Drone communication resilience management system supporting multi-link dynamic aggregation and intelligent switching

Abstract

The application discloses a UAV communication resilience management system supporting multi-link dynamic aggregation and intelligent switching, comprising a multi-mode communication hardware layer, a full-dimensional perception module, a link state prediction unit and a dynamic scheduling unit. The full-dimensional perception module collects physical layer parameters, transmission layer state parameters, UAV flight state data and environmental geographic information of each communication link in real time. The link state prediction unit predicts future link connectivity probability and bandwidth using a time series model based on flight state and environmental information, and outputs a quality score. The dynamic scheduling unit generates a strategy based on the score and real-time parameters. Its traffic classification submodule identifies uplink data stream characteristics and marks priority, and the transmission coding submodule performs redundant or fragmented coding accordingly, and maps to different links for parallel transmission, while controlling the hardware layer to perform link activation, hibernation or switching. The system combines prediction and real-time perception to achieve multi-link intelligent aggregation and switching, improving communication reliability and bandwidth utilization.

Description

Technical Field

[0001] This invention relates to the field of unmanned aerial vehicle (UAV) technology, and in particular to a UAV communication resilience management system that supports multi-link dynamic aggregation and intelligent switching. Background Technology

[0002] With the widespread application of drone technology in remote sensing and mapping, emergency communication and logistics delivery, higher requirements have been placed on the stability and bandwidth of its communication system. Existing drone communication systems usually adopt a multi-mode communication hardware architecture, integrating various heterogeneous modules such as 5G, satellite communication and self-organizing networks to establish communication links of different standards. Although some advanced systems have basic link quality monitoring functions, they can manually or automatically switch between different links based on simple indicators such as signal strength or packet loss rate.

[0003] However, in practical applications, this reactive switching mechanism often exhibits significant lag. When drones fly at high speeds in complex urban environments or areas with obstructed terrain, the primary link may be disconnected instantaneously, while the activation and handshake of the backup link takes time, resulting in unacceptable communication gaps. Furthermore, existing data transmission strategies often employ a "best-effort" load balancing approach, failing to differentiate between different services such as control signaling and high-definition video streams generated by the airborne application layer. Critical control commands may be lost due to link congestion or quality fluctuations, severely impacting flight safety and mission continuity. Simultaneously, due to the lack of in-depth fusion analysis of drone flight trajectories, attitude changes, and surrounding environmental geographic information, the system cannot predict the emergence of communication blind spots, making it difficult to pre-connect links or reserve bandwidth in advance. This significantly reduces the efficiency of multi-link aggregation, not only wasting spectrum resources but also failing to meet the stringent requirements for communication resilience in highly dynamic scenarios.

[0004] Therefore, there is an urgent need to develop an intelligent management system that can integrate full-dimensional state perception, link quality prediction, and differentiated service scheduling. By accurately inferring the future link state and encoding and mapping data streams of different priorities, it can achieve seamless aggregation and smooth switching of multiple links, thereby solving the problems of high communication interruption risk, low reliability of critical data transmission, and insufficient resource utilization in existing technologies. Summary of the Invention

[0005] To improve existing methods and systems, a UAV communication resilience management system supporting multi-link dynamic aggregation and intelligent switching is provided, including: The multimode communication hardware layer includes at least two heterogeneous communication modules, which are used to establish communication links of different standards respectively. The all-dimensional perception module is connected to the multi-mode communication hardware layer and the UAV flight control system. The all-dimensional perception module is configured to collect physical layer parameters, transmission layer status parameters, UAV flight status data and environmental geographic information of each communication link in real time. The link status prediction unit is connected to the full-dimensional perception module. The link status prediction unit is configured to infer the connectivity probability and available bandwidth of each communication link within a future preset time window based on UAV flight status data and environmental geographic information, and output the link quality score. The dynamic scheduling unit is connected to the full-dimensional perception module, the link state prediction unit, and the multi-mode communication hardware layer. The dynamic scheduling unit includes a service classification submodule and a transmission coding submodule. The service classification submodule is configured to identify the service characteristics of the uplink data stream generated by the airborne application layer and mark its priority. The transmission coding submodule is configured to perform redundant coding or fragmentation coding on the uplink data stream according to the priority, and map the encoded data packets to different communication links for parallel transmission. The dynamic scheduling unit is also configured to generate scheduling policies based on link quality scores, physical layer parameters, and transport layer status parameters, and control the multi-mode communication hardware layer to perform link activation, link hibernation, or link switching operations on the communication links based on the scheduling policies.

[0006] In some embodiments, the link state prediction unit employs a fusion filtering and neural network algorithm. The input data of the link state prediction unit includes building vector data within a preset radius around the current location of the UAV, historical link quality sequences, and UAV attitude angular velocity. The output of the link state prediction unit is the quality score of each link at a future time T. The dynamic scheduling unit is configured to trigger a pre-connection handshake process on the backup link in the communication link when the quality score is lower than a preset security threshold, and to establish a backup data channel before the primary link in the communication link is interrupted.

[0007] In some embodiments, the service classification submodule is configured to perform deep parsing of the protocol field in the packet header of the uplink data stream, and divide the uplink data stream into a high-priority control stream and a low-priority media stream. The transmission coding submodule employs a full copy strategy for high-priority control flows, enabling the same data packets to be broadcast simultaneously on all active links in all communication links. The transmission coding submodule employs a scalable coding strategy for low-priority media streams to generate base layer data packets and enhancement layer data packets.

[0008] In some embodiments, the transmission coding submodule uses system fountain codes to encode high-priority control flow, so that the receiving end can restore the original control command when it receives any number of encoded packets; The transmission coding submodule uses a modified version of the multipath transmission control protocol MPTCP to process low-priority media streams. The transmission coding submodule establishes multiple substreams corresponding to different communication links and performs out-of-order reordering of data packets through a common sequence number space. Among them, basic layer data packets are mapped to low-bandwidth links in the communication link for transmission, while enhancement layer data packets are mapped to high-bandwidth links in the communication link for transmission.

[0009] In some embodiments, the dynamic scheduling unit further includes a switching decision hysteresis suppression module, which is configured to set a hysteresis threshold and a penalty factor. The dynamic scheduling unit only triggers a link switching operation when the quality score of the primary link is lower than the difference between the quality score of the backup link and the lag threshold. Within a preset cooling-off period after the link switching operation is completed, the switching decision hysteresis suppression module prevents the system from switching back to the original primary link.

[0010] In some embodiments, the management system further includes a hardware resource dynamic scheduling module, which is connected to the dynamic scheduling unit and the airborne power management system. The hardware resource dynamic scheduling module is configured to monitor the load rate of the airborne processor and the power consumption status of each heterogeneous communication module. When the processor load exceeds the first preset threshold, the hardware resource dynamic scheduling module reduces the sampling frequency of the link state prediction unit and simplifies the scheduling algorithm complexity. When the remaining battery power is lower than the second preset threshold, the hardware resource dynamic scheduling module forcibly shuts down the low-priority communication link and switches to low-power narrowband communication mode.

