A method and apparatus for end-to-end transmission of live video in satellite-ground integrated networks
By dividing the management domain in the satellite-ground fusion network and using dual-homed edge nodes, combined with video frame priority and multi-agent reinforcement learning, the path loop and stuttering problems in live video transmission are solved, achieving low-latency and high-reliability video transmission, which is suitable for live broadcasting in remote areas and disaster emergency response scenarios.
Patent Information
- Application Number
- CN202611309336.5
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2026-08-27
- Publication Date
- 2026-11-10
AI Technical Summary
Existing satellite-ground integrated networks suffer from local congestion, repeated repairs, and path loops caused by unreasonable path selection in live video transmission. They also cannot effectively handle video stuttering caused by satellite switching or short-term link interruptions, especially when the topology of low-Earth orbit satellites changes rapidly.
Low-Earth orbit satellites are divided into multiple management domains, which are managed by medium-Earth orbit satellites. Inter-domain handover is achieved through dual-homed edge nodes. By combining video frame priority and multi-agent reinforcement learning, a cross-management domain routing method is constructed. An I-frame packet loss response mechanism and redundancy protection measures are set up to reduce the computational complexity and control overhead of routing.
Under conditions of rapid changes in low-Earth orbit satellite topology, it enables loop-free and repairable live video transmission, reduces video packet latency and loss, and improves the live video experience. It is suitable for scenarios such as live streaming in remote areas, maritime and aviation communications, field monitoring, and disaster emergency response.
Smart Images

Figure CN123075199A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of satellite communication and video transmission technology, and in particular to a method and apparatus for end-to-end transmission of live video for satellite-ground integrated networks. Background Technology
[0002] Services such as live video streaming, remote conferencing, online monitoring, and disaster emergency response require the real-time transmission of continuously generated video streams from cameras or mobile terminals to remote users. While existing terrestrial internet provides high bandwidth in urban areas, coverage remains insufficient in remote regions such as oceans, deserts, mountains, and aerospace applications. During natural disasters such as earthquakes, floods, and fires, terrestrial base stations and backbone links may also be disrupted. A space-terrestrial converged network, composed of low-Earth orbit satellites, ground stations, and terrestrial internet, can provide wide-area coverage and lower propagation latency, offering a new infrastructure for live video transmission in these scenarios.
[0003] The end-to-end transmission path in a space-ground integrated network typically consists of a ground-to-satellite link from the source ground station to the accessing LEO satellite, inter-satellite links between multiple LEO satellites, and a ground-to-satellite link from the destination LEO satellite to the destination ground station. LEO satellites orbit at low altitudes and move at high speeds, resulting in limited visibility time for individual satellites to ground stations. Furthermore, the adjacency relationships between satellites and the access relationships between ground stations change continuously over time. Simultaneously, constellation density, ground-to-satellite link bandwidth, round-trip time, signal-to-noise ratio, and node load vary significantly across different regions. Therefore, path selection requires not only determining the inter-satellite links but also simultaneously identifying the access satellites for both the source and destination ground stations.
[0004] Existing satellite routing methods often employ heuristic metrics such as shortest distance and minimum propagation delay, or periodically recalculate the end-to-end path based on link status. The former typically ignores satellite load, remaining link availability time, and ground-to-satellite link selection, easily leading to local congestion; the latter requires frequent synchronization of global topology and resource status, resulting in high control message and computational overhead in large-scale constellations. Some distributed learning methods select the next hop hop-by-hop based solely on local observations, lacking constraints on the stability of the complete path and the live broadcast period, potentially leading to repeated repairs, local detours, or path loops.
[0005] Live video transmission differs from ordinary data transmission; its user experience depends not only on average latency but also on startup latency, bitrate, number of stutters, stutter duration, bitrate fluctuations, and whether keyframes arrive completely. Mainstream video encoding typically uses image groups as the basic unit, where I-frames serve as references for decoding subsequent P-frames and B-frames. Losing an I-frame can lead to significant distortion within an image group. Existing routing mechanisms often treat all types of video frames equally; when satellite switching or short-term link interruptions occur, even if the average throughput meets requirements, prolonged stuttering may still occur. While research on adaptive video transmission using multi-focus single relay satellites for bend-tube relays has been conducted, the end-to-end transmission problem of live video under multi-hop inter-satellite link conditions remains unsolved.
[0006] Therefore, a transmission method is needed that can simultaneously sense satellite topology dynamics, ground station access quality, and the importance of video frames, complete cross-management domain routing without relying on global high-frequency synchronization, and take timely prefetching, repair, and redundancy protection measures when path failure, node congestion, and key frame packet loss risks increase, so as to meet the requirements of low latency, high reliability, and low stuttering for live video.
[0007] The information disclosed in this background section is intended only to enhance the understanding of the overall background of the invention and should not be construed as an admission or in any way implying that the information constitutes prior art known to those skilled in the art. Summary of the Invention
[0008] The purpose of this invention is to provide a method and apparatus for end-to-end live video transmission in satellite-ground integrated networks. It decomposes the global routing problem into multiple intra-domain path segments and combines ground station access selection, video frame priority, key frame time slot interleaving redundancy, and multi-agent reinforcement learning path decision-making. This enables loop-free, repairable, and controllable live video transmission under conditions of rapid changes in low-Earth orbit satellite topology. Thus, it can solve the end-to-end live video transmission problem under multi-hop inter-satellite link conditions and the problems of repeated repair, local detours, or path loops in the path.
