Micro-nano satellite swarm cooperative system and method

Abstract

The present application relates to the technical field of satellite communication, and proposes a micro-nano satellite constellation coordination system and a coordination method, the micro-nano satellite constellation coordination system comprising a multi-modal perception and fusion module, a frequency offset compensation and hybrid communication module, a resource game allocation module, and a lightweight intelligent decision execution module. A Doppler frequency offset compensation mechanism and LoRa+WiFi hybrid networking are introduced, which reduces the inter-satellite link load by ≥40% and the end-to-end delay jitter by ≥30% under equivalent traffic, ensuring the link connectivity rate under high-speed relative motion. Through a lightweight model distillation and quantization scheme, decision reasoning with a delay of <10 ms is successfully realized under the micro-nano satellite computing power constraint of ≤1 TOPS.

Description

Technical Field

[0001] This invention relates to the technical field of satellite communication, specifically to a micro / nano satellite constellation collaborative system and collaborative method. Background Technology

[0002] Currently, low-Earth orbit (LEO) microsatellites and nanosatellites are widely used in commercial spaceflight and Earth observation. Existing microsatellite operating modes primarily rely on a "centralized ground-based scheduling and data processing" architecture. The basic data processing procedures and connectivity relationships are as follows: Communication network architecture: Satellites typically carry traditional UHF / VHF band transceivers for telemetry and remote control, or use S / X bands for service data downlink. Inter-satellite links (ISLs) mostly employ a single radio frequency communication system, relaying data through a pre-set static routing table or a simple star topology.

[0003] Sensing and processing: After the sensors on the satellite (such as optical cameras) collect raw images or environmental data, they store them in full on the onboard solid-state drive.

[0004] Coordination and scheduling mechanism: When a satellite passes over a ground station, it centrally transmits massive amounts of raw data. The ground control center uses high-performance servers to centrally process the data, extract features, and plan tasks. Subsequently, it sends the next orbital maneuver, attitude adjustment, or collaborative observation commands to each satellite node via uplink.

[0005] While existing technologies can achieve basic satellite operation, they suffer from the following serious drawbacks when dealing with large-scale, highly dynamic micro- and nano-satellite constellation coordination scenarios: Inter-satellite links are extremely unstable and prone to disconnection because low-Earth orbit satellites have very low orbital altitudes and extremely high relative speeds (up to several kilometers per second), resulting in significant Doppler frequency offsets in inter-satellite communication. Traditional terrestrial commercial communication protocols (such as standard LoRa or WiFi) have not been optimized at the underlying physical layer for extremely dynamic scenarios, and frequency offsets can cause the receiver to be unable to perform phase-locked demodulation, resulting in frequent link interruptions.

[0006] The "sensor-download" model results in severe latency and bandwidth bottlenecks: Reason: Microsatellites have extremely limited downlink bandwidth resources, while the amount of raw multimodal data (especially high-resolution images) is enormous. Centralized downlink transmission not only leads to network congestion, but also makes the closed-loop cycle from "event occurrence" to "ground decision and uplink again" take several hours, which is simply unable to meet the demand for high-timeliness response at the minute level (such as disaster emergency monitoring and space debris avoidance).

[0007] Centralized scheduling lacks robustness and cannot achieve autonomous satellite constellation coordination because existing scheduling relies heavily on ground stations. Once a satellite is in a telemetry and control blind zone, it becomes an information island. In addition, micro and nano satellites are limited by size and power consumption, resulting in extremely low onboard computing power (usually ≤1 TOPS). Traditional complex artificial intelligence models cannot run on satellites, causing the satellites themselves to lack the decision-making capabilities for multi-target collaborative allocation and autonomous obstacle avoidance. Summary of the Invention

[0008] This invention proposes a micro-nano satellite constellation collaborative system and collaborative method, aiming to construct a new paradigm of constellation collaboration that replaces hardware stacking with algorithm optimization.