[0011] In some embodiments, the physical layer parameters collected by the full-dimensional sensing module include signal-to-noise ratio, received signal strength indication, bit error rate, and channel occupancy rate; The transmission layer state parameters collected by the full-dimensional sensing module include round-trip time, jitter, packet loss rate, and number of retransmissions; The flight status data collected by the all-dimensional perception module includes the drone's latitude and longitude coordinates, flight altitude, velocity vector, and remaining battery power; Physical layer parameters, transport layer status parameters, and flight status data are transmitted to the dynamic scheduling unit in real time through a standardized interface.

[0012] In some embodiments, the multimode communication hardware layer includes a 5G / 4G cellular module, a low-orbit satellite communication module, a Wi-Fi 6 self-organizing network module, and a LoRa narrowband module; 5G / 4G cellular modules, low-orbit satellite communication modules, Wi-Fi 6 self-organizing network modules, and LoRa narrowband modules connect to the onboard computer via M.2 or USB interfaces and share the same GPS / BeiDou time source to achieve microsecond-level time synchronization. The dynamic scheduling unit sends link selection instructions to each module via sockets.

[0013] In some embodiments, the management system further includes a closed-loop feedback optimization module, which is connected to the ground station control center and the dynamic scheduling unit respectively. The closed-loop feedback optimization module is configured to statistically analyze the end-to-end effective throughput and command loss rate, generate feedback results, and feed the feedback results back to the link status prediction unit. The link state prediction unit uses the feedback results to adjust the weight parameters of the time series prediction model online, and the dynamic scheduling unit uses the feedback results to update the flow table scheduling strategy.

[0014] In some embodiments, the dynamic scheduling unit is deployed between the airborne edge computing unit and the ground station control center; The dynamic scheduling unit is responsible for the aggregation, encoding and distribution of uplink data, while the ground station control center is responsible for the fragmentation, scheduling and multicasting of downlink data, forming a bidirectional asymmetric link aggregation channel; The ground station control center maintains the global link topology map and issues auxiliary decision-making instructions for link switching to the dynamic scheduling unit through the telemetry link.

[0015] This application collects link parameters, flight status, and environmental information through a full-dimensional sensing module, infers future link quality using a time-series prediction model, and dynamically schedules control signaling and media streams accordingly, performing differentiated redundancy or fragmented encoding and mapping them to heterogeneous links for parallel transmission, while intelligently controlling link switching. This scheme achieves predictive management of communication links, effectively eliminating switching lag and ping-pong effects in highly dynamic flight, significantly improving the transmission reliability of critical commands, and maximizing aggregated bandwidth utilization while ensuring communication resilience. Attached Figure Description

[0016] Figure 1 This is a system architecture diagram of the UAV communication resilience management system that supports multi-link dynamic aggregation and intelligent switching proposed in this invention.

[0017] In the diagram: 1. Multimode communication hardware layer; 2. Full-dimensional sensing module; 3. Link status prediction unit; 4. Dynamic scheduling unit; 5. Service classification sub-module; 6. Transmission coding sub-module; 7. UAV flight control system. Detailed Implementation

[0018] The following description is intended to disclose the invention and enable those skilled in the art to implement it. The preferred embodiments described below are merely examples, and other obvious variations will occur to those skilled in the art.

[0019] Firstly, this application provides a UAV communication resilience management system that supports multi-link dynamic aggregation and intelligent switching, such as... Figure 1 As shown, it includes: Multimode communication hardware layer 1 includes at least two heterogeneous communication modules, which are used to establish communication links of different standards respectively; The all-dimensional perception module 2 is connected to the multi-mode communication hardware layer 1 and the UAV flight control system 7. The all-dimensional perception module 2 is configured to collect physical layer parameters, transmission layer status parameters, UAV flight status data and environmental geographic information of each communication link in real time. Link status prediction unit 3 is connected to the full-dimensional perception module 2. The link status prediction unit 3 is configured to infer the connectivity probability and available bandwidth of each communication link within a future preset time window based on UAV flight status data and environmental geographic information, and output the link quality score. The dynamic scheduling unit 4 is connected to the full-dimensional perception module 2, the link state prediction unit 3, and the multi-mode communication hardware layer 1, respectively. The dynamic scheduling unit 4 includes a service classification submodule 5 and a transmission coding submodule 6. The service classification submodule 5 is configured to identify the service characteristics of the uplink data stream generated by the airborne application layer and mark the priority. The transmission coding submodule 6 is configured to perform redundant coding or fragmentation coding on the uplink data stream according to the priority, and map the encoded data packets to different communication links for parallel transmission. The dynamic scheduling unit 4 is also configured to generate a scheduling strategy based on the link quality score, physical layer parameters and transport layer status parameters, and control the multi-mode communication hardware layer 1 to perform link activation, link hibernation or link switching operations on the communication link based on the scheduling strategy.

[0020] In practical implementation, the system's technology can be applied across various scenarios. For example, when performing inspection tasks in complex urban environments with numerous high-rise buildings, the multi-mode communication hardware layer 1 simultaneously utilizes both the 5G module and the low-orbit satellite module. The all-dimensional perception module 2 not only captures the signal-to-noise ratio, bit error rate, and round-trip latency of each link in real time, but also closely integrates the latitude and longitude coordinates, flight altitude, and attitude angular velocity transmitted back by the UAV flight control system 7, and overlays building vector data from the electronic map. The link state prediction unit 3 employs a fusion algorithm of Long Short-Term Memory (LSTM) network and Kalman filter to infer the link connectivity probability within the next 3 seconds based on the aforementioned multi-dimensional data. When the UAV is about to enter a signal obstruction area, causing a sharp drop in the 5G link quality score, the system will trigger the pre-connection handshake process of the satellite module in advance to avoid communication interruption. Meanwhile, the service classification submodule 5 in the dynamic scheduling unit 4 deeply analyzes the data packet header, identifies flight control commands as high-priority control streams, and real-time video streams as low-priority media streams. The transmission encoding submodule 6 then uses the system fountain code to fully copy the control stream, enabling it to be broadcast and transmitted on both 5G and satellite links. Even if one link is completely interrupted, the ground station can still restore the commands through the other link. For the video stream, a modified MPTCP protocol combined with SVC scalable encoding is used to map the base layer to the low-bandwidth satellite link to ensure continuity, and the enhancement layer to the high-bandwidth 5G link to improve image quality.