[0009] To achieve the above objectives, this invention provides an end-to-end live video transmission method for satellite-ground converged networks. This method is applied to an end-to-end live video transmission system for satellite-ground converged networks, which includes a medium Earth orbit satellite, a low Earth orbit satellite, a source ground station, a destination ground station, a source low Earth orbit satellite, and a destination low Earth orbit satellite. The method specifically includes the following steps:
[0010] S1: Divide the low-Earth orbit satellite set into multiple management domains. Each management domain consists of multiple low-Earth orbit satellites, and assign a medium-Earth orbit satellite as a management node.
[0011] S2: At the boundary of adjacent management domains, select a low-orbit satellite that can maintain control contact with two management nodes at the same time as a dual-homed edge node;
[0012] S3: The source ground station and the destination ground station calculate the set of visible low-Earth orbit satellites in real time, select and access the optimal low-Earth orbit satellites as the source low-Earth orbit satellites and the destination low-Earth orbit satellites;
[0013] S4: The source ground station divides the encoded video into continuous task units according to the group of pictures (GOP). Each GOP contains multiple frames. Based on the video coding dependency, the frames are divided into intra-coded I-frames, predictive coded P-frames, and bidirectional predictive B-frames.
[0014] S5: The frame is divided into multiple network layer data packets for transport. Each data packet contains identification information, including GOP sequence number, frame sequence number, data packet sequence number, and frame type information.
[0015] S6: Generate the final path from the source ground station to the destination ground station, and send the data packet from the source ground station to the destination ground station;
[0016] S7: The destination ground station receives and rearranges the data packets according to the identification information to restore the encoded video.
[0017] In one embodiment of the present invention, generating the final path from the source ground station to the destination ground station specifically includes the following steps:
[0018] S61: Medium-Earth orbit satellites in the management domain of the source low-Earth orbit satellite will take the source low-Earth orbit satellite as the starting point of their path and the management domain of the source low-Earth orbit satellite as their current management domain.
[0019] S62: Determine the affiliation relationship between the target low-orbit satellite and the current management domain;
[0020] S63: If the target low-Earth orbit satellite is located in the current management domain, then the target low-Earth orbit satellite is taken as the path endpoint in the current management domain, and an optimal intra-domain path from the path start point to the path endpoint is generated.
[0021] If the target low-Earth orbit satellite is not located within the current management domain, then select an optimal dual-homed edge node within the current management domain as the path endpoint within the current management domain, and use the source low-Earth orbit satellite as the path starting point to generate the optimal intra-domain path within the current management domain.
[0022] S64: Take the management domain of the other management node that manages the optimal dual-homed edge node as the new current management domain, and take the optimal dual-homed edge node as the new path starting point;
[0023] S65: Repeat steps S62-S64 until the target low-Earth orbit satellite is within the current management domain;
[0024] S66: Output all optimal intra-domain paths and synthesize the final path;
[0025] In one embodiment of the present invention, selecting an optimal dual-homed edge node from the current management domain includes:
[0026] S631: Retrieve all dual-homed edge nodes in the current management domain;
[0027] Select the k-th station whose geographical location or estimated arrival time is closest to the destination. edge A set of dual-homed edge nodes as candidate nodes;
[0028] Based on the normalized queue load, the predicted delay of the current low-Earth orbit satellite to each dual-homed edge node, and the remaining dwell time of each dual-homed node, all dual-homed edge nodes in the candidate node set are scored.
[0029] The edge node with the highest score is selected as the optimal dual-homed edge node in the current management domain.
[0030] In one embodiment of the present invention, the step of generating the final path from the source ground station to the destination ground station and sending the data packet from the source ground station to the destination ground station further includes an I-frame packet loss handling mechanism, including:
[0031] Packet loss risk prediction: Medium-Earth orbit satellites estimate the packet loss probability of the optimal intra-domain path based on information such as historical packet loss rate, current queue occupancy, link contact window, repair rate, and recent jitter. ;
[0032] When the packet loss probability Greater than the threshold When the number of redundant tokens is not less than the preset value c2, a backup intra-domain path is configured for the I-frame. The backup intra-domain path does not intersect with any node of the optimal intra-domain path. When there is no path with completely non-intersecting nodes, the path with the fewest shared links is selected as the backup intra-domain path.
[0033] The backup intra-domain path and the optimal intra-domain path have a transmission time slot of Δt.
[0034] In one embodiment of the present invention, generating the optimal intra-domain path from the path start point to the path end point includes:
[0035] A shared model is constructed, which is an Actor-Critic model, including an Actor network and a Critic network; the shared model adopts a centralized training and decentralized execution CTDE mechanism.