[0009] A micro / nano satellite constellation coordination system designed for this purpose includes: The multimodal sensing and fusion module is used to collect local multimodal sensing data of the satellite, perform feature extraction and feature-level fusion on the multimodal sensing data to obtain a fused feature vector, and generate a trigger signal based on the comparison result of the fused feature vector and the preset reference information. The frequency offset compensation and hybrid communication module exchanges status information between satellite nodes through LoRa links, constructs and updates the constellation topology based on the status information, and estimates and compensates for the frequency offset of the communication carrier of the inter-satellite links. The resource game allocation module is used to respond to the trigger signal generated by the multimodal perception and fusion module. For the satellite cluster collaborative task, each satellite node acts as an independent intelligent agent and initiates resource bidding based on the remaining power, computing power and observation window of the satellite node. The Nash equilibrium allocation of computing, communication and observation resources among the satellite cluster is achieved through the game mechanism. The lightweight intelligent decision-making and execution module combines the resource allocation results of the resource game allocation module with the topology status of the frequency offset compensation and hybrid communication module to output instructions, thereby enabling collaborative scheduling and autonomous obstacle avoidance of multiple satellites. The multimodal perception and fusion module, frequency offset compensation and hybrid communication module, resource game allocation module and lightweight intelligent decision execution module communicate and connect with each other.

[0010] Each satellite carries a set of multimodal sensors for real-time acquisition of space environment and satellite attitude data; The multimodal sensor group includes a miniature optical camera, a MEMS inertial navigation unit, and a temperature and humidity sensor.

[0011] The hardware modules carried by each satellite also include a satellite edge computing payload control board with its internal operating feature fusion algorithm, resource game algorithm and lightweight MARL inference engine. This satellite edge computing payload control board communicates with the multimodal sensor group through its internal bus. The multimodal sensor array and the satellite edge computing payload control board together form a multimodal sensing and fusion module.

[0012] The hardware modules carried by each satellite also include a hybrid radio frequency communication payload control board for forming a frequency offset compensation and hybrid communication module. It has a built-in LoRa radio frequency front-end and WiFi radio frequency front-end, as well as a baseband processing FPGA. The baseband processing FPGA has a Doppler frequency offset compensation logic circuit programmed inside. The hybrid radio frequency communication payload control board and the multimodal sensing are connected to the satellite edge computing payload control board of the fusion module through the SPI / UART interface to be responsible for the interaction and topology maintenance between satellites.

[0013] Each satellite carries a hardware module that includes an actuator. The actuator is connected to the satellite edge computing payload control board of the fusion module, receives maneuver commands from the lightweight intelligent decision execution module, and drives the satellite to perform attitude adjustment and avoidance maneuvers.

[0014] A collaborative method for a micro / nano satellite constellation collaborative system as described above includes the following steps: Step 1: The multimodal perception and fusion module acquires multimodal data at the satellite edge and performs lightweight feature extraction and feature-level fusion. Step 2: To address the time-varying and intermittent reachability characteristics of inter-satellite links, a dynamic topology network communication protocol between satellites is designed using frequency offset compensation and a hybrid communication module. Step 3: Execute resource auctions and game-based allocation driven by each satellite mission through the resource game allocation module; Step 4: By combining the resource allocation results from Step 3 with the topology status from Step 2 through the lightweight intelligent decision-making and execution module, a closed-loop instruction of "perception-decision-execution" is output to realize the coordinated scheduling and autonomous obstacle avoidance of multiple satellites.

[0015] In step one, a satellite event-driven coding mechanism is introduced. The multimodal perception and fusion module generates downlink data or inter-satellite interaction instructions only when the state of the fused features changes abruptly or a specific task threshold is met. This filters redundant background data and reduces the amount of data transmitted from the ground. Among them, the feature difference value is defined. The calculation method uses the Euclidean distance or weighted difference of the feature vectors: ΔE = ||F_current - F_baseline||_2; The triggering condition is: ΔE≥Th_event, F_current is the currently extracted fusion feature vector, F_baseline is the background baseline feature vector, the background baseline feature vector is a pre-set feature vector used to characterize the typical features of multimodal sensing data of the satellite under normal operating conditions, and serves as a benchmark for judging whether the current sensing data contains the target mission or abnormal situation, and Th_event is a dynamically set event triggering threshold.

