Laser communication APT system based on AI accelerator and space laser communication system

Abstract

This invention relates to the field of space laser communication technology, specifically to an AI accelerator-based laser communication APT system and a space laser communication system. The system includes a coarse tracking unit, an AI accelerator, and a fine tracking unit. The AI ​​accelerator detects and tracks the beacon beam spot in the beacon beam image and determines the beacon beam spot position information, including its location and motion trajectory, based on tracking error information. The coarse and fine tracking units track the beacon beam spot based on the position information fed back from the AI ​​accelerator and feed it back to the corresponding servo system. This invention significantly reduces the processing latency of beacon beam spot detection, positioning, and tracking, compressing the acquisition and alignment time of traditional APT systems to the millisecond level, and significantly improving the real-time performance of spot acquisition and tracking. It maximizes the effective transmission time of inter-satellite laser communication and ensures long-term dynamic and stable aiming of the laser link.

Description

Technical Field

[0001] This invention relates to the field of space laser communication technology, specifically to an AI accelerator-based laser communication APT system and a space laser communication system. Background Technology

[0002] In the 6G era, the deep integration of communication networks and computing power scheduling has led to a continuous increase in the demand for bandwidth in space information transmission. Building a broadband space-based bearer network compatible with terrestrial networks has become an inevitable trend in the future development of communication technologies. Satellite microwave communication and free-space laser communication are the foundation for constructing space-based bearer networks. Compared to traditional microwave communication, laser communication has advantages such as high communication speed, strong anti-interference capability, low power consumption, and weak spectrum constraints, making space laser communication the basis for building broadband space-based bearer networks.

[0003] In space laser communication systems, the acquisition, pointing, tracking (APT) of transmitting and receiving lasers is the most critical technology in space laser communication and a prerequisite for establishing stable laser communication between satellites.

[0004] How to reduce the laser optomechanical alignment error and alignment time in the APT system, and ensure the rapid dynamic acquisition, alignment, tracking link establishment and communication of space lasers within a limited time, has become an urgent technical problem to be solved. Summary of the Invention

[0005] To address at least some of the technical problems in the prior art, the present invention provides an AI accelerator-based laser communication APT system and a space laser communication system.

[0006] In a first aspect, the present invention provides a laser communication APT system based on an AI accelerator, the laser communication APT system comprising: The coarse tracking unit outputs the detected beacon light spot image to the AI ​​accelerator; responds to the beacon light spot position information fed back by the AI ​​accelerator, and controls the beacon light spot to enter the tracking field of view of the fine tracking unit according to the beacon light spot position information; The AI ​​accelerator is used to track beacon beams based on beacon beam images input by a coarse tracking unit or a fine tracking unit, and to compensate for the tracked beacon beam position information based on tracking error information; the beacon beam position information is either the beacon beam position information detected by the coarse tracking unit or the beacon beam center position information detected by the fine tracking unit. The fine tracking unit is used to output the detected beacon light spot image to the AI ​​accelerator; it responds to the beacon light spot center position information fed back by the AI ​​accelerator, and controls the detected beacon light spot to enter the tracking field of view of the communication receiving field of view according to the beacon light spot center position information, so as to establish the coaxial state of the tracking line of view and the communication receiving line of view.

[0007] Optionally, the AI ​​accelerator is pre-configured with a dynamic target detection and tracking neural network and a compensation unit; The neural network is used to perform spot detection and trajectory tracking on the beacon spot image input by the coarse tracking unit or the fine tracking unit, and to track the position information of the beacon spot; The compensation unit is used to compensate for the position information of the tracked beacon spot based on the tracking error information.

[0008] Optionally, the dynamic target detection and tracking neural network is mainly constructed using the YOLOvx algorithm and the DeepSORT algorithm; the tracking error information includes spot detection error information, platform vibration residual information, and dynamic tracking error information.

[0009] Optionally, the tracking detector in the coarse tracking unit is a CCD detector; the operating frame rate of the CCD detector is more than 10 times the servo bandwidth of the servo mechanism in the coarse tracking unit.

[0010] Optionally, the tracking detector in the fine tracking unit is a four-quadrant detector (QD) or a position-sensitive detector (PSD); the resonant frequency of the servo mechanism in the fine tracking unit is matched with the sampling frequency of the tracking detector to form a broadband stable fine tracking unit.

[0011] Optionally, the laser communication APT system further includes a capture unit; The capture unit is used to capture the far-field beacon light spot using an AI accelerator.

[0012] Optionally, the laser communication APT system further includes a lead adjustment unit; The lead adjustment unit is used to align the tracking line of sight, the communication transmission line of sight, and the communication reception line of sight.