[0036] A candidate path set is generated based on ephemeris data and the latest link statistics. This candidate path set contains K... p Candidate paths;
[0037] Using the current time as the decision time t, and based on the network environment and candidate path characteristics, construct the state vector of the medium-Earth orbit satellite at decision time t. , the state vector Input the Actor network to obtain K p Probability distribution of candidate paths ;
[0038] Set constraints and a masking function for the candidate path set so that candidate paths that do not meet the constraints are not selected. If there are paths that meet the constraints, then select the appropriate path from among those paths based on the probability distribution. The final selected path P is obtained. sel If none of the paths meet the constraints, the candidate path set is regenerated until a selected path is obtained.
[0039] In one embodiment of the present invention, the shared model is further updated;
[0040] This includes establishing a segmented reward mechanism to calculate the immediate reward value obtained at decision time t. ;
[0041] Batch collection The trajectory serves as the input to the model;
[0042] When the cumulative number of collections reaches the preset batch size N batch Or the preset time T since the last update train The model is updated when the time is right.
[0043] In one embodiment of the present invention, the source routing header of the data packet includes at least a low-Earth orbit satellite sequence, a next-hop cursor, and a backup next hop;
[0044] After reading the expected next-hop satellite node from the low-Earth orbit satellite sequence, the current low-Earth orbit satellite performs a security check and effective neighbor detection on the node. If the detection is successful, the data packet is forwarded. If the next hop is temporarily unavailable, and the backup next hop meets the requirements of not revisiting the visited node, contact window, and time limit, the backup next hop performs local fast rerouting to avoid sending control messages to the medium-Earth orbit satellite.
[0045] In one embodiment of the present invention, if the expected next hop and backup next hop are both unavailable, or the path lifetime expires, or the cursor goes out of bounds or the security check rejects the current path, the current low-Earth orbit satellite sends a repair request to the medium-Earth orbit satellite it manages.
[0046] Medium-Earth orbit (MEO) satellites use the current low-Earth orbit (LEO) satellites as a new starting point, planning a new suffix from the current LEO satellites to the end of the original path segment and covering the list of incomplete path nodes; the number of repairs is accumulated to K. repmax Then, it triggers the reselection of the path endpoint and the optimal intra-domain path from the current management domain.
[0047] Another object of the present invention is to provide an end-to-end live video transmission apparatus for satellite-ground converged networks, for implementing the end-to-end live video transmission method for satellite-ground converged networks as described in any of the preceding claims, comprising:
[0048] The management domain partitioning module is used to divide the low-Earth orbit satellite set into multiple management domains. Each management domain consists of multiple low-Earth orbit satellites and is assigned a medium-Earth orbit satellite as a management node.
[0049] The node selection module is used to select low-orbit satellites that can maintain control communication with two management nodes at the boundary of adjacent management domains as dual-homed edge nodes.
[0050] The access module is used to calculate the set of visible low-Earth orbit satellites in real time for the source ground station and the destination ground station, and to select and access the optimal low-Earth orbit satellites as the source low-Earth orbit satellites and the destination low-Earth orbit satellites.
[0051] The task partitioning module is used by the source ground station to divide the encoded video into continuous task units according to the group of pictures (GOP). Each GOP contains multiple frames. According to the video coding dependency, the frames are divided into intra-coded I-frames, predictive coded P-frames, and bidirectional predictive B-frames.
[0052] The segmentation module is used to segment frames into multiple network layer data packets for transport. Each data packet contains identification information, including GOP sequence number, frame sequence number, data packet sequence number, and frame type information.
[0053] The path generation module is used to generate the final path from the source ground station to the destination ground station and send the data packet from the source ground station to the destination ground station.
[0054] The receiving module is used to receive data packets at the destination ground station and rearrange them according to the identification information to recover the encoded video.
[0055] Another object of the present invention is to provide an electronic device, including a memory, a processor, a RAM, and a computer program stored in the memory and executable on the processor, characterized in that, when the processor executes the computer program, it implements the steps of a live video end-to-end transmission method for a satellite-ground converged network as described in any of the preceding claims.
[0056] Compared with existing technologies, this invention provides a method and apparatus for end-to-end live video transmission in a space-ground integrated network. Utilizing hierarchical management, it divides management domains for highly dynamic low-Earth orbit (LEO) satellites using relatively stable medium-Earth orbit (MEO) satellites. It configures primary and backup management nodes and dual-homed edge nodes for LEO satellites and selects ground stations to access satellites based on visible link duration, round-trip delay, signal-to-noise ratio (SNR), and load, thereby achieving hierarchical satellite management and significantly reducing the complexity of routing calculations and control signaling overhead. Cross-domain segmented routing is implemented, with each MEO satellite only planning path segments within its own management domain. Inter-domain handover is completed at the dual-homed edge nodes. Loop-free, low-overhead forwarding is achieved through next segment prefetching, source route header forwarding, local fast rerouting, intra-domain repair, and lifetime constraints. An I-frame packet loss prevention mechanism is implemented, maintaining intra-domain path consistency at the image group level. I-frames are weighted and prioritized according to their importance (I-frames, P-frames, and B-frames). When the predicted packet loss rate is high and the redundancy budget allows, I-frames are time-slot interleaved and copied on paths that are as non-overlapping as possible. Simultaneously, intelligent path decision-making is employed to construct a multi-agent reinforcement learning model with a centralized training and decentralized execution mechanism. Based on orbital phase, node load, source and destination states, and the latency, hop count, stability, remaining time before failure, and deadline margin of candidate paths, the optimal intra-domain path is selected. Furthermore, safety action shielding and segmented feedback constrain the path to meet the live broadcast deadline. This adapts to rapid changes and frequent switching in low-Earth orbit satellite topology, reducing video packet latency and loss under limited control overhead, minimizing buffering, and improving the live video experience. It is suitable for scenarios such as live broadcasting in remote areas, maritime and aviation communications, field monitoring, disaster emergency response, and global video distribution. Attached Figure Description
[0057] Figure 1 This is a flowchart illustrating an end-to-end live video transmission method for a satellite-ground converged network according to an embodiment of the present invention.