[0016] In step two, for low-Earth orbit high-dynamic scenarios, LoRa / WiFi Doppler frequency offset estimation and compensation algorithms are introduced through frequency offset compensation and hybrid communication modules. This is to overcome the frequency offset loss-of-lock problem caused by high-speed relative motion through frequency offset pre-compensation and carrier tracking in the digital domain. LoRa is used for long-distance topology heartbeat maintenance between satellites, and WiFi is used for short-distance high-bandwidth data transmission to construct a dynamic topology graph neural network. Doppler frequency deviation The classic computational model: ; Pre-compensation frequency at the transmitting end for: .

[0017] In step three, for collaborative tasks among satellites and for each satellite node as an independent intelligent agent, resource bidding is initiated based on the satellite's remaining power, computing power and observation window. The Nash equilibrium allocation of computing, communication and observation resources among satellites is achieved through a game mechanism. Define the formula for calculating the "utility / bid price" for satellites participating in the bidding: ,in For satellite The bidding utility value; The remaining computing power; Remaining battery level; For satellite With the mission initiator Communication delay or distance between them; These are the weighting coefficients.

[0018] In step four, the lightweight MARL model, which has undergone model distillation and quantization compression, is deployed on the onboard computing node with a computing power of ≤1 TOPS. Define the reward function for the agent satellite in its interaction with the environment. : ,in To achieve positive benefits in completing collaborative observation tasks; Penalties for power consumption during motorized or communication operations; Extreme penalties for triggering obstacle avoidance danger distance; To adjust the weights.

[0019] The beneficial technical effects of the present invention are as follows: 1. Overcoming the challenges of high dynamic communication: By introducing a Doppler frequency offset compensation mechanism and LoRa+WiFi hybrid networking, the inter-satellite link load is reduced by ≥40% and the end-to-end latency jitter is reduced by ≥30% under equivalent service volume, ensuring link connectivity under high-speed relative motion.

[0020] 2. Breakthrough in extremely low computing power bottleneck: Through lightweight model distillation and quantization schemes, decision reasoning with a latency of <10 ms was successfully achieved under the constraint of micro-nano satellite computing power of ≤1 TOPS.

[0021] 3. Greatly alleviates bandwidth pressure: Relying on the feature-level fusion and event-driven coding mechanism at the satellite edge, the amount of data transmitted from the ground has been reduced by ≥70%, meeting the needs of minute-level emergency response.

[0022] 4. Achieve highly reliable on-orbit autonomous collaboration: In ground-based semi-physical closed-loop simulation (≥50 node scale) and engineering prototype verification, highly reliable multi-target collaborative observation and autonomous obstacle avoidance were achieved, with an obstacle avoidance success rate of ≥92%. Attached Figure Description

[0023] Figure 1 This is a flowchart of the intelligent collaborative method for micro-nano satellite constellations of the present invention, which shows the four core steps and closed-loop feedback from multi-source sensing fusion, frequency offset compensation and hybrid networking, resource game theory to lightweight MARL decision-making.

[0024] Figure 2 This diagram illustrates the LoRa+WiFi hybrid networking architecture of this invention, which includes Doppler frequency offset compensation, showing the communication links between constellation nodes. The relative velocity vectors between nodes and the frequency offset estimation and compensation modules included in the physical layer are specifically marked. Dashed lines represent LoRa links, and solid lines represent WiFi links.

[0025] Figure 3 This is a flowchart illustrating the principle of the collaborative method based on MARL and dynamic topology of this invention.