[0013] Optionally, the coarse tracking unit includes a tracking detector for detecting the beacon light spot captured by the capture unit and outputting the beacon light spot image to the AI ​​accelerator; The tracking controller in the coarse tracking unit is used to respond to the beacon light spot position information fed back by the AI ​​accelerator and send control signals according to the beacon light spot position information; The servo mechanism in the coarse tracking unit is used to control the beacon beam spot to enter the tracking field of view of the fine tracking unit according to the control signal of the tracking controller in the coarse tracking unit.

[0014] Optionally, the tracking detector in the fine tracking unit is used to detect the beacon light spot entering its tracking field of view and output the beacon light spot image to the AI ​​accelerator; The tracking controller in the fine tracking unit is used to respond to the beacon beam spot center position information fed back by the AI ​​accelerator and send control signals according to the beacon beam spot center position information; The servo mechanism in the fine tracking unit is used to control the beacon light spot to enter the tracking field of view of the communication receiving field of view according to the control signal of the tracking controller in the fine tracking unit, so as to establish the coaxial state of the tracking line of view and the communication receiving line of view.

[0015] In a second aspect, the present invention provides a space laser communication system, the space laser communication system comprising the laser communication APT system as described in any one of the above descriptions.

[0016] The embodiments of the present invention have the following technical effects: 1. The embodiments of this invention significantly reduce the processing latency of beacon spot detection, positioning, and tracking, compressing the acquisition and alignment time of traditional APT systems to the millisecond level, and significantly improving the real-time performance of spot acquisition and tracking. This maximizes the effective transmission time of inter-satellite laser communication, ensuring long-term dynamic and stable aiming of the laser link.

[0017] 2. This invention combines servo control with an AI accelerator, enabling high-precision locking of beacon beams. Even under complex conditions such as vibration, attitude disturbances, and relative motion on the satellite platform, it maintains tracking accuracy at the microradian (μrad) level. This technology not only enhances the anti-interference capability of space laser communication systems but also provides crucial support for the future construction of high-throughput, low-latency space-based communication networks.

[0018] 3. The satellite laser communication APT technology based on the dynamic target detection and tracking neural network of the AI ​​accelerator in this embodiment of the invention can simultaneously track, detect and locate multiple light spots. It is not only suitable for one-to-one node laser link access, but also for space multi-node multi-target laser link establishment APT system, laying the technical foundation for the next generation of space laser phased array antennas and multi-satellite networking laser communication.

[0019] 4. In this embodiment of the invention, the AI ​​accelerator is reconfigurable and upgradable in orbit, adaptable to signal processing of various types of detectors such as high-resolution CCD and QD, without relying on hardware modifications, and compatible with multiple APT architectures. The accelerator is lightweight and operates with low power consumption, which can better adapt to the stringent size, power consumption, and heat dissipation constraints of satellite payloads.

[0020] 5. The embodiments of the present invention break through the bottlenecks of traditional algorithms that rely on manually set thresholds and have poor dynamic adaptability. The AI ​​accelerator executes model inference and can autonomously learn the characteristics of light spots in complex spatial environments. It has stronger tracking robustness in highly dynamic and multi-perturbation scenarios, and meets the laser rapid link establishment requirements of highly mobile satellite nodes in 6G space-based bearer networks. Attached Figure Description

[0021] Figure 1 This is a schematic diagram of the composition of a space laser communication system provided in an embodiment of the present invention; Figure 2 This is a schematic diagram of the APT process of the laser communication system provided in an embodiment of the present invention; Figure 3 This is a schematic diagram of the transmitting and receiving areas during the acquisition phase provided in an embodiment of the present invention; Figure 4 The APT system error provided in the embodiments of the present invention; Figure 5 This is a schematic diagram illustrating the deployment of the AI ​​accelerator for the pre-configured dynamic target tracking and detection neural network provided in an embodiment of the present invention within the APT system. Figure 6 This is a schematic diagram illustrating the use of an AI accelerator for spot detection and tracking in an APT system, as provided in an embodiment of the present invention. Detailed Implementation

[0022] Various exemplary embodiments of the present invention will now be described in detail. This detailed description should not be considered as a limitation of the present invention, but rather as a more detailed description of certain aspects, features, and embodiments of the present invention.

[0023] It should be understood that the terminology used in this invention is merely for describing particular embodiments and is not intended to limit the invention. Furthermore, with respect to numerical ranges in this invention, it should be understood that the upper and lower limits of the range and each intermediate value between them are specifically disclosed. Any stated value or intermediate value within a stated range, as well as each smaller range between any other stated value or intermediate value within said range, are also included in this invention. The upper and lower limits of these smaller ranges may be independently included or excluded from the range.