[0058] Figure 2 This is a schematic diagram of the final path of a live video end-to-end transmission method for a satellite-ground converged network according to an embodiment of the present invention.
[0059] Figure 3 This is a schematic diagram of the transmission framework at each level of a live video end-to-end transmission method for a satellite-ground converged network according to an embodiment of the present invention.
[0060] Figure 4 This is a schematic diagram of the repair process for an end-to-end live video transmission method for a satellite-ground converged network according to an embodiment of the present invention.
[0061] Figure 5 This is a schematic diagram of an I-frame packet loss response mechanism for an end-to-end live video transmission method for a satellite-ground converged network according to an embodiment of the present invention.
[0062] Figure 6 This is a schematic diagram of the shared model decision-making process of a live video end-to-end transmission method for satellite-ground converged networks according to an embodiment of the present invention.
[0063] Figure 7 This is a schematic diagram of the structure of an end-to-end live video transmission device for a satellite-ground converged network according to an embodiment of the present invention. Detailed Implementation
[0064] The specific embodiments of the present invention will now be described in detail with reference to the accompanying drawings, but it should be understood that the scope of protection of the present invention is not limited to the specific embodiments.
[0065] Unless otherwise expressly stated, throughout the specification and claims, the term "comprising" or its variations such as "including" or "comprises" shall be understood to include the stated elements or components without excluding other elements or other components.
[0066] Inventive concept:
[0067] This invention employs a medium-Earth orbit (MEO) satellite to manage the data plane of low-Earth orbit (LEO) satellites, decomposing large-scale global path calculations into multiple intra-domain path segments. Each MEO satellite maintains only the fine-grained state of a limited number of LEO satellites, and cross-domain exchanges are limited to edge prefetching and feedback messages. This allows the state space, computational complexity, and control overhead to increase with the size of the management domain rather than with the size of the global constellation, making it suitable for large-scale multi-orbit constellation deployments.
[0068] By considering the selection of satellites for ground station access in conjunction with inter-satellite routing, and by comprehensively evaluating factors such as remaining visibility time, RTT, SNR, and load, we can avoid frequent handovers and uplink bottlenecks caused by selecting only the nearest satellite. We can also reduce cross-domain handover waiting by using dual-homed edge nodes and next-segment prefetching.
[0069] By employing source routing headers, backup next hops, intra-domain repair, path lifetime, and access node constraints, forwarding can continue even when the low-orbit topology changes rapidly, and the path must be loop-free. The repair count limit and exponential backoff can prevent control storms, and the primary and backup mid-orbit management relationship can improve the availability of the control plane.
[0070] By incorporating video coding dependencies into network forwarding, different priorities are set for I-frames, P-frames, and B-frames, and I-frames are only copied when the risk is high and the token budget allows. Off-path and time-slot interleaving makes redundant copies less susceptible to simultaneous fluctuations on the same link, improving keyframe integrity and reducing decoding distortion with controllable bandwidth overhead.
[0071] The multi-agent reinforcement learning model directly optimizes the live broadcast duration, path stability, latency, packet loss, jitter, and repair costs. At the same time, it uses safety actions to filter out candidate paths that do not meet the hard constraints. Segmented rewards enable each mid-track agent to obtain clear feedback on their responsibilities, taking into account learning effectiveness, interpretability, and operational safety.
[0072] This enables the construction of a low-latency live video platform covering remote areas, oceans, and airspace, providing stable video transmission capabilities for wildlife and infrastructure monitoring, maritime and aviation communications, distance education, news broadcasting, and disaster emergency response, and has high engineering application value and economic and social benefits.
[0073] Example 1:
[0074] like Figures 1 to 7 As shown, a preferred embodiment of the present invention provides an end-to-end live video transmission method for a satellite-ground integrated network, applied to a live video end-to-end transmission system for a satellite-ground integrated network. The system includes a medium Earth orbit satellite, a low Earth orbit satellite, a source ground station, a destination ground station, a source low Earth orbit satellite, and a destination low Earth orbit satellite. The method specifically includes the following steps:
[0075] S1: Divide the low-Earth orbit satellite set into multiple management domains. Each management domain consists of multiple low-Earth orbit satellites, and assign a medium-Earth orbit satellite as a management node.