[0026] Figure 4 This is a system architecture diagram of the micro-nano satellite constellation collaborative system module of the present invention. Detailed Implementation

[0027] The technical solutions of the present invention will be clearly and completely described below with reference to the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. In order to make the above-mentioned objects, features and advantages of this application more apparent and understandable, many specific details are set forth in the following description in order to provide a full understanding of this application. However, this application can be implemented in many other ways different from those described herein, and those skilled in the art can make similar improvements without departing from the spirit of this application. Therefore, this application is not limited to the specific embodiments disclosed below.

[0028] See Figures 1-4 A micro / nano satellite constellation collaborative system, comprising: The multimodal sensing and fusion module is used to collect local multimodal sensing data of the satellite, perform feature extraction and feature-level fusion on the multimodal sensing data to obtain a fused feature vector, and generate a trigger signal based on the comparison result of the fused feature vector and the preset reference information. The frequency offset compensation and hybrid communication module exchanges status information between satellite nodes through LoRa links, constructs and updates the constellation topology based on the status information, and estimates and compensates for the frequency offset of the communication carrier of the inter-satellite links. The resource game allocation module is used to respond to the trigger signal generated by the multimodal perception and fusion module. For the satellite cluster collaborative task, each satellite node acts as an independent intelligent agent and initiates resource bidding based on the remaining power, computing power and observation window of the satellite node. The Nash equilibrium allocation of computing, communication and observation resources among the satellite cluster is achieved through the game mechanism. The lightweight intelligent decision-making and execution module combines the resource allocation results of the resource game allocation module with the topology status of the frequency offset compensation and hybrid communication module to output instructions, thereby enabling collaborative scheduling and autonomous obstacle avoidance of multiple satellites. The multimodal perception and fusion module, frequency offset compensation and hybrid communication module, resource game allocation module, and lightweight intelligent decision execution module are interconnected. A complete constellation collaborative system architecture has been constructed, achieving closed-loop autonomous collaboration from perception to execution through the organic cooperation of multimodal perception, frequency offset compensation communication, resource game, and lightweight decision modules.

[0029] Each satellite carries a set of multimodal sensors for real-time acquisition of space environment and satellite attitude data; The multimodal sensor group includes a miniature optical camera, a MEMS inertial navigation unit, and a temperature and humidity sensor. The combination of the miniature optical camera, IMU, and temperature and humidity sensor enables multi-source acquisition of space environment and satellite attitude data, providing a rich data foundation for subsequent feature fusion.

[0030] The hardware modules carried by each satellite also include a satellite edge computing payload control board with its internal operating feature fusion algorithm, resource game algorithm and lightweight MARL inference engine. This satellite edge computing payload control board communicates with the multimodal sensor group through its internal bus. The multimodal sensor array and the satellite edge computing payload control board together form a multimodal sensing and fusion module.

[0031] The satellite edge computing payload control board uses low-power AI chips (such as NPU / MCU with computing power limited to around 1 TOPS). By deploying low-power AI chips locally on the satellite and running feature fusion, resource game theory, and MARL inference algorithms, the limitations of onboard computing power are overcome, enabling real-time on-orbit processing.

[0032] The hardware modules carried by each satellite also include a hybrid radio frequency communication payload control board for forming a frequency offset compensation and hybrid communication module. It has a built-in LoRa radio frequency front-end and WiFi radio frequency front-end, as well as a baseband processing FPGA. The baseband processing FPGA has a Doppler frequency offset compensation logic circuit programmed inside. The hybrid radio frequency communication payload control board and the multimodal sensing are connected to the satellite edge computing payload control board of the fusion module through the SPI / UART interface to be responsible for the interaction and topology maintenance between satellites.

[0033] It integrates LoRa and WiFi dual links and FPGA frequency offset compensation circuit to achieve long-distance topology maintenance and short-distance high-speed transmission, overcoming the link loss problem caused by high dynamic Doppler frequency offset.