[0024] Unless otherwise stated, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains. While only preferred methods and materials have been described herein, any methods and materials similar or equivalent to those described herein may be used in the implementation or testing of this invention. All references to this specification are incorporated by way of citation to disclose and describe methods and / or materials associated with those references. In the event of any conflict with any incorporated reference, the content of this specification shall prevail.

[0025] Example 1 To address the shortcomings of existing technologies, embodiments of the present invention provide a laser communication APT system based on an AI accelerator, such as... Figure 1 As shown, the laser communication APT system (hereinafter referred to as the APT system) is a component of a space laser communication system. The space laser communication system has four main components: the APT system, the optomechanical system, the communication system, and the electronic control system. The APT system includes a capture unit, a coarse tracking unit, a fine tracking unit, and a lead adjustment unit. The electronic control system includes a system main control unit, a remote control and telemetry unit, and a power management unit. The optomechanical system includes a precision mechanical stage, an optical telescope unit, and an optical path relay unit. The communication system includes a transmitting unit, a receiving unit, a communication bus, and a modulation and coding unit. The laser communication APT system includes: The coarse tracking unit outputs the detected beacon light spot image to the AI ​​accelerator; responds to the beacon light spot position information fed back by the AI ​​accelerator, and controls the beacon light spot to enter the tracking field of view of the fine tracking unit according to the beacon light spot position information; The AI ​​accelerator is used to track beacon beams based on beacon beam images input by the coarse tracking unit or the fine tracking unit, and to compensate for the beacon beam position information, which includes the position and motion trajectory, based on the tracking error information; the beacon beam position information is either the beacon beam position information detected by the coarse tracking unit or the beacon beam center position information detected by the fine tracking unit. The fine tracking unit is used to output the detected beacon light spot image to the AI ​​accelerator; respond to the beacon light spot center position information fed back by the AI ​​accelerator, and control the detected beacon light spot to enter the tracking field of view of the communication receiving field of view according to the beacon light spot center position information, so as to establish the coaxial state of the tracking line of view and the communication receiving line of view.

[0026] In the APT system, an AI accelerator with AI target detection and tracking network inference computing capabilities is deployed between the tracking detector (CCD or QD) with narrowband filter and the electronic control system. This accelerator identifies and tracks the light spot and feeds the results back to the servo control unit. Ultimately, through the cooperation of coarse acquisition tracking, fine acquisition tracking, and lead control systems, dynamic and stable link establishment of the communication optical network is achieved.

[0027] For APT systems, they can be divided into two categories based on the presence or absence of beacon light in the laser link: beacon-based acquisition and beacon-less acquisition. In order to increase the far-field coverage of the beam energy to achieve rapid acquisition, APT systems can use independent beacon light emitting units to assist in acquisition and tracking. The beacon light is a laser with a large beam divergence angle.

[0028] In APT systems, reliable acquisition is a prerequisite for establishing a space laser communication link. Generally, the communication time between satellite laser links is limited. To maximize the effective transmission time of the laser communication link, the acquisition process needs to be minimized. During the establishment of a space laser communication link, the line-of-sight acquisition performance is affected by factors such as the line-of-sight pointing accuracy of the communication terminal, the acquisition scheme procedure, the scanning control strategy, and the signal detection and processing algorithm. To establish a laser communication link, the line-of-sight of the laser communication terminal must first be pointed according to ephemeris information in the initial stage of acquisition to achieve initial laser acquisition. The initial pointing accuracy has a significant impact on the acquisition performance of the laser communication link. The laser communication terminal includes a communication laser.

[0029] Tracking refers to the process where, after laser communication link acquisition is completed, the laser communication terminal controls a servo mechanism to compensate for link errors detected by the tracking detector. Platform vibration interference and relative motion between terminals can cause the tracking beam to deviate from the center of the detector's field of view. Therefore, the tracking system needs to be able to accurately correct the line of sight in real time to ensure stable transmission of the subsequent laser communication link. Tracking detectors such as charge-coupled devices (CCDs), quadrant detectors (QDs), and position-sensitive detectors (PSDs) can be used. However, regardless of the type of beam tracking detector used, an image processing system is required to detect the beam position and suppress errors.