[0076] Specifically, the management domain is divided as follows: based on the orbital ephemeris, geographical location, and relative motion relationships of the medium-Earth orbit (MEO) and low-Earth orbit (LEO) satellites, the LEO satellite set is divided into multiple management domains. Each MEO satellite manages a subset of LEO satellites, maintaining the adjacency relationships, inter-satellite link status, queue load, contact windows, and coverage mapping from ground stations to LEO satellites within its domain. Adjacent MEO satellites exchange boundary summaries and prefetch messages via inter-satellite links without needing to synchronize the complete global LEO topology. Within each management domain, in addition to selecting an optimal MEO satellite as the management node, a suboptimal MEO satellite is also selected as a backup management node. LEO satellites periodically receive heartbeats from the primary and backup management nodes. When a management node's heartbeat is continuously lost, the path request is switched to the backup management node to avoid single points of failure. The method for selecting the best medium-Earth orbit (MEO) satellite is as follows: Candidate nodes are scored according to the principles of "longer expected remaining control link duration, lower control round-trip delay, and shorter spatial distance," after normalizing these indicators. Candidate nodes that are currently invisible, have an expected control link duration below a threshold, or a control delay above a threshold are not included in the ranking. The MEO satellite with the highest score is configured as the primary MEO management node. The medium-orbit satellite with the second-highest score and meeting the above validity conditions will be configured as a backup management node. .
[0077] S2: At the boundary of adjacent management domains, select a low-orbit satellite that can maintain control contact with two management nodes at the same time as a dual-homed edge node;
[0078] Specifically, a management domain can be adjacent to multiple management domains, and multiple dual-homed edge nodes can be simultaneously retained at the boundary of any pair of adjacent management domains. Therefore, candidate edge nodes of this management domain can come from multiple nodes on the same boundary or from multiple boundaries of different adjacent management domains; each edge node records the identifiers of its interleaved adjacent management domains. Dual-homed edge nodes simultaneously store the identifiers and control channels of two adjacent management domains, used to receive cross-domain data packets and trigger the next routing request. Management domains and edge nodes follow T... update Periodic recalculation; because intra-domain topology changes are typically slower than global topology changes, T update It can be set from a few seconds to tens of seconds.
[0079] S3: The source ground station and the destination ground station calculate the set of visible low-Earth orbit satellites in real time, select and access the optimal low-Earth orbit satellites as the source low-Earth orbit satellites and the destination low-Earth orbit satellites;
[0080] Specifically, after the source ground station determines the source LEO satellite to be accessed, it sends information such as the destination ground station identifier, video stream identifier, image group sequence number, playback period, and frame type to the accessing LEO satellite. The accessing LEO satellite then initiates a routing request to its main MOR management node. The destination ground station also maintains the current destination LEO access satellite using the same scoring method, and the relevant mapping is periodically synchronized by the MOR satellite.
[0081] The definition of the visible low-Earth orbit satellite set is as follows:
[0082] (1)
[0083] in, The elevation angle of the low-Earth orbit satellite L. min The minimum elevation angle threshold can be set from 5° to 15°.
[0084] For each visible low-Earth orbit satellite L, estimate the remaining link duration T before it leaves the visible range of the ground station. link (L), to measure or predict the round-trip delay of the current Earth-Satellite link. Signal-to-noise ratio and load The features are normalized to [0,1] respectively, and the access feature vector is constructed and the access score is calculated according to equation (3):
[0085] (2)
[0086] (3)
[0087] Among them, w T w R w S and w L These are non-negative weights, used to evaluate balanced link stability, latency, signal quality, and congestion level, respectively. Access score is determined. The visible low-Earth orbit satellite with the highest access score is selected as the access satellite.
[0088] S4: The source ground station divides the encoded video into continuous task units according to the group of pictures (GOP). Each GOP contains multiple frames. Based on the video coding dependency, the frames are divided into intra-coded I-frames, predictive coded P-frames, and bidirectional predictive B-frames.
[0089] Specifically, the weighting relationship:
[0090] (4)
[0091] Both low-Earth orbit (LEO) and medium-Earth orbit (MEO) satellites can maintain weighted priority queues, directly read lightweight frame type tags, prioritize forwarding I-frames, then P-frames, and finally B-frames; no deep parsing of the video payload is required.
[0092] S5: The frame is divided into multiple network layer data packets for transport. Each data packet contains identification information, including GOP sequence number, frame sequence number, data packet sequence number, and frame type information.
[0093] Specifically, within the same management domain, the same medium-Earth orbit satellite uniformly selects the main path for the same GOP, avoiding serious out-of-order issues caused by frequent route switching of data packets within the same image group.
[0094] S6: Generate the final path from the source ground station to the destination ground station, and send the data packet from the source ground station to the destination ground station;
[0095] Specifically, the source low-Earth orbit satellite L src After accessing the source ground station, a routing request is sent to the management node M1 of its management domain. The request must include at least the destination ground station identifier DEST_GS, video stream and image group identifiers, frame type, allowed end-to-end latency or playback duration, and current queue and link state summary. Then, M1 queries the coverage mapping from the ground station to the low-Earth orbit satellite to obtain the target low-Earth orbit satellite L for the current coverage destination ground station. dest If L dest If it is located within the management domain of M1, then L will be... dest As the endpoint of this path segment; if L dest If located in another administration domain, one of the dual-homed edge nodes in this administration domain will be selected as the endpoint of this path segment. The destination ground station can be covered by multiple low-Earth orbit satellites at the same time. Query the latest coverage mapping from the ground station to the low-Earth orbit satellite, and score the currently visible target low-Earth orbit satellites according to the access scoring method. Select the one with the highest score as the current target access satellite L. dest If L already exists dest If the satellite remains visible and its score differs from the current best value by no more than a preset hysteresis threshold, the original access satellite will be maintained to avoid frequent switching. Each new management domain reconfirms the current L upon receiving a prefetch or on-demand request. dest If the L dest If it is located within the current management domain, then it is taken as the endpoint of this path segment; otherwise, one of the dual-homed edge nodes in this management domain is selected as the endpoint of this path segment.