[0034] Each satellite's onboard hardware modules also include actuators. These actuators are communicatively connected to the satellite edge computing payload control board of the fusion module, receiving maneuvering commands from the lightweight intelligent decision-making and execution module to drive the satellite to perform attitude adjustments and avoidance maneuvers. By receiving decision commands and executing attitude adjustments and avoidance maneuvers through momentum wheels and magnetic torque converters, a closed-loop control system from decision-making to physical action is completed.

[0035] A collaborative method for a micro / nano satellite constellation collaborative system as described above includes the following steps: Step 1: The multimodal perception and fusion module acquires multimodal data at the satellite edge and performs lightweight feature extraction and feature-level fusion. Step 2: To address the time-varying and intermittent reachability characteristics of inter-satellite links, a dynamic topology network communication protocol between satellites is designed using frequency offset compensation and a hybrid communication module. Step 3: Execute resource auctions and game-based allocation driven by each satellite mission through the resource game allocation module; Step 4: By combining the resource allocation results from Step 3 with the topology status from Step 2 through the lightweight intelligent decision-making and execution module, a closed-loop instruction of "perception-decision-execution" is output to realize the coordinated scheduling and autonomous obstacle avoidance of multiple satellites.

[0036] The system functions are mapped into four steps, forming a complete process of "perception-communication-game-decision-execution", which enables autonomous collaboration of the constellation.

[0037] In step one, a satellite event-driven coding mechanism is introduced. The multimodal perception and fusion module generates downlink data or inter-satellite interaction instructions only when the state of the fused features changes abruptly or a specific task threshold is met. This filters redundant background data and reduces the amount of data transmitted from the ground. Among them, the feature difference value is defined. The calculation method uses the Euclidean distance or weighted difference of the feature vectors: ΔE = ||F_current - F_baseline||_2; The triggering condition is: ΔE≥Th_event, F_current is the currently extracted fusion feature vector, F_baseline is the background baseline feature vector, the background baseline feature vector is a pre-set feature vector used to characterize the typical features of multimodal sensing data of the satellite under normal operating conditions, and serves as a benchmark for judging whether the current sensing data contains the target mission or abnormal situation, and Th_event is a dynamically set event triggering threshold.

[0038] The feature difference is calculated by ΔE=‖F_current-F_baseline‖_2, and the interaction is triggered only when ΔE≥Th_event, reducing the amount of ground-based downlink data by ≥70% and meeting the minute-level emergency response requirements.

[0039] In step two, for low-Earth orbit high-dynamic scenarios, LoRa / WiFi Doppler frequency offset estimation and compensation algorithms are introduced through frequency offset compensation and hybrid communication modules. This is to overcome the frequency offset loss-of-lock problem caused by high-speed relative motion through frequency offset pre-compensation and carrier tracking in the digital domain. LoRa is used for long-distance topology heartbeat maintenance between satellites, and WiFi is used for short-distance high-bandwidth data transmission to construct a dynamic topology graph neural network. Doppler frequency deviation The classic computational model: ; Pre-compensation frequency at the transmitting end for: .

[0040] Detailed Explanation of Key Steps in Doppler Frequency Offset Compensation and Hybrid Networking: In step two, due to the relative velocity of the satellite The maximum value will cause a Doppler frequency shift in the received signal frequency. .

[0041] Frequency Offset Estimation and Compensation: The baseband FPGA continuously monitors the phase deflection rate of the preamble. Using the Extended Kalman Filter (EKF) algorithm combined with shared ephemeris data from the local and neighboring satellites, the relative radial velocity is predicted, and the estimated frequency offset is calculated. Carrier frequency pre-compensation (reverse offset) is performed at the transmitting end. At the receiving end, a phase-locked loop (PLL) is used to correct residual frequency offset, ensuring that LoRa and WiFi signals can still be correctly demodulated when they meet at high speed.