[0030] To achieve high dynamic and high-precision tracking, a composite tracking structure combining coarse and fine tracking can be employed, where coarse and fine tracking units work together for control. The coarse tracking unit has a large field of view and a low servo bandwidth, primarily enabling rapid acquisition and stable coarse tracking, allowing the beacon spot to reliably enter the fine tracking field of view. The fine tracking unit has a smaller dynamic range, higher servo bandwidth, and higher tracking accuracy (on the order of μrad), and fine tracking can further suppress coarse tracking residuals. The final tracking speed of the APT system with the composite tracking structure mainly depends on the coarse tracking speed, while the tracking accuracy mainly depends on the performance of the fine tracking unit.

[0031] The alignment phase involves detecting the spot position deviation during the tracking phase to determine the direction of the incident beam, and then guiding the local laser beam to coaxially emit in the opposite direction to the detected beam direction. This achieves alignment at both ends of the APT system. The beam alignment and tracking phases are closely related and together determine the overall accuracy of the APT system.

[0032] The following sections will describe in detail the laser communication APT system provided in this invention from three perspectives: time sequence, spatial composition, and key algorithms.

[0033] I. The workflow of the laser communication APT system is introduced in chronological order.

[0034] The laser communication APT system provided in this embodiment of the invention has a total workflow of five stages: initial pointing, acquisition stage, coarse tracking alignment, fine tracking alignment, and dynamic communication. The first four stages are mainly driven by the beacon light, while the fifth stage is when the communication light begins to operate. The workflow of these five stages is summarized as follows: Figure 2 As shown.

[0035] Phase 1: Initial Pointing. First, the laser needs to establish an initial pointing based on the ephemeris, inertial sensor, and star-sensor information sent to the satellite's integrated electronic system or laser control box, ensuring the beacon's line-of-sight enters its respective uncertain region, such as... Figure 3 As shown.

[0036] The second stage: Acquisition. The acquisition unit directs the far-field beacon light beam towards the communication receiver's field of view, ensuring that the field of view encompasses the beacon light spot. In other words, APT system acquisition involves aligning the optical axis of the transmitting beam with the receiver's detector's field of view, allowing the receiver's detector's field of view to encompass the arriving spot of the transmitted beam. For example... Figure 3 As shown, the system enters its working state, first capturing the far-field beacon light spot. The beam from transmitter A mechanically scans the uncertain area of ​​receiver B, while simultaneously preparing the transmitter to receive beacon light feedback from the other end. A crucial condition for this process is that the receiving field of view at both ends must completely cover the divergence angle of the beacon light at their respective ends. Then, the beacon light spot is allowed to enter the receiver's detector field of view. Once the beacon light signal enters the receiver's camera field of view, a light spot will appear on the receiver, indicating that the system has completed the capture phase.

[0037] To achieve rapid and high-probability acquisition, the uncertainty area should be reduced first, and the acquisition mode and scanning method should be optimized second. The acquisition phase mainly involves scanning and searching the receiver detector's field of view over a large area. Since the detection field of view of the acquisition detector is relatively large, this phase is not significantly affected by mechanical vibration and spot jitter.

[0038] The third stage: coarse tracking. In the APT system's workflow, the tracking stage follows the acquisition stage. Its function is to stabilize the acquired beacon beam spot within a specific range on the surface of the fine tracking detector at the receiving end. The position signal of the detected beacon beam spot (referred to as the spot) is a jitter signal with a large residual. Therefore, the tracking stage needs to minimize the residual of this jittery target spot as much as possible, and also acquire the coordinate parameters of the beam direction in real time to prepare for the subsequent alignment stage and the reception of communication signals. Due to many uncertainties such as mechanical jitter and spot drift, the spot may move out of the field of view. In this case, it is necessary to provide error signal feedback for the spot to further track it.

[0039] For long-distance space laser communication systems, a two-stage coarse and fine beacon light emission is commonly used. The coarse tracking unit consists of three parts: a coarse tracking controller, a coarse tracking detector, and a servo turntable (servo mechanism). Its main functions include initial line-of-sight pointing, open-loop scanning and acquisition, and dynamic coarse tracking. The tracking detector (coarse tracking detector) in the coarse tracking unit detects the beacon light spot captured by the acquisition unit and outputs the image of that beacon light spot to the AI ​​accelerator. The tracking controller (coarse tracking controller) in the coarse tracking unit responds to the beacon light spot position information fed back from the AI ​​accelerator and sends control signals based on this information. The servo mechanism in the coarse tracking unit controls the beacon light spot to enter the tracking field of view of the fine tracking unit based on the control signals from the tracking controller.

[0040] To suppress the impact of detector measurement noise on overall tracking accuracy, improving detection accuracy is crucial. The coarse tracking detector of the APT system uses a high-resolution, large-format, high-frequency CCD detector; the CCD detector can also be called a CCD camera.