[0096] S7: The destination ground station receives and rearranges the data packets according to the identification information to restore the encoded video.
[0097] Specifically, after receiving the data packet, the destination ground station summarizes the actual latency, packet loss rate, jitter, hop count, repair count, early failure status, and first-delivery copy identifier for each management domain, in units of image groups, and returns the feedback in batches to the medium-Earth orbit satellites participating in the decision-making process to update the candidate path statistics and the shared model.
[0098] As an optional implementation, generating the final path from the source ground station to the destination ground station specifically includes the following steps:
[0099] S61: Medium-Earth orbit satellites in the management domain of the source low-Earth orbit satellite will take the source low-Earth orbit satellite as the starting point of their path and the management domain of the source low-Earth orbit satellite as their current management domain.
[0100] S62: Determine the affiliation relationship between the target low-orbit satellite and the current management domain;
[0101] S63: If the target low-Earth orbit satellite is located in the current management domain, then the target low-Earth orbit satellite is taken as the path endpoint in the current management domain, and an optimal intra-domain path from the path start point to the path endpoint is generated.
[0102] If the target low-Earth orbit satellite is not located within the current management domain, then select an optimal dual-homed edge node within the current management domain as the path endpoint within the current management domain, and use the source low-Earth orbit satellite as the path starting point to generate the optimal intra-domain path within the current management domain.
[0103] S64: Take the management domain of the other management node that manages the optimal dual-homed edge node as the new current management domain, and take the optimal dual-homed edge node as the new path starting point;
[0104] S65: Repeat steps S62-S64 until the target low-Earth orbit satellite is within the current management domain;
[0105] S66: Output all optimal intra-domain paths and synthesize the final path.
[0106] Specifically, after generating the optimal intra-domain path, the low-Earth orbit satellite sequence is formed in the following order:
[0107] (5)
[0108] This sequence is then written into the compact source routing header of the data packet. When the endpoint of this path segment is a dual-homed edge node, M1 sends a prefetch notification to the neighboring medium-Earth orbit satellite M2, which manages the edge node, via the medium-Earth orbit inter-satellite link. The notification carries L... edge The data packet contains DEST_GS, video stream and image group identifiers, and estimated arrival times. M2 pre-calculates the next path segment before the data packet arrives; if the prefetch message is lost or the lead time is insufficient, the edge node initiates an on-demand request to M2 after the data packet arrives to ensure correct functionality. After the data packet arrives at the dual-homed edge node, the edge node switches to the adjacent medium-Earth orbit satellite M2, which is closer to the destination ground station. M2 overwrites PATH_VEC with the prefetched or real-time calculated next path segment, resets PATH_IDX, and adds its own identifier to MEO_ID_List. The above process is executed recursively until the path segment endpoint is L. dest And complete the downward movement.
[0109] As an optional implementation, selecting an optimal dual-homed edge node from the current management domain includes:
[0110] S631: Retrieve all dual-homed edge nodes in the current management domain;
[0111] Select the k-th station whose geographical location or estimated arrival time is closest to the destination. edge A set of dual-homed edge nodes as candidate nodes;
[0112] Based on the normalized queue load, the predicted delay of the current low-Earth orbit satellite to each dual-homed edge node, and the remaining dwell time of each dual-homed node, all dual-homed edge nodes in the candidate node set are scored.
[0113] The edge node with the highest score is selected as the optimal dual-homed edge node in the current management domain.
[0114] Specifically, k edge These are preset values. The edge node scores are calculated using the following formula:
[0115] (6)
[0116] Among them, T dwell (L e ) represents the candidate edge node L in the candidate node set.e As the remaining residency time of the dual-homed node, Load(L) e ) represents the normalized queue load, d(L) e ) represents the predicted time delay from the current low-Earth orbit satellite to this edge node. Candidate dual-homed edge node L e The orbital period of the given orbit is used as a normalized scale for the remaining dwell time. If candidate nodes are located in the same orbital layer, the orbital period of that layer can be used. γ=[γ1,γ2,γ3] T Let be the weight vector, and , , All are non-negative weights, and their sum can be set to 1; A preset normalized upper bound for the prediction latency of candidate edge nodes; Candidate dual-homed edge nodes Overall score. Edge node scoring. Largest dual-homed edge node L edge As the end point of the path segment.