[0021] This deeply integrated mechanism of perception, prediction, and scheduling has yielded significant benefits. First, through precise prediction of future link states using a time-series prediction model, the system shifts from passive "link reconnection" to proactive "pre-connection," completely eliminating communication gaps caused by handover lag during highly dynamic flights and resolving the ping-pong effect problem of traditional threshold handover. Second, differentiated coding and mapping strategies based on service priorities enable critical control commands to have extremely high redundancy, ensuring reliable transmission even on severe links with packet loss rates as high as 30%. Furthermore, the fragmentation and aggregation of media streams maximizes the utilization of remaining bandwidth in heterogeneous links, avoiding overall stuttering caused by single-link congestion. Finally, the real-time fusion of flight status and environmental geographic information by the full-dimensional perception module 2 allows the scheduling strategy to dynamically adjust with changes in UAV attitude. For example, it automatically adjusts antenna beam pointing and link weights when the UAV climbs or turns, thereby significantly improving spectrum resource utilization and mission execution continuity while ensuring communication resilience.

[0022] In some embodiments, the link state prediction unit 3 employs a fusion filtering and neural network algorithm. The input data of the link state prediction unit 3 includes building vector data within a preset radius around the current position of the UAV, historical link quality sequences, and UAV attitude angular velocity. The output of the link state prediction unit 3 is the quality score of each link at a future time T. The dynamic scheduling unit 4 is configured to trigger the pre-connection handshake process of the backup link in the communication link when the quality score is lower than the preset security threshold, and to establish a backup data channel before the primary link in the communication link is interrupted.

[0023] In practical implementation, the effectiveness of this technical solution is mainly reflected in the deep collaborative mechanism between link state prediction and dynamic scheduling. For example, in a densely populated urban "canyon" scenario, the link state prediction unit 3 not only receives the historical signal-to-noise ratio sequence from the full-dimensional perception module 2, but also accesses the attitude angular velocity data from the UAV flight control system 7 and the vector data of buildings within 200 meters provided by the GIS system in real time. After cleaning the multi-source noise data using Kalman filtering, it is input into a Long Short-Term Memory (LSTM) network for training and inference, outputting a 5G link quality score for the next 3 seconds. When the score drops below the safety threshold due to the UAV's attitude adjustment and its impending entry into a signal obstruction area, the dynamic scheduling unit 4 does not wait for the link to actually break, but immediately triggers the pre-connection handshake process of the low-orbit satellite module. Within hundreds of milliseconds before the physical disconnection of the 5G primary link, it completes the IP allocation and port binding of the backup link in advance, establishing a redundant data channel. For example, in mountainous logistics and distribution scenarios, the system uses historical Doppler frequency shift sequences and real-time terrain occlusion models to predict the impending loss of Wi-Fi link. The scheduling unit then activates the 4G module in advance and warms up the protocol stack in the "gray area" where the Wi-Fi signal is attenuated but not interrupted, to ensure the continuity of data transmission.

[0024] The derivation process of this prediction-based pre-connection mechanism reveals significant benefits. First, by introducing building vector data and attitude angular velocity as inputs to the neural network, the system no longer relies solely on instantaneous signal strength indicators but gains the ability to "see through" environmental occlusion and motion trends. This results in a high-confidence score for the link quality at time T, resolving the lag problem of traditional threshold-based switching. Second, the dynamic scheduling unit 4 uses this score to trigger a pre-connection handshake in advance, transforming the passive reconnection that occurs after a link interruption into an active backup before the interruption. Since the pre-connection process is executed in parallel in the background while the main link is still transmitting data, the backup channel can immediately take over the data stream the moment the main link is truly interrupted, completely eliminating the communication gap and effectively preventing the loss of control commands due to switching delays. Finally, this "soft switching" mode avoids frequent oscillations in link quality at threshold edges, i.e., the ping-pong effect. This not only protects the power consumption and lifespan of the airborne module but also ensures zero-packet-loss transmission of high-priority services (such as flight control commands) through seamless link aggregation, thereby fundamentally improving the resilience and safety of UAVs operating in complex electromagnetic environments.

[0025] In some embodiments, the service classification submodule 5 is configured to perform deep parsing of the protocol field in the header of the uplink data stream, and divide the uplink data stream into a high-priority control stream and a low-priority media stream. Transmission coding submodule 6 adopts a full copy strategy for high-priority control flows, so that the same data packet is broadcast and transmitted simultaneously on all active links in all communication links. The transmission coding submodule 6 employs a scalable coding strategy for low-priority media streams to generate base layer data packets and enhancement layer data packets.

[0026] In practical implementation, this technical solution achieves efficient data distribution through deep packet inspection and differentiated coding strategies. For example, in a densely packed urban "canyon" scenario, the service classification submodule 5 intercepts uplink data packets in real time. By parsing the TOS field and transport layer port number in the IP header, it accurately identifies small UDP packets originating from the flight control system as high-priority control streams, while large RTP packets destined for video streaming are low-priority media streams. The transmission coding submodule 6 then performs a full copy strategy on the control stream, generating multiple identical data packets, which are then injected into all currently active links such as 5G, Wi-Fi 6, and satellite communication for broadcast transmission. This ensures that even if a link is momentarily interrupted due to building obstruction, the receiving end can still receive complete instructions from other links. Simultaneously, for the media stream, the transmission coding submodule 6 uses SVC scalable video coding technology to split the high-definition video stream into basic layer data packets that ensure basic smoothness and enhancement layer data packets that improve image quality. In another mountainous logistics delivery scenario, the system dynamically maps layered data based on link bandwidth, fixing the basic layer to a narrow but stable LoRa link to ensure a minimum, and mapping the enhancement layer to a 5G link with ample bandwidth to pursue high image quality.

[0027] The derivation process of this business-aware differentiated scheduling mechanism reveals significant benefits. First, by employing a full-copy broadcast strategy for the control flow, the system constructs spatial diversity gain, making critical control commands no longer dependent on the stability of a single link. Even in extreme cases where high-dynamic maneuvers cause a sudden drop in signal strength on a particular link, the receiver only needs to capture data packets from any one link to reconstruct the command. This completely eliminates the risk of control command loss due to switching delays in traditional single-link transmission, greatly improving flight safety. Second, the scalable coding strategy adopted for the media stream enables refined utilization of bandwidth resources. The base layer data packets are small in size and have strong anti-packet loss capabilities, maintaining the continuity of video surveillance in weak network environments. The enhancement layer utilizes remaining bandwidth to improve image quality. This layered mechanism avoids the precipitous drop in user experience caused by a "one-size-fits-all" approach to quality degradation, while also preventing high-volume media streams from crowding out control signaling bandwidth during congestion. Finally, the deep parsing of protocol fields in the business classification submodule 5 enables precise traffic isolation, ensuring that high-priority services always have exclusive transmission channels and will not experience queuing delays due to sudden traffic from low-priority services. This maximizes the effective throughput after heterogeneous link aggregation while ensuring communication resilience.

[0028] In some embodiments, the transmission encoding submodule 6 uses system fountain codes to encode the high-priority control flow, so that the receiving end can restore the original control command when it receives any number of encoded packets; The transmission coding submodule 6 uses a modified version of the multipath transmission control protocol MPTCP to process low-priority media streams. The transmission coding submodule 6 establishes multiple substreams corresponding to different communication links and performs out-of-order reordering of data packets through a common sequence number space. Among them, basic layer data packets are mapped to low-bandwidth links in the communication link for transmission, while enhancement layer data packets are mapped to high-bandwidth links in the communication link for transmission.