[0041] The fourth stage: fine tracking. The fine tracking unit is mainly used to further suppress coarse tracking residuals. This stage requires more precise beam alignment, allowing the beam spot to enter a smaller field of view, thus achieving the alignment goal. The tracking accuracy obtained by the fine tracking unit is the final accuracy of the compound axis APT system.

[0042] The fine tracking unit consists of a fine tracking detector, a spot centroid detection unit, a fine tracking servo mechanism, and a fine tracking controller. For the fine tracking unit, achieving precise spot center detection is a crucial measure to improve tracking accuracy. The fine tracking detector employs a four-quadrant detector (QD) or a position-sensitive detector (PSD). The tracking detector (referred to as the fine tracking detector) in the fine tracking unit detects the beacon light spot entering its tracking field of view and outputs the image of the beacon light spot to the AI ​​accelerator. The tracking controller (referred to as the fine tracking controller) responds to the beacon light spot center position information fed back from the AI ​​accelerator and sends control signals based on this information. The servo mechanism (referred to as the fine tracking servo mechanism) in the fine tracking unit controls the beacon light spot to enter the tracking field of view of the communication receiving field of view according to the control signals from the tracking controller, thereby establishing a coaxial state between the tracking line of sight and the communication receiving line of sight.

[0043] Phase 5: Dynamic Communication.

[0044] After the APT system completes precise tracking, the communication laser transceiver lines of sight become coaxial. At this point, the communication laser and the peer communication detector initiate data transmission. Simultaneously, the lead alignment unit compensates for the line-of-sight advance caused by the relative motion of the satellite, enabling dynamic communication.

[0045] II. Introduction to the APT system architecture of laser communication based on AI accelerator from the perspective of spatial composition.

[0046] like Figure 4 As shown, the errors involved in the APT system include tracking error and aiming error. The tracking error mainly includes: spot detection error, platform vibration residual, and dynamic tracking error, which are random errors with a Gaussian distribution. The aiming error is a systematic error, which is the algebraic sum of lead error and system mechanical alignment line-of-sight error. The dynamic error accuracy of the space laser communication system is usually at the microradian level (μrad). To achieve such high-precision tracking, on the one hand, high-performance sensors, controllers, and drivers are required. On the other hand, advanced control strategies are required, specifically including (1) multiple integral compensation control, (2) variable structure control, (3) adaptive control, and (4) coarse-fine composite axis control.

[0047] Figure 5This invention demonstrates the components of an APT system based on a pre-installed high dynamic target detection and tracking neural network (HVRLC) in an embodiment of the present invention. The AI ​​accelerator is deployed between the coarse and fine tracking detection unit and the electronically controlled servo system (including a coarse tracking HVRLC and a fine tracking HVRLC). The AI ​​accelerator executes the inference process of the YOLOvx+DeepSORT target detection and tracking AI model to process the initial position and accuracy data of the beacon light spot during the capture and tracking process. The compensation unit compensates for the position information of the tracked beacon light spot based on the tracking error information, obtaining beacon light spot position information including position and motion trajectory. In conjunction with the coarse and fine tracking servo systems, it achieves rapid capture and tracking of the beacon light spot, thereby reducing laser optomechanical alignment errors and alignment time, and ensuring rapid dynamic tracking, link establishment, and communication of spatial lasers within a limited time.

[0048] This invention employs a two-stage beacon light emission system, coarse and fine. The coarse tracking unit also functions as a capture unit. To reduce capture time, the coarse tracking field of view should be as large as possible; therefore, a high-resolution, large-array CCD detector is used for the coarse tracking detector. The video signal output from the CCD is pre-amplified, filtered, and thresholded before being fed into a high dynamic range target detection and tracking neural network (AI) accelerator. The AI ​​algorithm tracks and calculates the azimuth and elevation error control signals, which are then sent to the coarse tracking unit via the electronic control system to drive the servo turntable. The coarse tracking accuracy only needs to ensure the light spot enters the fine tracking field of view. Since there is a time delay between spot detection and signal processing, this delay reduces the stability phase margin of the control system, thus affecting the system's bandwidth and stability. To minimize the impact of this time delay on the control system, the operating frame rate of the CCD detector should be at least 10 times the servo bandwidth of the coarse tracking servo turntable.

