array discrete control quantity, preferably, in step S104, the azimuth angle is... With pitch angle The input is fed into a pre-trained neural network to obtain the discrete control input for the RIS array. This method can also cover angles that are not densely sampled in the codebook.

[0056] The two encoding methods mentioned above can be automatically switched according to conditions such as "angle change speed", "confidence level" and "system load" to ensure that the control link can operate under different computing power conditions.

[0057] Preferably, in step S104, the azimuth angle of the UAV relative to the RIS is determined. With pitch angle Before calculating the array discrete control variables of the RIS, the following steps are also included:

[0058] azimuth angle With pitch angle As the current azimuth angle With pitch angle Each of these is individually mapped to and fused with historical azimuth and historical elevation angles to determine the current azimuth angle. With pitch angle Smoothing is performed to avoid frequent jumps in the array's discrete control values ​​caused by false detections in a single frame or jitter in the detection frame. Any smoothing method can be used, such as exponential smoothing.

[0059] Preferably, the neural network is a CNN-Transformer network.

[0060] In step S2, based on the pixel velocity and voltage signals, the improved ByteTrack algorithm is used to combine the detection results with the previous frame of the UAV image. By performing data correlation, the corrected UAV pixel coordinates are obtained, including:

[0061] S201, employing an improved Kalman filter based on the pixel speed. With voltage signal Predict the previous frame of the drone image Correction results of the position of the UAV ;

[0062] S202, the previous frame of drone image Correction results of the position of the UAV The data is correlated with the detection results to obtain the corrected UAV pixel coordinates.

[0063] Specifically, the improved ByteTrack algorithm used in step S2 is improved in the Kalman filter. In the conventional ByteTrack algorithm, the Kalman filter is based on the previous frame of the UAV image. The target state is used to predict the current target position; however, in the method provided by this invention, an improved Kalman filter is used to further introduce the target position from the previous frame of UAV image. Extracted pixel speed With voltage signal The state of the UAV target is predicted and corrected to obtain the corrected position of the UAV in the previous frame. .

[0064] Specifically, in step S2, “alignment change information” is extracted from the voltage signal corresponding to the power signal fed back after the UAV’s beam state is updated in the RIS, and used to correct the UAV’s angle (i.e., pitch angle and azimuth angle) state, thereby updating the RIS control quantity.

[0065] After receiving the voltage signal corresponding to the power signal fed back by the UAV after updating its beam state at RIS, a smoothed voltage is first obtained through exponential smoothing. :

[0066]

[0067] in, This represents the smoothing coefficient; a larger value indicates a larger proportion of the current frame. For example, a value of 0.3 could be used to reduce voltage fluctuations. Its changing trend, including differential values, is then calculated. With slope It is used to characterize whether the beam pointing is accurate, thereby indirectly reflecting the alignment changes between the beam and the UAV.

[0068] The relevant calculation formulas are as follows:

[0069]

[0070]

[0071] Meanwhile, the pixel velocity of the drone is introduced to measure the magnitude of this trend, and its calculation is based on the following definition, where and These represent the speed magnitudes in the image's length and width, respectively.

[0072]

[0073] in, , ,in, , The center of the drone is in the previous frame of the drone image. In coordinate, , The center of the drone is located in the two previous drone images. The central location, for , The sampling time interval.

[0074] The ground terminal maps voltage variation trends and pixel speeds to filter modulation factors. It is used to control the improved Kalman filter.

[0075]

[0076] Improved Kalman filter parameter tuning, such as Figure 6 As shown, modulation factor Responsible for controlling process noise Size. Setting the value too low may cause the predicted drone location to lag behind the actual location. Setting it too high has the opposite effect; simulation tests have shown this. The value range is [0.5, 2.5]. From a physical perspective, This represents the system model's subjective confidence in the uncertainty of the angle state evolution; a larger value indicates that the system is less confident in the angle prediction obtained by extrapolating from the motion model. Based on this, the width of the confidence interval of the current beam pointing relative to the actual UAV direction, i.e., the system's covariance, is... It can be calculated using the following formula:

[0077]

[0078]

[0079] Through With process noise It is obtained recursively; specifically, it is initialized with an initial value. In the prediction process from frame 0 to frame 1, process noise is introduced to obtain the prediction covariance. After the state update is completed within the first frame, no new process noise terms are introduced again, therefore the corresponding posterior error covariance satisfies... The above recursive method ensures that the error covariance only expands as time progresses, remaining consistent throughout the update process of the same frame. In this embodiment, The initial value is set to 10 to characterize the initial uncertainty of the target center position and scale parameters, i.e., the initial position standard deviation is approximately The size of each pixel.