[0042] Hybrid networking strategy: When the distance between two satellites is greater than 10km, the edge computing payload controls the communication module to turn off WiFi and only use LoRa to send a "topology heartbeat packet" containing its own ID, location, and remaining computing power; when the distance is less than 10km and there is a need for collaborative computing or image transmission, the WiFi module is woken up to establish a broadband connection. This mechanism greatly reduces communication power consumption and maintains the dynamic topology graph (GNN) of the entire network.

[0043] By calculating and pre-compensating for frequency offset, and combining LoRa (>10km heartbeat) and WiFi (<10km data transmission) hybrid networking, the inter-satellite link load is reduced by ≥40% and the latency jitter is reduced by ≥30%.

[0044] In step three, for collaborative tasks among satellites and for each satellite node as an independent intelligent agent, resource bidding is initiated based on the satellite's remaining power, computing power and observation window. The Nash equilibrium allocation of computing, communication and observation resources among satellites is achieved through a game mechanism. Define the formula for calculating the "utility / bid price" for satellites participating in the bidding: ,in For satellite The bidding utility value; The remaining computing power; Remaining battery level; For satellite With the mission initiator Communication delay or distance between them; These are the weighting coefficients.

[0045] Calculate the utility of bidding to achieve Nash equilibrium allocation of inter-satellite computing, communication and observation resources, thereby improving resource utilization.

[0046] In step four, the lightweight MARL model, which has undergone model distillation and quantization compression, is deployed on the onboard computing node with a computing power of ≤1 TOPS. Define the reward function for the agent satellite in its interaction with the environment. : ,in To achieve positive benefits in completing collaborative observation tasks; Penalties for power consumption during motorized or communication operations; This is an extreme penalty for triggering the obstacle avoidance danger distance. To adjust the weights.

[0047] Event-driven and MARL-based collaborative decision-making process: Event Triggered: Satellite A's optical camera captures what appears to be space debris. The edge computing payload extracts image features and fuses them with IMU data. The difference between the current features and the background baseline is calculated. .when When the value exceeds a set threshold, an "obstacle avoidance alarm" event is triggered instead of downloading the entire image.

[0048] Resource Game Theory: Satellite A discovers its computing power is insufficient to accurately calculate the orbital parameters of debris, so it broadcasts a "computing power demand auction" to surrounding satellites via LoRa. Satellites B and C bid based on their remaining power and current mission priorities. Through Nash equilibrium calculation, Satellite B, with the most abundant computing power, wins the bid. Satellite A then transmits the compressed feature vector to Satellite B via WiFi.

[0049] Lightweight MARL Execution: The lightweight MARL model on Satellite B (compressed to less than 2MB on the ground through knowledge distillation) receives the feature vectors and the current constellation topology, and infers the optimal obstacle avoidance strategy within <10ms. Satellite B transmits the maneuver commands back to Satellite A, and Satellite A's actuators (momentum wheels) are then activated to complete autonomous obstacle avoidance. The entire process requires no ground station intervention, achieving true constellation intelligent collaboration.

[0050] Explanation of related terms: Dynamic Topology Network Communication Protocol: This protocol is an adaptive routing and link establishment rule designed to address the drastic changes in inter-satellite distances and relative positions caused by the high-speed motion of low-Earth orbit satellites. It can autonomously determine the establishment, maintenance, and disconnection of links based on real-time distances and signal quality between satellites, without requiring a pre-set static routing table on the ground.

[0051] LoRa / WiFi Doppler Frequency Offset Estimation and Compensation Algorithm: Due to the extremely high relative speed of satellites (up to several kilometers per second), electromagnetic wave transmission will produce a severe Doppler effect, resulting in frequency offset. This algorithm uses an extended Kalman filter (EKF) combined with ephemeris data to estimate the relative radial velocity and calculate the frequency offset value. At the transmitting end, digital domain reverse pre-compensation (offset) of the carrier frequency is performed. At the receiving end, a phase-locked loop (PLL) is used to correct residual frequency offset, ensuring signal demodulation during high-speed rendezvous.