[0029] In practical implementation, this technical solution achieves high reliability and efficient data transmission by combining system fountain codes with multipath transmission protocols. For example, in urban emergency inspection scenarios, the transmission coding submodule 6, for high-priority flight control command streams, abandons the traditional fixed code rate encoding method and instead uses system fountain codes for processing, generating a large number of encoded packets and broadcasting them simultaneously through multiple links such as 5G, satellite, and ad hoc networks. The ground station receiver does not need to wait for data packets with specific numbers; it only needs to capture any number of encoded packets to restore the original commands through decoding algorithms. This ensures that even during the handover gap when the primary link is momentarily interrupted due to tall buildings blocking the signal, any encoded packet on the backup link can maintain the continuity of the control link. Meanwhile, for low-priority high-definition video streams, the transmission encoding submodule 6 runs a modified version of the Multipath Transmission Control Protocol (MPTCP). At the transport layer, multiple logical substreams are established, each bound to a different physical link. All substreams share a common sequence number space, which the receiver uses to precisely reassemble data packets from different paths that may arrive out of order. During this process, submodule 6 also performs intelligent mapping based on the real-time bandwidth characteristics of each link. It forcibly schedules the base layer data packets generated by SVC encoding to narrower but more stable LoRa or narrowband satellite links to ensure minimum image quality continuity, while scheduling enhancement layer data packets to high-bandwidth 5G or Wi-Fi 6 links to improve clarity. In another example of mountainous logistics delivery, when a drone traverses a canyon with complex signal coverage, the MPTCP protocol can dynamically adjust the congestion windows of each substream, transferring data from congested links to idle links. The redundancy of fountain codes ensures that critical obstacle avoidance commands are not lost due to sudden interference from a single link.

[0030] This deeply integrated coding and transmission mechanism brings significant benefits, as explained in the derivation process. First, the application of fountain codes in the system frees the transmission of high-priority control streams from being constrained by the instantaneous state of the link. Traditional retransmission mechanisms often fail due to delays in highly dynamic environments, while the "no fixed code rate" characteristic of fountain codes allows the receiver to recover instructions at any packet loss rate, fundamentally eliminating control blind spots caused by link switching or instantaneous interference, and greatly improving flight safety. Second, the modified MPTCP solves the problem of out-of-order delivery and reassembly between heterogeneous links through a common sequence number space, enabling multiple links with huge differences in bandwidth and latency to work collaboratively like a single logical pipeline. This not only avoids head-of-line congestion but also significantly improves the effective throughput of the media stream by aggregating bandwidth. Finally, the precise mapping strategy between data packets and link bandwidth in the base layer and enhancement layer achieves optimal resource allocation. The base layer's guaranteed transmission in weak network environments prevents video black screens, while the enhancement layer's best-effort transmission in strong network environments ensures image quality. This layered guarantee mechanism maximizes the user's service experience quality without increasing additional bandwidth overhead, while also avoiding the crowding out of critical control signaling by low-priority, high-volume data, thus ensuring the overall communication resilience of the system.

[0031] In some embodiments, the dynamic scheduling unit 4 further includes a switching decision hysteresis suppression module, which is configured to set a hysteresis threshold and a penalty factor. The dynamic scheduling unit 4 only triggers a link switching operation when the quality score of the primary link is lower than the difference between the quality score of the backup link and the lag threshold. Within a preset cooling-off period after the link switching operation is completed, the switching decision hysteresis suppression module prevents the system from switching back to the original primary link.

[0032] In practical implementation, this handover decision hysteresis suppression module effectively solves the problem of frequent handovers caused by link signal jitter in highly dynamic environments by introducing a hysteresis threshold and a penalty factor. For example, in the complex electromagnetic environment of a city with many high-rise buildings, when a drone is rapidly moving between two buildings, the signal strength of the primary 5G link may drop momentarily due to multipath effects and then recover quickly. Judging by traditional instantaneous values ​​can easily lead to false handovers. At this time, the dynamic scheduling unit 4 calculates the quality score difference between the primary link and the backup satellite link in real time. Only when the primary link score is continuously lower than the backup link score minus the hysteresis threshold (e.g., set to 3 points) is it determined to be a real link degradation and a handover operation is performed, avoiding invalid oscillations caused by signal spikes. In another mountain patrol scenario, when the drone flies over a ridge and obstructs the line of sight, after the system completes the switch from Wi-Fi to 4G, the switch decision lag suppression module will immediately start the preset cooling time (e.g., 5 seconds). During this period, even if the Wi-Fi signal briefly rebounds due to the drone's attitude adjustment, the system will forcibly prohibit the switchback operation to ensure that the protocol stack of the new link completes a stable handshake and routing table convergence.

[0033] This strategy, based on hysteresis suppression and cooling mechanisms, yields significant beneficial effects. First, by setting hysteresis thresholds and penalty factors, the system upgrades switching decisions from instantaneous point-to-point comparisons to trend judgments over a time period. Switching only occurs when the quality advantage of the backup link is established and the degradation of the primary link is irreversible. This directly eliminates the "ping-pong effect" caused by link quality fluctuations at threshold edges, significantly reducing signaling overhead and module power consumption due to repeated switching. Second, the introduction of a preset cooling time eliminates the possibility of backswitching in the time dimension, preventing the system from repeatedly hopping between old and new links, ensuring the stability of the data transmission path, and enabling transport layer protocols such as TCP to smoothly establish connections on new links, preventing out-of-order packet retransmissions caused by frequent routing changes. Finally, this mechanism significantly improves the transmission continuity of high-priority services, especially in video streaming, effectively avoiding screen stuttering or glitches caused by frequent link oscillations. Thus, while ensuring communication resilience, it maximizes the overall performance and reliability of the heterogeneous link aggregation system.

[0034] In some embodiments, the management system further includes a hardware resource dynamic scheduling module, which is connected to the dynamic scheduling unit 4 and the airborne power management system. The hardware resource dynamic scheduling module is configured to monitor the load rate of the airborne processor and the power consumption status of each heterogeneous communication module. When the processor load rate exceeds the first preset threshold, the hardware resource dynamic scheduling module reduces the sampling frequency of the link state prediction unit 3 and simplifies the scheduling algorithm complexity. When the remaining battery power is lower than the second preset threshold, the hardware resource dynamic scheduling module forcibly shuts down the low-priority communication link and switches to low-power narrowband communication mode.