[0080] Based on the updated error covariance, the Kalman gain of the improved Kalman filter is... Calculated by the following formula:

[0081]

[0082] Noise measurement This reflects the reliability of the calculated azimuth or pitch angle of the UAV at the current moment. In this invention, It is not a fixed constant, but rather changes dynamically during the observation process. Specifically, , that is, The value is proportional to the feedback voltage and represents the standard deviation between the predicted and actual drone positions. A higher feedback voltage indicates a worse alignment effect, meaning a larger observation error. Simulation experiments have shown this. The value is approximately 4.

[0083] Ultimately, the improved Kalman filter predicts the corrected UAV position. It is estimated that updates will be made in the following format:

[0084]

[0085] When the drone accelerates or the received voltage drops rapidly (corresponding to a decrease in received power), the system's confidence in the model's predictions decreases. and As it increases, it thus improves. This causes the angle estimate to be more strongly pulled toward the offset correction direction indicated by physical feedback, accelerating the alignment recovery process after beam misalignment; while when the voltage and motion state are stable, and It automatically reduces and estimates updates more conservatively, thereby effectively suppressing angle jitter and overcorrection. for The drone pixel coordinates in the data, i.e. The UAV pixel coordinates are obtained by using target detection algorithms and the ByteTrack algorithm.

[0086] Based on the improved Kalman filter described above, the target tracking box is corrected, i.e., the corrected UAV position ( , , ) can be made by Figure 7 The corrected azimuth angle of the UAV is determined by calculations performed using Bytetrack. and pitch angle The calculation method is as follows:

[0087]

[0088]

[0089]

[0090] This invention provides a dual-ring beam tracking system for unmanned aerial vehicles (UAVs) based on a smart metasurface, comprising:

[0091] RIS is used to receive and reflect communication signals sent by the base station;

[0092] The ground-based control device includes a binocular depth camera, a dual-loop coupled control module, and a phase / voltage transmission module;

[0093] The binocular depth camera is located on the same plane as the RIS and is used to acquire UAV images in real time.

[0094] The dual-ring coupling control module is used to execute the method described in any of the above embodiments to update the beam state of the RIS;

[0095] The phase / voltage transmission module (which may be an FPGA) is used to convert the discrete control quantity of the array into a voltage control signal of the RIS and transmit it to the voltage control board of the RIS so that the RIS can update the beam state.

[0096] The UAV-side power feedback unit is used to receive the power signal received by the UAV after the RIS updates the beam state, convert it into a voltage signal, perform analog-to-digital conversion to obtain a digital voltage signal, and then send it to the control module.

[0097] Specifically, the system provided by this invention spatially comprises a ground-based control device and a UAV terminal: the ground-based control device is responsible for visual perception, angle estimation, and the generation and transmission of control quantities; the UAV terminal is responsible for energy domain observation (power / voltage) and data transmission; the intelligent metasurface is deployed on the ground side as a controlled radiation / reflection unit, updating its array state according to the control quantities to change the beam pointing or radiation pattern. That is, the ground-based control device acquires sensor information and obtains direction-related parameters; generates and transmits discrete array control quantities based on the direction-related parameters to update the beam state of the intelligent metasurface; acquires the received power feedback sequence transmitted back by the UAV terminal; and corrects the direction-related parameters, the discrete array control quantities, or their update strategies based on the feedback sequence, thereby achieving closed-loop updating.

[0098] The dual-loop coupling control module constructs a scaling factor based on the voltage variation trend and the UAV speed. The improved Kalman filter parameters are dynamically adjusted to correct the angle state or filter parameters. The dual-loop coupled control module employs a PAC adaptive coupling mechanism based on... Adjusting the process noise covariance of the improved Kalman filter With observation noise .

[0099] Within a closed-loop cycle, the ground-based binocular depth camera outputs synchronized UAV image frames and depth data. The processor performs target localization and trajectory maintenance on the image frames, obtaining the pixel position and corresponding depth of the target (i.e., the UAV) in the image. Subsequently, the pixels and depth are restored to three-dimensional points in the camera coordinate system, and coordinate correction is performed based on the installation offset between the camera and the metasurface center, ensuring that the angle output (i.e., pitch and azimuth angles) is referenced to the metasurface center, thereby reducing fixed deviations caused by differences in installation position. Then, the ground end generates array control codes (i.e., encoding matrices) based on the obtained azimuth and pitch angles, and obtains the array offset voltage matrix through voltage mapping. The voltage control board applies the corresponding offset to the array units to complete one metasurface state update. Simultaneously, the UAV continuously acquires the voltage sequence corresponding to the received power and transmits it back to the ground end at a rate higher than the metasurface update frequency. The ground end uses this voltage sequence as an "energy domain observation of the alignment state" for fine-tuning the angle state, adaptive tuning of filter parameters, or determination of control quantity update trigger conditions, ensuring that the system remains aware of state changes until the next metasurface update.