[0052] Dynamic Topology Graph Neural Network (GNN): This approach treats each microsatellite or nanosatellite in a constellation as a "node" in a graph, and inter-satellite communication links as "edges." Due to satellite motion, the structure of this graph changes dynamically over time. GNN is used to extract features from this spatiotemporally dynamic network, helping satellites perceive the global topology and providing environmental state input for subsequent resource allocation.

[0053] Game Theory Mechanism (Nash Equilibrium): In a constellation with limited resources, each satellite is considered a rational individual seeking to maximize its own mission gains. By introducing an auction model (such as a Vickrey auction), satellites bid for computing / communication resources based on their remaining power, computing power, and mission urgency, ultimately reaching a stable state (Nash equilibrium) where none of the parties can increase their gains by unilaterally changing their strategies, thus achieving optimal distributed allocation of resources.

[0054] Multi-Agent Reinforcement Learning (MARL): An artificial intelligence algorithm architecture. In this system, each microsatellite or nanosatellite is an independent "agent." Through interaction with the space environment and other satellites, they learn through trial and error (reward / punishment mechanism), gradually optimizing their own decision-making strategies (such as attitude adjustment and obstacle avoidance maneuvers), ultimately maximizing the overall collaborative utility of the constellation.

[0055] The above are merely preferred embodiments of the present invention and are not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A micro / nano satellite constellation collaborative system, characterized in that, include: The multimodal sensing and fusion module is used to collect local multimodal sensing data of the satellite, perform feature extraction and feature-level fusion on the multimodal sensing data to obtain a fused feature vector, and generate a trigger signal based on the comparison result of the fused feature vector and the preset reference information. The frequency offset compensation and hybrid communication module exchanges status information between satellite nodes through LoRa links, constructs and updates the constellation topology based on the status information, and estimates and compensates for the frequency offset of the communication carrier of the inter-satellite links. The resource game allocation module is used to respond to the trigger signal generated by the multimodal perception and fusion module. For the satellite cluster collaborative task, each satellite node acts as an independent intelligent agent and initiates resource bidding based on the remaining power, computing power and observation window of the satellite node. The Nash equilibrium allocation of computing, communication and observation resources among the satellite cluster is achieved through the game mechanism. The lightweight intelligent decision-making and execution module combines the resource allocation results of the resource game allocation module with the topology status of the frequency offset compensation and hybrid communication module to output instructions, thereby enabling collaborative scheduling and autonomous obstacle avoidance of multiple satellites. The multimodal perception and fusion module, frequency offset compensation and hybrid communication module, resource game allocation module and lightweight intelligent decision execution module communicate and connect with each other.

2. The micro / nano satellite constellation collaborative system according to claim 1, characterized in that: Each satellite carries a set of multimodal sensors for real-time acquisition of space environment and satellite attitude data; The multimodal sensor group includes a miniature optical camera, a MEMS inertial navigation unit, and a temperature and humidity sensor.

3. The micro / nano satellite constellation coordination system according to claim 2, characterized in that: The hardware modules carried by each satellite also include a satellite edge computing payload control board with its internal operating feature fusion algorithm, resource game algorithm and lightweight MARL inference engine. This satellite edge computing payload control board communicates with the multimodal sensor group through its internal bus. The multimodal sensor array and the satellite edge computing payload control board together form a multimodal sensing and fusion module.

4. The micro / nano satellite constellation coordination system according to claim 2, characterized in that: The hardware modules carried by each satellite also include a hybrid radio frequency communication payload control board for forming a frequency offset compensation and hybrid communication module. It has a built-in LoRa radio frequency front-end and WiFi radio frequency front-end, as well as a baseband processing FPGA. The baseband processing FPGA has a Doppler frequency offset compensation logic circuit programmed inside. The hybrid radio frequency communication payload control board and the multimodal sensing are connected to the satellite edge computing payload control board of the fusion module through the SPI / UART interface to be responsible for the interaction and topology maintenance between satellites.