[0035] In practical implementation, the hardware resource dynamic scheduling module achieves refined management of computing power and energy consumption through deep coupling with the airborne power management system and dynamic scheduling unit 4. For example, when performing high-precision mapping tasks, the airborne edge computing unit experiences a processor load rate spike to over 90% due to the simultaneous operation of the visual SLAM algorithm and communication protocol stack. At this time, the hardware resource dynamic scheduling module immediately intervenes, reducing the sampling frequency of the link state prediction unit 3 from once every 10 milliseconds to once every 100 milliseconds, and temporarily shutting down the nonlinear optimization module in the prediction model, retaining only the linear extrapolation algorithm to maintain basic link quality assessment functions, thereby freeing up CPU computing power to ensure the real-time performance of core flight control and data encoding. Similarly, during the final stages of long-distance inspections, when the remaining battery power drops to the second preset threshold of 15%, the module forcibly cuts off the power supply to the high-power 5G communication module, retaining only the low-power narrowband communication link to transmit critical telemetry data. Simultaneously, it instructs the dynamic scheduling unit 4 to stop all redundant encoding operations of media streams, maintaining only the most basic control command pass-through. In another overheat protection scenario, if the temperature of the communication module is detected to be too high, the module will dynamically reduce the radio frequency transmission power and adjust the modulation method, sacrificing some bandwidth to ensure that the connection is not interrupted in order to improve hardware stability.

[0036] The derivation process of this hardware-software collaborative resource scheduling mechanism reveals significant benefits. First, by monitoring processor load in real time and dynamically reducing algorithm complexity, the system avoids crashes or scheduling stagnation caused by computing power exhaustion, ensuring the "survivability" of the communication management system under high-load tasks. This prevents the drone from losing connection due to competition for computing resources when performing complex tasks. Second, the link forced switching strategy based on power thresholds prioritizes flight safety and control links with limited power resources. By sacrificing unnecessary high-bandwidth services, it significantly extends the drone's loiter time and safe return probability under low power conditions, solving the risk of "sudden crashes" caused by the lax power management of traditional systems. Finally, closed-loop control of communication module power consumption and temperature not only reduces the probability of hardware overheating and damage but also reduces unnecessary energy waste through intelligent frequency reduction and sleep strategies. This allows the heterogeneous communication hardware layer to operate with a better energy efficiency ratio, thereby maximizing the overall operating radius and mission continuity of the drone while ensuring communication resilience.

[0037] In some embodiments, the physical layer parameters collected by the full-dimensional sensing module 2 include signal-to-noise ratio, received signal strength indication, bit error rate, and channel occupancy rate; The transmission layer status parameters collected by the full-dimensional sensing module 2 include round-trip time, jitter, packet loss rate, and number of retransmissions; The flight status data collected by the all-dimensional perception module 2 includes the UAV's latitude and longitude coordinates, flight altitude, velocity vector, and remaining battery power; Physical layer parameters, transport layer status parameters, and flight status data are transmitted to the dynamic scheduling unit 4 in real time through a standardized interface.

[0038] In practical implementation, the all-dimensional perception module 2 achieves deep data fusion of heterogeneous communication modules and flight control systems through underlying driver and middleware technologies. For example, in inspection operations in complex urban electromagnetic environments, the all-dimensional perception module 2 captures the physical layer signal-to-noise ratio, received signal strength indication (RSSI), hard demodulation bit error rate, and channel occupancy rate of the 5G module in real time at millisecond intervals. Simultaneously, it monitors the round-trip time (RTT), packet jitter, packet loss rate, and TCP retransmission count of the transport layer through the Socket interface, and synchronously obtains the UAV's WGS84 latitude and longitude coordinates, barometer-corrected flight altitude, three-dimensional velocity vector, and remaining battery power from the flight control system through the MAVLink protocol. After being timestamped and aligned within the module, this heterogeneous data is pushed at high speed to the dynamic scheduling unit 4 in a structured message format through a standardized middleware interface based on DDS (Data Distribution Service). In another mountainous logistics delivery scenario, when a drone traversing a canyon causes severe fluctuations in the satellite link signal, the all-dimensional perception module 2 not only collects the bit error rate and channel occupancy rate of the satellite link, but also closely combines this with the drone's vertical velocity vector and flight altitude data to determine whether the signal attenuation is due to obstruction or increased distance, and transmits these raw parameters to the scheduling unit without loss. Similarly, in high-bandwidth video backhaul tasks, the module focuses on monitoring the channel occupancy rate and transmission layer jitter of the Wi-Fi link. Once channel congestion is detected, it immediately packages and sends the current latitude and longitude coordinates and battery level to assist the scheduling unit in deciding whether to switch to the 5G link.

[0039] This comprehensive and standardized data acquisition and transmission mechanism brings significant benefits. First, by simultaneously incorporating the physical layer's signal-to-noise ratio (SNR) and bit error rate (BER) with the transmission layer's latency and packet loss rate, the system constructs a more accurate link quality profile than a single RSSI. This allows it to distinguish between packet loss caused by weak signals and retransmissions caused by network congestion, enabling the dynamic scheduling unit 4 to selectively choose coding strategies or switch links, avoiding ineffective switching due to misjudgments based on a single indicator. Second, the real-time fusion of flight status data (latitude, longitude, velocity vector, altitude) and communication parameters endows the system with "geographical and motion perception" capabilities. For example, when it senses that a drone is flying at high speed towards an obstructed area and its battery is low, the scheduling unit can predict the risk of link interruption in advance and execute a soft switch, rather than passively waiting for the connection to be lost. This fundamentally improves the continuity and resilience of communication. Finally, the application of standardized interfaces breaks the coupling between the underlying hardware and the upper-level scheduling algorithm, enabling communication modules and flight control data from different manufacturers to seamlessly access the scheduling logic. This not only reduces the complexity of system integration but also ensures low latency and high fidelity of multi-dimensional parameters during transmission, ensuring that the strategy generated by the dynamic scheduling unit 4 is always based on the freshest global state, thereby achieving millisecond-level link adaptive adjustment in highly dynamic flight.

[0040] In some embodiments, the multimode communication hardware layer 1 includes a 5G / 4G cellular module, a low-orbit satellite communication module, a Wi-Fi 6 self-organizing network module, and a LoRa narrowband module; 5G / 4G cellular modules, low-orbit satellite communication modules, Wi-Fi 6 self-organizing network modules, and LoRa narrowband modules connect to the onboard computer via M.2 or USB interfaces and share the same GPS / BeiDou time source to achieve microsecond-level time synchronization. The dynamic scheduling unit 4 sends link selection instructions to each module through the socket.

[0041] In practical implementation, this multi-mode communication hardware layer 1 constructs a robust physical transmission foundation through heterogeneous interfaces and a high-precision timing mechanism. For example, in urban counter-terrorism and emergency response scenarios, the airborne computer connects to a 5G cellular module via an M.2 high-speed interface to carry 4K high-definition video streams. Simultaneously, it connects to an external Wi-Fi 6 self-organizing network module via a USB interface to build a temporary local area network with ground-based individual soldier equipment. In addition, it also has an onboard LoRa narrowband module for low-speed telemetry in extreme situations. All modules are connected to the same GPS / BeiDou dual-mode timing module via hardware cables, ensuring that the timestamp error of data packets on each link is controlled within the microsecond level, providing a benchmark for subsequent out-of-order reordering. In remote power line inspection scenarios, low-Earth orbit (LEO) satellite communication modules serve as the primary link for wide-area coverage, establishing long-term connections with ground stations, while LoRa modules act as the last-line backup link. When a drone enters a valley blind spot, the dynamic scheduling unit 4 sends a "sleep" command to the 5G module via a socket, while simultaneously sending a "wake up and increase transmission power" command to the satellite module. The entire switching process utilizes microsecond-level time synchronization to achieve seamless connection. Similarly, in formation flight tests, multiple drones share location information through a Wi-Fi 6 ad hoc network module. The scheduling unit uses a unified timing source to calculate the relative latency of each link and dynamically adjusts the data packet transmission sequence to avoid airborne collisions.