[0049] The primary function of the fine tracking unit is to further suppress line-of-sight errors. The tracking accuracy achieved by the fine tracking unit is the final accuracy of the composite mechanism APT system. To obtain higher tracking accuracy, it is necessary to increase the resonant frequency of the fine tracking servo mechanism and the sampling frequency of the fine tracking detector element, matching the resonant frequency of the fine tracking servo mechanism with the sampling frequency of the fine tracking detector. A frequency greater than 1000Hz is used to form a broadband and stable active photoelectric tracking unit. In this embodiment of the invention, the fine tracking detector uses QD detection, which drives the fine tracking servo mechanism after processing by the image processing unit. Once the APT system achieves precise dynamic tracking, it means that the communication transmitting line of sight and the communication receiving line of sight have established a high-precision coaxial state, which can then be used in conjunction with advance calibration for small-beam divergence laser communication.

[0050] III. Achieving rapid acquisition and tracking of light spots based on a high dynamic target detection and tracking neural network.

[0051] like Figure 5As shown, in this embodiment of the invention, the tracking detector and optical narrowband filter in the APT system are deployed with an AI accelerator capable of executing a high dynamic target detection and tracking neural network. The accelerator executes the inference process of YOLOvx+DeepSORT target detection and trajectory tracking, which is used to process the position and accuracy data of the beacon light spot during the acquisition and tracking process. In conjunction with coarse tracking and fine tracking units, it realizes rapid acquisition and tracking of the beacon light spot, thereby reducing laser optomechanical alignment error and alignment time, and ensuring that rapid dynamic tracking, link establishment and communication of spatial lasers can be achieved within a limited time.

[0052] An AI accelerator is a custom-designed dedicated circuit system that accelerates artificial intelligence computation. Compared to traditional CPUs and GPUs, AI accelerators have matrix and cache structures more adapted to AI models and can perform point-to-point compression of data formats, enabling them to process more image data per second. In the coarse and fine tracking stages of an APT system, the AI ​​accelerator processes the image data input from a CCD or QD detector, accelerating the tracking and aiming speed of the APT system. The principle of AI accelerator spot detection is as follows... Figure 6 As shown.

[0053] High dynamic range (HDR) target detection and tracking neural networks address the questions of "where is the target" and "where is it going?". Target detection solves the "where is the target?" problem, while target tracking solves the "where is it going?" problem. Target detection is commonly used for coordinate detection and discrimination of fixed-type targets, while target tracking is commonly used for trajectory determination and target locking of specific targets. Details of the target detection and trajectory tracking networks are shown in Table 1. This embodiment of the invention combines target detection and trajectory tracking, and deploying them in an AI accelerator can more accurately capture beacon light position information. The AI ​​accelerator processes the beacon light image data, reducing system errors while improving the tracking speed of the APT system.

[0054] Table 1. YOLO+DeepSORT Network Involved in This Invention For the YOLOvx+DeepSORT algorithm, YOLOvx (such as YOLOv5s) is responsible for finding the spot targets in each frame (capture), while DeepSORT associates these spot targets in time and assigns a unique ID to the same spot target for tracking.

[0055] The YOLOvx processing flow is as follows: Step A: Mesh Generation: First, adjust the input image to be detected to a fixed size (e.g., 640×640 pixels) and divide it evenly into S×S grid cells. Each grid cell is responsible for detecting the center point of the light spot target within its region. The image to be detected is the output image of either a coarse tracking detector or a fine tracking detector.

[0056] Step B: Bounding Box and Class Prediction: Each grid cell independently predicts a fixed number (e.g., 3) of bounding boxes. For each predicted bounding box, five core parameters are regressed: center coordinates (x, y), width (w), height (h), and confidence that the box contains a spot.

[0057] Step C: Multi-scale feature fusion: To improve the detection capability of small targets, independent predictions are made on feature maps of three different scales, corresponding to the detection of the first target, the second target, and the third target, which are large, medium, and small targets in descending order.

[0058] Step D: Result Filtering and Output: After generating a large number of redundant pre-boundary bounding boxes, post-processing is performed using Non-Maximum Suppression (NMS) technology. This process sorts all predicted bounding boxes according to their confidence scores and removes adjacent bounding boxes whose Intersection over Union (IoU) with the highest bounding box is higher than a preset threshold (e.g., 0.5). Finally, unique and accurate spot location information is output.

[0059] The core of the DeepSORT algorithm lies in fusing the appearance and motion features of the light spot target. Through the following cascaded matching process, a unique and persistent identity (ID) is assigned to the light spot output by the YOLOvx detector in each frame: Step A: Trajectory State Prediction (Kalman Filter): For each existing trajectory, a Kalman filter is used to predict its state in the next frame. This predicts the bounding box position and velocity change of the light spot in the next frame, while simultaneously estimating the uncertainty (covariance matrix) of the predicted state. This prediction step provides prior positional information for subsequent matching.