[0100] Furthermore, to ensure link timing consistency, the ground end aligns the camera frame timestamp, angle estimation timestamp, and voltage feedback timestamp. A circular buffer can be used to cache the angle state and voltage samples of the most recent frames respectively, and the voltage features can be aligned to the angle state update time using the "nearest neighbor" or linear interpolation method, thereby avoiding miscorrection caused by asynchronous sampling.

[0101] like Figure 9 As shown, taking RIS using 2-bit discrete control as an example, each discrete state corresponds to a preset bias voltage level. In engineering implementation, the two-dimensional encoding matrix can be expanded into control frames by rows and columns. The control frames carry address / check information and are sent to the voltage control board via a wired bus. The voltage control board then outputs multiple biases to drive the array units. To enhance maintainability, the preset voltage levels, mapping tables, and codebooks can be stored in the ground-side configuration file, enabling the system to be quickly reconfigured after replacing arrays, frequency bands, or unit devices.

[0102] For each frame of image output by the camera, the control module first performs object detection to obtain candidate boxes; then it performs data association to maintain the continuity of object IDs and robustly handles situations such as short-term occlusion and interference from similar objects.

[0103] When occlusion or detection failure occurs, the system does not immediately interrupt control, but enters "prediction hold mode": In this mode, the trajectory estimation module outputs the predicted angle and holds it for several frames (or several milliseconds) until a reliable observation is obtained again; at the same time, it combines voltage feedback to determine whether the current beam is still in the effective alignment range. If the energy index drops rapidly, the prediction hold window is shortened and the aggressiveness of the recovery strategy is increased in order to return to the observable state as soon as possible.

[0104] The UAV-side power feedback unit outputs a feedback sequence (i.e., a voltage signal) characterizing the received power of the UAV's beam signal to the RIS and transmits it back to the ground control device. The UAV-side power feedback unit includes a receiving antenna, a transmitting antenna, a microcontroller, and a power detection circuit. It converts the received UAV power into an analog voltage, which is then sampled and digitized by the microcontroller and transmitted back to the ground control device via a wireless link. The sampling interval satisfies... Furthermore, the wireless link is a Bluetooth link or an equivalent wireless link.

[0105] The receiving antenna of the UAV-side power feedback unit couples the electromagnetic signal from the smart metasurface to the power detection circuit, which outputs an analog voltage that monotonically corresponds to the received power. Microcontrollers The voltage is sampled and digitized, and then transmitted back to the ground via a wireless link. To ensure the voltage feedback has "controllable" stability, filtering can be performed before or after transmission: for example, using a moving average to weaken instantaneous spikes, or using a first-order low-pass filter to suppress high-frequency noise while preserving the trend. The filtered voltage is denoted as... The drone pixel velocity obtained from the previous frame of drone image and the filtered voltage are fed back to the improved Kalman filter to correct the predicted position and beam pointing in the Bytetrack tracking process, forming a complete physical loop.

[0106] like Figure 8 As shown, voltage samples and velocity information continuously arrive at the ground station between two metasurface state updates; therefore, the ground station can continuously evaluate the alignment trend while waiting for the next state update, and provide a basis for triggering subsequent angle corrections or control quantity updates. To avoid time mismatch caused by wireless backhaul jitter, the ground station can attach an arrival timestamp to each voltage sample and use the "true time interval" instead of a fixed step size when calculating trend characteristics.

[0107] The system provided in this embodiment of the invention allows for the following extensions and replacements without changing the basic link:

[0108] Firstly, when the number of targets is expanded, the digital side can maintain multiple trajectories and output multiple sets of candidate angle states simultaneously. The control side can use time-division multiplexing to output different control codes in different time slices, or use codeword allocation to generate multi-beam / wide-beam strategies in the same frame (whether to use them depends on array capabilities and application requirements).

[0109] Secondly, the array size, operating frequency band, coding granularity, and voltage mapping method can be reconfigured according to different hardware platforms. The ground end only needs to replace the mapping table / codebook or the network output dimension to reuse the same control process.