[0042] The derivation process of the significant beneficial effects brought about by this hardware architecture and synchronization mechanism is as follows. First, by integrating four heterogeneous modules—5G / 4G, satellite, Wi-Fi 6, and LoRa—the system constructs a communication capability covering the entire spectrum of "high bandwidth, wide area, local area, and low power consumption." The failure of a single link (such as 5G base station disconnection or satellite signal obstruction) no longer leads to overall communication paralysis, but can instantly switch to other available links, greatly improving the system's resilience and survivability. Second, the microsecond-level time synchronization achieved by sharing GPS / BeiDou time sources solves the problem of inconsistent time bases between heterogeneous networks. This allows the dynamic scheduling unit 4 to accurately align the timestamps of data packets from different links when performing multi-link aggregation or switching, avoiding out-of-order packet loss or scheduling logic confusion caused by clock drift, and ensuring the efficient operation of protocols such as MPTCP. Finally, the Socket-based instruction interaction method has extremely low system overhead and cross-platform compatibility, enabling scheduling strategies to be delivered to the underlying drivers of each module with millisecond-level latency. Combined with high-precision time synchronization, it achieves the minimization of closed-loop latency from link awareness to execution switching, ensuring that communication link switching is as smooth as "shifting gears" in high-speed flight or complex electromagnetic interference environments, thereby fundamentally guaranteeing the continuity and real-time performance of UAV operations.

[0043] In some embodiments, the management system further includes a closed-loop feedback optimization module, which is connected to the ground station control center and the dynamic scheduling unit 4 respectively. The closed-loop feedback optimization module is configured to statistically analyze the end-to-end effective throughput and instruction loss rate, generate feedback results, and feed the feedback results back to the link state prediction unit 3; Link state prediction unit 3 uses feedback results to adjust the weight parameters of the time series prediction model online, and dynamic scheduling unit 4 uses feedback results to update the flow table scheduling strategy.

[0044] In practical implementation, the closed-loop feedback optimization module achieves continuous self-evolution of communication strategies by constructing a complete "perception-decision-execution-evaluation" link. For example, when performing long-term inspection tasks in the complex electromagnetic environment of densely populated urban buildings, the closed-loop feedback optimization module continuously monitors the effective throughput of the video stream received by the ground station and the loss rate of flight control commands. When it finds that the actual throughput is consistently lower than the theoretical estimate of the link state prediction unit 3, or when the command loss rate shows a slight increase, the module quantifies this "prediction deviation" to generate feedback results and sends them back to the link state prediction unit 3. After receiving this feedback, the link state prediction unit 3 uses an online learning algorithm to fine-tune the weight parameters in the time-series prediction model, such as reducing the dependence on historical signal strength and increasing the sensitivity to building occlusion features, so that the subsequent link quality score is more in line with the current actual environment. At the same time, the dynamic scheduling unit 4 dynamically updates the flow table scheduling strategy based on the command loss rate index in the feedback results. For example, when it detects that a link has a strong signal but a high packet loss rate, it automatically lowers the weight of that link in load balancing and increases the redundancy of redundant coding. In another long-distance communication scenario across the sea, as the drone gradually moves away from the coast, the closed-loop feedback optimization module detects an increase in the actual time delay jitter of the satellite link and then feeds it back to the prediction unit to adjust the coefficients of the Doppler frequency shift compensation algorithm. At the same time, the command scheduling unit downgrades the high-order modulation scheme originally used for the satellite link to a more robust low-order modulation and increases the length of the forward error correction code in the flow table.

[0045] The derivation process of this closed-loop feedback mechanism reveals significant beneficial effects. First, by introducing end-to-end effective throughput and command loss rate as "ground truth," the system breaks the deadlock of model prediction being disconnected from the actual environment in open-loop control. This allows the link state prediction unit 3 to correct driving predictions based on actual road conditions, much like a human driver. As flight time increases, the model's adaptability to the electromagnetic environment of a specific area becomes stronger, and the prediction accuracy increases spirally, fundamentally solving the problem that static models cannot cope with dynamically changing channels. Second, the dynamic scheduling unit 4 uses feedback results to update the flow table strategy in real time, realizing a shift from "passive response" to "active optimization." For example, when a hidden packet loss is detected on a link, the mapping relationship of the data flow can be adjusted in advance without waiting for the link to be completely interrupted. This fine-tuning mechanism avoids drastic link switching and ensures the smoothness of the service flow. Finally, the closed-loop feedback optimization module incorporates the ground station control center into the loop, making human-machine collaboration possible. The ground station can macroscopically intervene in the feedback weights from a global task perspective, ensuring the autonomy of the algorithm while retaining the interface for manual control. Ultimately, without increasing hardware costs, the overall communication resilience and resource utilization of the heterogeneous link aggregation system are maximized through continuous software iteration.

[0046] In some embodiments, the dynamic scheduling unit 4 is deployed between the airborne edge computing unit and the ground station control center; Dynamic scheduling unit 4 is responsible for the aggregation, encoding and distribution of uplink data, while the ground station control center is responsible for the fragmentation, scheduling and multicasting of downlink data, forming a bidirectional asymmetric link aggregation channel; The ground station control center maintains the global link topology map and sends auxiliary decision-making instructions for link switching to the dynamic scheduling unit 4 through the telemetry link.

[0047] In practical implementation, this architecture constructs an efficient link aggregation channel through the division of computing power between airborne and ground-based terminals and the design of asymmetric data flow. For example, in wide-area mapping scenarios, the dynamic scheduling unit 4 is deployed inside the airborne edge computing unit. In the uplink direction, it aggregates data packets from 5G, satellite, and ad hoc network modules, and performs fountain coding or SVC hierarchical coding according to service priority. Then, it distributes the packets to each radio frequency module for concurrent transmission via a high-speed bus. In the downlink direction, the ground station control center acts as the "brain," maintaining a global link topology map that includes all available link bandwidth, latency, and coverage. When it is necessary to issue high-definition maps or control commands, the ground station splits large files into multiple sub-streams and pushes them to the UAV simultaneously via multicast based on the optimal path calculated from the global topology. The dynamic scheduling unit 4 on the UAV side is only responsible for receiving and reassembling the data, without needing to perform complex routing calculations. In another emergency communication vehicle collaboration scenario, the ground station detected a surge in the probability of satellite signal obstruction in a certain area through telemetry links. It then proactively issued an auxiliary decision-making instruction to the airborne dynamic scheduling unit 4 to "pre-switch to relay UAV link". After receiving this instruction with a global perspective, the airborne unit combined local perception data to perform a rapid switch, avoiding misjudgments caused by relying solely on local airborne perception.