5. The micro / nano satellite constellation coordination system according to claim 2, characterized in that: Each satellite carries a hardware module that includes an actuator. The actuator is connected to the satellite edge computing payload control board of the fusion module, receives maneuver commands from the lightweight intelligent decision execution module, and drives the satellite to perform attitude adjustment and avoidance maneuvers.

6. A collaborative method for a micro / nano satellite constellation collaborative system as described in any one of claims 1-5, characterized in that, Includes the following steps: Step 1: The multimodal perception and fusion module acquires multimodal data at the satellite edge and performs lightweight feature extraction and feature-level fusion. Step 2: To address the time-varying and intermittent reachability characteristics of inter-satellite links, a dynamic topology network communication protocol between satellites is designed using frequency offset compensation and a hybrid communication module. Step 3: Execute resource auctions and game-based allocation driven by each satellite mission through the resource game allocation module; Step 4: By combining the resource allocation results from Step 3 and the topology status from Step 2 with the lightweight intelligent decision-making and execution module, a closed-loop instruction of "perception-decision-execution" is output to realize the coordinated scheduling and autonomous obstacle avoidance of multiple satellites.

7. The collaborative method for a micro / nano satellite constellation collaborative system according to claim 6, characterized in that: In step one, a satellite event-driven coding mechanism is introduced. The multimodal perception and fusion module generates downlink data or inter-satellite interaction instructions only when the state of the fused features changes abruptly or a specific task threshold is met. This filters redundant background data and reduces the amount of data transmitted from the ground. Among them, the feature difference value is defined. The calculation method uses the Euclidean distance or weighted difference of the feature vectors: ΔE = ||F_current - F_baseline||_2; The triggering condition is: ΔE≥Th_event, F_current is the currently extracted fusion feature vector, F_baseline is the background baseline feature vector, the background baseline feature vector is a pre-set feature vector used to characterize the typical features of multimodal sensing data of the satellite under normal operating conditions, and serves as a benchmark for judging whether the current sensing data contains the target mission or abnormal situation, and Th_event is a dynamically set event triggering threshold.

8. The collaborative method for a micro / nano satellite constellation collaborative system according to claim 6, characterized in that: In step two, for low-Earth orbit high-dynamic scenarios, LoRa / WiFi Doppler frequency offset estimation and compensation algorithms are introduced through frequency offset compensation and hybrid communication modules. This is to overcome the frequency offset loss-of-lock problem caused by high-speed relative motion through frequency offset pre-compensation and carrier tracking in the digital domain. LoRa is used for long-distance topology heartbeat maintenance between satellites, and WiFi is used for short-distance high-bandwidth data transmission to construct a dynamic topology graph neural network. Doppler frequency deviation The classic computational model: ； Pre-compensation frequency at the transmitting end for: .

9. The collaborative method for a micro / nano satellite constellation collaborative system according to claim 6, characterized in that: In step three, for collaborative tasks among satellites and for each satellite node as an independent intelligent agent, resource bidding is initiated based on the satellite's remaining power, computing power and observation window. The Nash equilibrium allocation of computing, communication and observation resources among satellites is achieved through a game mechanism. Define the formula for calculating the "utility / bid price" for satellites participating in the bidding: ,in satellite Bidding utility value; The remaining computing power; Remaining battery level; For satellite With the mission initiator Communication delay or distance between them; These are the weighting coefficients.

10. The collaborative method for a micro / nano satellite constellation collaborative system according to claim 6, characterized in that: In step four, the lightweight MARL model, which has undergone model distillation and quantization compression, is deployed on the onboard computing node with a computing power of ≤1 TOPS. Define the reward function for the agent satellite in its interaction with the environment. : ,in To achieve positive benefits in completing collaborative observation tasks; Penalties for power consumption during motorized or communication operations; This is an extreme penalty for triggering the obstacle avoidance danger distance. To adjust the weights.

Images