[0060] Step B: Appearance Feature Extraction (Deep ReID Network): Using a lightweight deep convolutional neural network, a high-dimensional appearance feature vector (such as 128-dimensional or 256-dimensional) is extracted for each detected spot bounding box.

[0061] Step C: Cascade Matching (Hungarian Algorithm): The core innovation of Deep SORT lies in its Cascade Matching strategy. This strategy prioritizes matching the most recently occurring trajectories with high confidence, effectively solving the ID switching problem caused by long-term occlusion. Specifically, the algorithm constructs a cost matrix whose elements are composed of motion similarity (Mahanobis distance d). (1) (i,j), which measures the degree of fit between the predicted bounding box and the detected bounding box, and appearance similarity (cosine distance d). (2)(i,j), a weighted sum that measures the similarity of appearance feature vectors, is composed of: C i,j =λd (1) (i,j)+(1 λ)d (2) (i,j) Where λ is the balancing weight coefficient (usually set to 0.98). Subsequently, the Hungarian Algorithm is used to solve the optimal matching problem under this cost matrix, thereby optimally associating the detection results of the current frame with the existing trajectories.

[0062] Step D: Track Management and Update: Based on the matching results, the track is processed in three ways: Successfully matched trajectories: Update the Kalman filter state with the associated bounding box and update the appearance feature library of the trajectory (retaining the feature vectors of the most recent successful matches). Unmatched trajectories: Judged according to a preset threshold (e.g., max_age = 30 frames). If the number of consecutive unmatched frames of a trajectory exceeds the threshold, it is determined that the spot has left the field of view and the trajectory is terminated; otherwise, its prediction state is temporarily retained, waiting for subsequent frames to rematch. Unmatched detection: Initialize new trajectories for these newly appearing spots, and confirm them as official trajectories after 3 consecutive successful matches to filter out false detections.

[0063] Final output: Output the status of all active trajectory spots, including: unique spot ID, current frame bounding box coordinates, target category, and trajectory history information.

[0064] To achieve lightweight design, this embodiment of the invention appropriately prunes the YOLOvx algorithm for the APT (Alternate Threat) space laser communication system and adds a channel attention mechanism, enabling the model to capture target light spots more accurately and quickly. In the DeepSORT architecture, Kalman filtering and the Hungarian algorithm are computational logic; the real computational drain is the ReID feature extraction network. This embodiment of the invention replaces the original ReID network with a smaller CNN (such as MobileNetV2) in the accelerator.

[0065] Compared with the prior art, the various embodiments of the present invention have the following technical effects: 1. This invention, through the deployment of an AI accelerator with a pre-installed dynamic target detection and tracking neural network in the APT system, and the adoption of the YOLOvx+DeepSORT detection and tracking fusion algorithm, significantly reduces the processing latency of beacon spot detection, localization, and tracking, compressing the acquisition and alignment time of the traditional APT system to the millisecond level, and significantly improving the real-time performance of spot acquisition and tracking. It also maximizes the effective transmission time of inter-satellite laser communication, ensuring long-term dynamic and stable aiming of the laser link.

[0066] 2. This invention combines servo control with an AI accelerator, enabling high-precision locking of beacon beams. Even under complex conditions such as vibration, attitude disturbances, and relative motion on the satellite platform, it maintains tracking accuracy at the microradian (μrad) level. This technology not only enhances the anti-interference capability of space laser communication systems but also provides crucial support for the future construction of high-throughput, low-latency space-based communication networks.

[0067] 3. The satellite laser communication APT technology based on the dynamic target detection and tracking neural network of the AI ​​accelerator can simultaneously track, detect and locate multiple light spots. It is not only suitable for one-to-one node laser link access, but also for space multi-node multi-target laser link establishment APT system, laying the technical foundation for the next generation of space laser phased array antennas and multi-satellite networking laser communication.

[0068] 4. The AI ​​accelerator is reconfigurable and upgradeable in orbit, adaptable to signal processing from various types of detectors such as high-resolution CCDs and QDs, without requiring hardware modifications, and compatible with multiple APT architectures. The accelerator features lightweight deployment and low-power operation, effectively meeting the stringent size, power consumption, and heat dissipation constraints of satellite payloads.

[0069] 5. Breaking through the bottlenecks of traditional algorithms that rely on manually set thresholds and have poor dynamic adaptability, the AI ​​accelerator performs model inference and can autonomously learn the characteristics of light spots in complex spatial environments. It has stronger tracking robustness in highly dynamic and multi-perturbation scenarios, meeting the laser rapid link establishment requirements of highly mobile satellite nodes in 6G space-based carrier networks.