[0048] The derivation process of this bidirectional asymmetric link aggregation mechanism reveals significant benefits. First, by moving downlink data fragmentation, scheduling, and multicast functions to the ground station control center, the near-unlimited computing and storage resources on the ground are fully utilized. This allows the airborne dynamic scheduling unit 4 to focus on uplink data processing, which has extremely high real-time requirements. This offloading of computing power not only reduces the load rate of the airborne computer but also avoids scheduling delays caused by airborne computing bottlenecks. Second, the global link topology map maintained by the ground station breaks through the limitations of the UAV's single perspective. It can predict link congestion and weather changes over a wide area. When it issues auxiliary decision-making instructions to the airborne unit, it actually provides a high-confidence "prior knowledge" for local decision-making, thereby significantly reducing the trial time for link switching and making switching decisions more accurate and forward-looking. Finally, the asymmetric channel design perfectly matches the business characteristics of "high uplink video backhaul and low downlink control commands" in UAV operations. Uplink aggregation coding improves spectrum efficiency, while downlink multicast fragmentation utilizes the advantages of broadcast channels. The combination of the two significantly improves the effective end-to-end throughput without increasing additional spectrum resources. Furthermore, the "ping-pong effect" is effectively suppressed through global intervention from the ground station, ensuring the ultimate stability and reliability of the communication link in a highly dynamic environment.

[0049] In practical implementation, the technical implementation of this UAV communication resilience management system can be achieved through deep hardware and software integration of the airborne edge computing unit and the multi-mode communication module. For example, the multi-mode communication hardware layer 1 can use a 5G industrial module based on the M.2 interface, a low-orbit satellite communication terminal based on the USB interface, and a Wi-Fi 6 self-organizing network card based on the PCIe interface as heterogeneous communication modules. These three types of modules correspond to three completely different communication standards: wide-area cellular, long-range satellite, and local self-organizing network. They achieve microsecond-level time synchronization by sharing an airborne GPS / BeiDou dual-mode timing source, ensuring that the timestamps of data packets are aligned during multi-link aggregation. The full-dimensional perception module 2 can be deployed in the underlying driver of the airborne computer to capture the physical layer parameters of each module in real time, such as signal-to-noise ratio (SNR), received signal strength indication (RSSI), and channel occupancy rate. At the same time, it obtains the UAV's latitude and longitude, velocity vector, and attitude angular velocity from the flight control system through the MAVLink protocol, and combines them with the building vector layer in the pre-loaded geographic information system (GIS) data to construct a four-dimensional state space that includes the electromagnetic environment and the physical environment. Link state prediction unit 3 can use a hybrid algorithm that combines extended Kalman filter (EKF) and long short-term memory network (LSTM). It takes historical link quality sequence, current attitude angular velocity and surrounding building occlusion as input, and predicts the link connectivity probability within the next 5 seconds through time series extrapolation. For example, when a drone is about to fly into the leeward side of a tall building, this unit can predict the attenuation trend of the 5G link in advance and output a low quality score. The dynamic scheduling unit 4, acting as the core decision-making brain, has a business classification submodule 5 that can deeply analyze the DSCP field and port number in the IP packet header, marking key data such as flight control commands and attitude feedback as high-priority control flows, while marking video downloads and telemetry logs as low-priority media flows. The transmission encoding submodule 6 uses the system fountain code (RaptorQ) for full-scale replication encoding of high-priority flows, so that the ground station only needs to receive any set of encoded packets to restore the original commands. For low-priority flows, it uses Scalable Video Coding (SVC), splitting the base layer and enhancement layer, and combining it with a modified version of the Multipath Transmission Control Protocol (MPTCP) to establish sub-streams corresponding to different physical links, and using the common sequence number space to perform out-of-order reordering at the receiving end. In addition, the system can also deploy a hardware resource dynamic scheduling module, which is connected to the airborne power management system (PMU) via the I2C bus. When the processor load rate is detected to exceed 80%, the sampling frequency of the link prediction model is automatically reduced to 1Hz and switched to a lightweight scheduling algorithm. When the battery power is below 15%, the high-power 5G module is forcibly shut down, and only the LoRa narrowband link is retained to maintain a minimum telemetry transmission.

[0050] This multi-level, multi-dimensional technical architecture brings significant benefits, and its derivation process closely follows the technical features in the claims. First, through physical layer isolation and time synchronization of heterogeneous communication modules, the system constructs a highly redundant communication base, so that single-system network failures (such as base station coverage blind spots or satellite signal blockage) no longer cause communication interruptions, but instead become a schedulable resource pool. Second, the link state prediction unit 3 introduces geographical environment and flight attitude data, solving the problem of lag in traditional communication systems that rely solely on instantaneous signal strength for decision-making. For example, before the drone crosses an urban canyon, the system has already predicted the link quality degradation through a predictive model, thereby triggering the pre-connection handshake of the backup link in advance, achieving "seamless switching" rather than "disconnection and reconnection," greatly improving flight safety. Furthermore, the service classification and differentiated coding strategies in the dynamic scheduling unit 4 enable refined configuration of communication resources: full replication and broadcasting of high-priority control flows ensures extremely high reliability of flight control commands, reaching their destination even under complex electromagnetic interference through multi-link redundancy; while scalable coding and multi-path mapping of media streams guarantee basic image quality transmission on bandwidth-constrained links, automatically overlaying enhancement layers when bandwidth is ample, achieving a dynamic balance between image quality and bandwidth. In addition, the handover decision hysteresis suppression module effectively avoids the "ping-pong handover" phenomenon caused by signal jitter by setting hysteresis thresholds and cooling times, reducing ineffective consumption of system computing power and link oscillations. The introduction of the hardware resource dynamic scheduling module extends the system's resilience boundary from an energy perspective, ensuring that the system can maintain core communication functions in degraded mode rather than completely fail under extreme conditions of low power or high computing load. Finally, the closed-loop feedback optimization module, from the global perspective of the ground station control center, statistically analyzes the end-to-end effective throughput and command loss rate, and reversely corrects the weight parameters of the airborne prediction model. This enables the system to possess self-learning and self-evolution capabilities, allowing it to adapt to electromagnetic environment changes in different regions and seasons. In summary, through a fully closed-loop design of perception, prediction, scheduling, and feedback, this system transforms passive link connections into proactive resource management, maximizing communication bandwidth utilization and system survivability while ensuring the absolute reliability of UAV control commands.