[0117] As an optional implementation, the process of generating the final path from the source ground station to the destination ground station and sending the data packet from the source ground station to the destination ground station also includes an I-frame packet loss handling mechanism, including:
[0118] Packet loss risk prediction: Medium-Earth orbit satellites estimate the packet loss probability of the optimal intra-domain path based on information such as historical packet loss rate, current queue occupancy, link contact window, repair rate, and recent jitter. ;
[0119] Specifically, the probability of packet loss on the main path Existing online statistical or lightweight prediction methods can be used for calculation. For example, the historical packet loss rate, current queue occupancy rate, link contact window exhaustion risk, repair rate, and normalized jitter of several recent GOPs or path segments can be calculated using sliding window statistics or exponentially weighted moving averages, and then fused using non-negative weighting and truncated to [0,1]. .
[0120] When the packet loss probability Greater than the threshold When the number of redundant tokens is not less than the preset value c2, a backup intra-domain path is configured for the I-frame. The backup intra-domain path does not intersect with any node of the optimal intra-domain path. When there is no path with completely non-intersecting nodes, the path with the fewest shared links is selected as the backup intra-domain path.
[0121] The backup intra-domain path and the optimal intra-domain path have a transmission time slot of Δt.
[0122] Specifically, each management domain sets up an I-frame redundancy token bucket B. dup Each additional I-frame replica consumes c2 tokens, which are restored at a set rate to limit redundant traffic.
[0123] The number of I-frame copies is determined by the following formula:
[0124] (7)
[0125]
[0126] That is, when the predicted packet loss rate p̂ of the main path loss Greater than the threshold And N is set when the number of redundant tokens is not less than c2. dup =2, otherwise set N dup =1. The replica count rule is independent of reinforcement learning actions and is controlled by deterministic safety rules, preventing the model from infinitely increasing redundant throughput for short-term rewards. For N dup For each I-frame, two intra-domain paths with non-intersecting nodes are selected. When no completely non-intersecting paths exist, the system degenerates into the two paths with the fewest shared links. The two replicas are not sent consecutively, but rather in two time slots with an interval of Δt, to reduce the probability of simultaneous impact from transient fluctuations, short-term blockages, or handover interruptions on the same link on both replicas. After receiving the data packets, the destination ground station rearranges the packets according to the GOP sequence number, frame sequence number, and timestamp. For I-frame replicas, the first complete replica that arrives and meets the playback deadline participates in decoding; subsequent duplicate replicas are discarded. For lost P-frames or B-frames, error concealment methods such as adjacent frame interpolation and reference frame duplication can be used. After playback is complete, the actual keyframe recovery status and redundancy overhead are fed back to the relevant medium-Earth orbit satellites.
[0127] As an optional implementation, generating the optimal intra-domain path from the path's starting point to its ending point includes:
[0128] A shared model is constructed, which is an Actor-Critic model, including an Actor network and a Critic network; the shared model adopts a CTDE mechanism of centralized training and decentralized execution.
[0129] Specifically, the centralized training and decentralized execution CTDE mechanism means that during the training phase, trajectories and feedback updates of shared parameters from multiple management domains are aggregated, while during the operation phase, only parameters from the current management domain are used for inference, and prefetching and feedback messages are exchanged only with medium-Earth orbit satellites in adjacent management domains.
[0130] A candidate path set is generated based on ephemeris data and the latest link statistics. This candidate path set contains K... p Candidate paths;
[0131] Specifically, the medium-Earth orbit satellite constructs a domain contact map with a time dimension based on ephemeris data, and asynchronously pre-calculates the Top-K candidate paths from each low-Earth orbit satellite to the target low-Earth orbit satellite or edge node in the background. Each path records estimated latency, hop count, link stability probability, shortest remaining contact time, node load, and remaining playback time, for later path quality feedback. When a real-time request arrives, intelligent path decision-making can proceed without traversing the entire map.
[0132] Using the current time as the decision time t, and based on the network environment and candidate path characteristics, construct the state vector of the medium-Earth orbit satellite at decision time t. , the state vector Input the Actor network to obtain K p Probability distribution of candidate paths ;
[0133] Specifically, the Actor network for K p The output probability distribution of the candidate paths is defined by the action space according to equation (8):
[0134] (8)
[0135] Action a t This indicates the selection of one of the paths; the Critic network estimates the expected cumulative reward V(s) of the current state. t Candidate paths are pre-calculated by the backend based on ephemeris and the latest link statistics, enabling online inference to primarily perform neural network forward computation without requiring real-time traversal of large-scale satellite imagery. During the training phase, sampling can be performed according to this probability distribution to maintain exploration. When the model is unavailable, the path with the minimum expected latency and stability threshold is selected from the feasible candidates. Once selected, these paths are sequentially arranged to form a low-Earth orbit satellite sequence.
[0136] At decision time t, the state vector The construction is shown in the following formula:
[0137] (9)
[0138] (10)
[0139]
[0140]
[0141]
[0142] Among them, P1 to P Candidate paths from the requested low-Earth orbit satellite to the endpoint of this path segment; global features Including the sine value of the orbital phase cosine value of orbital phase Average load within the domain and recent repair rate Source and destination characteristics Including source node queue usage Round-trip time to the destination station / and target low-orbit satellite visibility ; d is the predicted segment delay, h is the number of hops, p stab The probability that a path remains available during its soft lifetime is given by TTB, which is the shortest remaining contact time within the path. This represents the remaining playback time. All continuous features are truncated and normalized to [0,1]; there are insufficient candidate paths (K). p Zero padding and validity masks are used for the entries. Among them, A preset normalized upper bound is used to predict segmented delays and timeout margins for candidate paths. The available segment duration for the current video task on this path segment. This represents the maximum number of hops allowed for a path within a single domain; additionally... The source node queue capacity or a predefined upper bound for queue normalization. This is the preset normalized upper bound for the round-trip time delay to the destination station.