[0070] Example 2 This invention provides a space laser communication system, which includes the laser communication APT system as described in any one of Embodiment 1.

[0071] In the specific implementation process of Example 2, reference can be made to Example 1, and it has the corresponding technical effects.

[0072] Although the invention has been described with reference to exemplary embodiments, it should be understood that the invention is not limited to the disclosed exemplary embodiments. Various adjustments or changes may be made to the exemplary embodiments described in this specification without departing from the scope or spirit of the invention. The scope of the claims should be interpreted in the broadest possible sense to cover all modifications and equivalent structures and functions.

Claims

1. A laser communication APT system based on an AI accelerator, characterized in that, The laser communication APT system includes: The coarse tracking unit outputs the detected beacon light spot image to the AI ​​accelerator; responds to the beacon light spot position information fed back by the AI ​​accelerator, and controls the beacon light spot to enter the tracking field of view of the fine tracking unit according to the beacon light spot position information; The AI ​​accelerator is used to track beacon beams based on beacon beam images input by a coarse tracking unit or a fine tracking unit, and to compensate for the tracked beacon beam position information based on tracking error information; the beacon beam position information is either the beacon beam position information detected by the coarse tracking unit or the beacon beam center position information detected by the fine tracking unit. The fine tracking unit is used to output the detected beacon light spot image to the AI ​​accelerator; it responds to the beacon light spot center position information fed back by the AI ​​accelerator, and controls the detected beacon light spot to enter the tracking field of view of the communication receiving field of view according to the beacon light spot center position information, so as to establish the coaxial state of the tracking line of view and the communication receiving line of view.

2. The laser communication APT system according to claim 1, characterized in that, The AI ​​accelerator is pre-installed with a dynamic target detection and tracking neural network and a compensation unit; The neural network is used to perform spot detection and trajectory tracking on the beacon spot image input by the coarse tracking unit or the fine tracking unit, and to track the position information of the beacon spot; The compensation unit is used to compensate for the position information of the tracked beacon spot based on the tracking error information.

3. The laser communication APT system according to claim 2, characterized in that, The dynamic target detection and tracking neural network is mainly constructed using the YOLOvx algorithm and the DeepSORT algorithm; the tracking error information includes spot detection error information, platform vibration residual information, and dynamic tracking error information.

4. The laser communication APT system according to claim 1, characterized in that, The tracking detector in the coarse tracking unit is a CCD detector; the operating frame rate of the CCD detector is more than 10 times the servo bandwidth of the servo mechanism in the coarse tracking unit.

5. The laser communication APT system according to claim 1, characterized in that, The tracking detector in the fine tracking unit is a four-quadrant detector (QD) or a position-sensitive detector (PSD); the resonant frequency of the servo mechanism in the fine tracking unit is matched with the sampling frequency of the tracking detector to form a broadband and stable fine tracking unit.

6. The laser communication APT system according to any one of claims 1-5, characterized in that, The laser communication APT system also includes a capture unit; The capture unit is used to capture the far-field beacon light spot using an AI accelerator.

7. The laser communication APT system according to claim 6, characterized in that, The laser communication APT system also includes a lead adjustment unit; The lead adjustment unit is used to align the tracking line of sight, the communication transmission line of sight, and the communication reception line of sight.

8. The laser communication APT system according to claim 6, characterized in that, The tracking detector in the coarse tracking unit is used to detect the beacon light spot captured by the capture unit and output the beacon light spot image to the AI ​​accelerator; The tracking controller in the coarse tracking unit is used to respond to the beacon light spot position information fed back by the AI ​​accelerator and send control signals according to the beacon light spot position information; The servo mechanism in the coarse tracking unit is used to control the beacon beam spot to enter the tracking field of view of the fine tracking unit according to the control signal of the tracking controller in the coarse tracking unit.

9. The laser communication APT system according to claim 6, characterized in that, The tracking detector in the fine tracking unit is used to detect the beacon light spot entering its tracking field of view and output the beacon light spot image to the AI ​​accelerator. The tracking controller in the fine tracking unit is used to respond to the beacon beam spot center position information fed back by the AI ​​accelerator and send control signals according to the beacon beam spot center position information; The servo mechanism in the fine tracking unit is used to control the beacon light spot to enter the tracking field of view of the communication receiving field of view according to the control signal of the tracking controller in the fine tracking unit, so as to establish the coaxial state of the tracking line of view and the communication receiving line of view.

10. A space laser communication system, characterized in that, The space laser communication system includes the laser communication APT system as described in any one of claims 1-9.

Images