[0051] It should be noted that the order of the above embodiments of the present invention is merely for descriptive purposes and does not represent the superiority or inferiority of the embodiments. Furthermore, the above description focuses on specific embodiments of this specification. Additionally, the processes depicted in the accompanying drawings do not necessarily require a specific or sequential order to achieve the desired results. In some embodiments, multitasking and parallel processing are possible or may be advantageous.

[0052] The various embodiments in this specification are described in a progressive manner. The same or similar parts between the various embodiments can be referred to each other. Each embodiment focuses on describing the differences from other embodiments.

[0053] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A UAV communication resilience management system supporting multi-link dynamic aggregation and intelligent switching, characterized in that, include: The multimode communication hardware layer (1) includes at least two heterogeneous communication modules, which are used to establish communication links of different standards respectively. The full-dimensional perception module (2) is connected to the multi-mode communication hardware layer (1) and the UAV flight control system (7). The full-dimensional perception module (2) is configured to collect physical layer parameters, transmission layer status parameters, UAV flight status data and environmental geographic information of each communication link in real time. The link status prediction unit (3) is connected to the full-dimensional perception module (2). The link status prediction unit (3) is configured to infer the connectivity probability and available bandwidth of each communication link within a future preset time window based on the UAV flight status data and the environmental geographic information, and output the link quality score. The dynamic scheduling unit (4) is connected to the full-dimensional perception module (2), the link state prediction unit (3), and the multi-mode communication hardware layer (1), respectively. The dynamic scheduling unit (4) includes a service classification submodule (5) and a transmission coding submodule (6). The service classification submodule (5) is configured to identify the service characteristics of the uplink data stream generated by the airborne application layer and mark the priority. The transmission coding submodule (6) is configured to perform redundant coding or fragmentation coding on the uplink data stream according to the priority, and map the encoded data packets to different communication links for parallel transmission. The dynamic scheduling unit (4) is also configured to generate a scheduling strategy based on the link quality score, the physical layer parameters and the transport layer state parameters, and control the multi-mode communication hardware layer (1) to perform link activation, link hibernation or link switching operations on the communication link based on the scheduling strategy.

2. The management system according to claim 1, characterized in that, The link state prediction unit (3) adopts a fusion filtering and neural network algorithm. The input data of the link state prediction unit (3) includes building vector data within a preset radius around the current position of the UAV, historical link quality sequence and UAV attitude angular velocity. The output of the link state prediction unit (3) is the quality score of each link at a future time T. The dynamic scheduling unit (4) is configured to trigger the pre-connection handshake process of the backup link in the communication link when the quality score is lower than the preset security threshold, and to establish a backup data channel before the primary link in the communication link is interrupted.

3. The management system according to claim 1, characterized in that, The business classification submodule (5) is configured to perform deep parsing of the protocol field in the header of the uplink data stream and divide the uplink data stream into a high-priority control stream and a low-priority media stream. The transmission coding submodule (6) adopts a full copy strategy for the high-priority control flow, so that the same data packet is broadcast and transmitted simultaneously on all active links in the communication links. The transmission coding submodule (6) uses a scalable coding strategy for the low-priority media stream to generate base layer data packets and enhancement layer data packets.

4. The management system according to claim 3, characterized in that, The transmission coding submodule (6) uses the system fountain code to encode the high-priority control flow, so that the receiving end can restore the original control command when it receives any number of encoded packets; The transmission coding submodule (6) uses a modified version of the multipath transmission control protocol MPTCP to process the low-priority media stream. The transmission coding submodule (6) establishes multiple substreams corresponding to different communication links and performs out-of-order reassembly of data packets through a common sequence number space. The base layer data packets are mapped to low-bandwidth links in the communication link for transmission, and the enhancement layer data packets are mapped to high-bandwidth links in the communication link for transmission.

5. The management system according to claim 4, characterized in that, The dynamic scheduling unit (4) further includes a switching decision lag suppression module, which is configured to set a lag threshold and a penalty factor. The dynamic scheduling unit (4) triggers the link switching operation only when the quality score of the primary link is lower than the difference between the quality score of the backup link and the hysteresis threshold. During the preset cooling time after the link switching operation is completed, the switching decision hysteresis suppression module prohibits the system from switching back to the original primary link.

6. The management system according to claim 5, characterized in that, The management system also includes a hardware resource dynamic scheduling module, which is connected to the dynamic scheduling unit (4) and the airborne power management system. The hardware resource dynamic scheduling module is configured to monitor the load rate of the airborne processor and the power consumption status of each of the heterogeneous communication modules. When the processor load rate exceeds the first preset threshold, the hardware resource dynamic scheduling module reduces the sampling frequency of the link state prediction unit (3) and simplifies the scheduling algorithm complexity. When the remaining battery power is lower than the second preset threshold, the hardware resource dynamic scheduling module forcibly shuts down the low-priority communication link and switches to low-power narrowband communication mode.

7. The management system according to claim 6, characterized in that, The physical layer parameters collected by the full-dimensional sensing module (2) include signal-to-noise ratio, received signal strength indication, bit error rate, and channel occupancy rate; The transmission layer state parameters collected by the full-dimensional sensing module (2) include round-trip time, jitter, packet loss rate and number of retransmissions; The flight status data collected by the full-dimensional perception module (2) includes the latitude and longitude coordinates, flight altitude, velocity vector and remaining battery power of the UAV; The physical layer parameters, the transport layer state parameters, and the flight state data are transmitted to the dynamic scheduling unit (4) in real time through a standardized interface.

8. The management system according to claim 7, characterized in that, The multi-mode communication hardware layer (1) includes a 5G / 4G cellular module, a low-orbit satellite communication module, a Wi-Fi 6 self-organizing network module, and a LoRa narrowband module; The 5G / 4G cellular module, the low-orbit satellite communication module, the Wi-Fi 6 self-organizing network module, and the LoRa narrowband module are connected to the airborne computer via an M.2 interface or a USB interface and share the same GPS / BeiDou time source to achieve microsecond-level time synchronization. The dynamic scheduling unit (4) sends link selection instructions to each module through the socket.

9. The management system according to claim 8, characterized in that, The management system also includes a closed-loop feedback optimization module, which is connected to the ground station control center and the dynamic scheduling unit (4) respectively. The closed-loop feedback optimization module is configured to statistically analyze the end-to-end effective throughput and instruction loss rate, generate feedback results, and feed the feedback results back to the link state prediction unit (3). The link state prediction unit (3) uses the feedback result to adjust the weight parameters of the time series prediction model online, and the dynamic scheduling unit (4) uses the feedback result to update the flow table scheduling strategy.

10. The management system according to claim 9, characterized in that, The dynamic scheduling unit (4) is deployed between the airborne edge computing unit and the ground station control center; The dynamic scheduling unit (4) is responsible for the aggregation, encoding and distribution of uplink data, and the ground station control center is responsible for the fragmentation, scheduling and multicasting of downlink data, forming a bidirectional asymmetric link aggregation channel; The ground station control center maintains the global link topology map and sends auxiliary decision-making instructions for link switching to the dynamic scheduling unit (4) through the telemetry link.

Images