[0110] Those skilled in the art will readily understand that the above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A dual-ring beam tracking method for unmanned aerial vehicles based on a smart metasurface, characterized in that, include: S1, Receive the UAV image of the current frame acquired by the binocular camera at the current sampling time n+1. The target detection algorithm is used to analyze the UAV image in the current frame. Target detection is performed to obtain detection results, and the ByteTrack algorithm is used to compare the detection results with the previous frame of the UAV image. Data association is performed to obtain the UAV pixel coordinates; the array discrete control quantity of RIS is generated based on the UAV pixel coordinates to update the beam state of RIS; S2, calculate the previous frame of the drone image. The pixel speed of drones in The voltage signal corresponding to the power signal fed back by the UAV after updating the beam state of the RIS. According to the pixel speed With voltage signal The improved ByteTrack algorithm is used to compare the detection results with the previous frame of the drone image. Data association is performed to obtain the corrected UAV pixel coordinates; the corrected array discrete control quantity of RIS is generated based on the corrected UAV pixel coordinates to update the beam state of RIS again. S3 updates n to n+1 and returns to S1 until the termination condition is met.

2. The method as described in claim 1, characterized in that, In step S2, based on the pixel velocity and voltage signals, the improved ByteTrack algorithm is used to combine the detection results with the previous frame of the UAV image. By performing data correlation, the corrected UAV pixel coordinates are obtained, including: S201, employing an improved Kalman filter based on the pixel speed. With voltage signal Predict the previous frame of the drone image Correction results of the position of the UAV ; S202, the previous frame of drone image Correction results of the position of the UAV By correlating the detection results with the data, the corrected UAV pixel coordinates are obtained; in, The calculation formula is: ； For the output of the improved Kalman filter The correction results for the position of the UAV in the middle, The first two frames of UAV images output by the improved Kalman filter. The correction results for the position of the UAV in the middle, for The drone pixel coordinates in the data To improve the Kalman gain of the Kalman filter, , To measure noise, , For preset coefficients, for The covariance of the UAV position correction results is calculated using the iterative formula. Iterative calculation yielded the following results: , The initial value for process noise is given. , , , , For smoothing coefficients, The previous two frames of drone images voltage signal, for , The sampling time interval.

3. The method as described in claim 1 or 2, characterized in that, The previous frame of the drone image The pixel speed of drones in The calculation formula is: in, , ,in, , These are the previous drone images. drone pixels coordinate, , These are the two drone images from above. drone pixels coordinate, for , The sampling time interval.

4. The method as described in claim 1, characterized in that, In step S1, generating the array discrete control quantity of RIS based on the UAV pixel coordinates includes: S101, based on camera intrinsic parameters, determines the drone's pixel coordinates. and depth Convert to 3D points in camera coordinate system ; S102, based on the calibration bias of the camera and RIS center for 3D points Coordinate correction is performed to obtain three-dimensional points with RIS as the reference. ; in, , , = , Calibration bias between the binocular camera and the RIS center; S103, according to Calculate the azimuth angle of the UAV relative to the RIS With pitch angle ; S104, based on the azimuth angle of the UAV relative to RIS With pitch angle Calculate the array discrete control variables of the RIS.

5. The method as described in claim 4, characterized in that, In step S104, the azimuth angle is... With pitch angle The input is fed into a pre-trained neural network to obtain the discrete control quantity of the RIS array.

6. The method as described in claim 5, characterized in that, In step S104, based on the azimuth angle of the UAV relative to the RIS... With pitch angle Before calculating the array discrete control variables of the RIS, the following steps are also included: azimuth angle With pitch angle As the current azimuth angle With pitch angle Each of these is individually mapped to and fused with historical azimuth and historical elevation angles to determine the current azimuth angle. With pitch angle Perform smoothing processing.

7. The method as described in claim 5, characterized in that, The neural network is a CNN-Transformer network.

8. A dual-ring beam tracking system for unmanned aerial vehicles based on a smart metasurface, characterized in that, include: RIS is used to receive and reflect communication signals sent by the base station; The ground-based control device includes a binocular depth camera, a dual-loop coupled control module, and a phase / voltage transmission module; The binocular depth camera is located on the same plane as the RIS and is used to acquire UAV images in real time. The dual-ring coupling control module is used to perform the method as described in any one of claims 1-4 to update the beam state of the RIS; The phase / voltage transmission module is used to convert the discrete control quantity of the array into a voltage control signal of the RIS and transmit it to the voltage control board of the RIS so that the RIS can update the beam state. The UAV-side power feedback unit is used to receive the power signal received by the UAV after the RIS updates the beam state, convert it into a voltage signal, perform analog-to-digital conversion to obtain a digital voltage signal, and then send it to the control module.

Images