[0143] Set constraints and a masking function for the candidate path set so that candidate paths that do not meet the constraints are not selected. If there are paths that meet the constraints, then select the appropriate path from among those paths based on the probability distribution. The final selected path P is obtained. sel If none of the paths meet the constraints, the candidate path set is regenerated until a selected path is obtained.
[0144] Specifically, for candidate path P j Define three constraints and a masking function:
[0145]
[0146]
[0147]
[0148]
[0149] The three conditions C1, C2, and C3 represent the scenarios in which the candidate path violates the time slack, stability, and single-hop contact remaining time requirements, respectively. The path is retained when none of the three violation conditions are met, and the path is blocked as soon as any one of the violation conditions is met. This represents the remaining playback time margin of the candidate path after satisfying the predicted segmentation delay. This is the preset minimum segmentation deadline margin. Indicates the stability of the candidate path. Used to represent the minimum permissible stability threshold. This represents the remaining contact time of a single-hop link in the candidate path. The minimum single-hop contact remaining time threshold is set to ensure that the path remains available during forwarding.
[0150] When Mask(P) j When )=1, the policy logarithm of the path is set to negative infinity before Softmax, so that it cannot be selected. If all candidate paths are blocked, candidate set expansion, edge node reselection, or conservative shortest stable path is triggered.
[0151] As an optional implementation, updating the shared model is also included;
[0152] This includes establishing a segmented reward mechanism to calculate the immediate reward value obtained at decision time t. ;
[0153] Batch collection (s t ,a t ,r t ,s t +1) The trajectory, used as input to the model;
[0154] When the cumulative number of collections reaches the preset batch size N batch Or the preset time T since the last update train The model is updated when the time is right.
[0155] Specifically, medium-Earth orbit satellites collect data in batches of one group of orbiting orbits (GOPs) or multiple GOPs. t ,a t ,r t ,s t +1) The trajectory is updated on a ground training center or a medium-Earth orbit satellite with computing capabilities, and then the new version is distributed to each medium-Earth orbit satellite. During operation, version number and PlanID are used to ensure policy consistency within the same GOP; when the model is unavailable, it degenerates to the minimum latency stable path that meets the safety shielding conditions to ensure continuous operation of the data plane. When the cumulative number of collections reaches the preset batch size N... batch Or the preset time T since the last update train The model update is triggered at specific times. In one embodiment, N batch Take 64 valid GOPs / path segments as feedback, T trainA 60-second timeframe is used; however, if the recent repair rate, packet loss rate, or average reward deteriorates beyond a threshold for multiple consecutive statistical windows, it can be triggered earlier. The model is updated and validated at a ground training center or on a medium-Earth orbit satellite with computing capabilities, and a new version is distributed after successful validation. During operation, the version is switched only at GOP boundaries, and version number and PlanID are used to ensure policy consistency within the same GOP. When the model is unavailable, it degrades to the minimum latency stable path that meets the safety shielding conditions. The calculation formula is as follows:
[0156] (11)
[0157]
[0158]
[0159] (12)
[0160] in, This represents the path progress at the time of the repair. W f The importance weights of I / P / B frames are assigned based on the current data packet frame type, i.e., I frames, P frames, and B frames are each assigned a preset weight. , and And usually , Indicates whether the data packet arrives at the end of the path segment before the deadline, l̂, p̂ loss and ĵ represent normalized piecewise latency, packet loss rate, and jitter, respectively, h(P) sel ) represents the number of hops for the selected path; and These indicate that repair has occurred and the path has failed prematurely, respectively. This represents the segmented reward value. In the formula, To ensure timely delivery of the bonus coefficient, , , and These are the segmented latency, packet loss rate, jitter, and path hop count penalty coefficients, respectively. The maximum number of hops allowed for a single domain path; The penalty factor for each additional redundant copy, The number of copies sent in the current I-frame; The penalty coefficient for each in-domain repair. This is the penalty coefficient for premature path failure; This is the repair progress compensation coefficient. This indicates the path progress when a repair occurs, where This represents the number of remaining hops in the original path segment that were not yet completed during the repair process. This is the original planned total number of hops for this path segment, therefore Each medium-Earth orbit satellite is only responsible for its own planned path segment to avoid difficulties in allocating end-to-end rewards. Only the first successful delivery receives a positive delivery reward; late deliveries are not rewarded repeatedly.
[0161] As an optional implementation, the source routing header of the data packet includes at least a low-Earth orbit satellite sequence, a next-hop cursor, and an alternate next hop;
[0162] After reading the expected next-hop satellite node from the low-Earth orbit satellite sequence, the current low-Earth orbit satellite performs a security check and effective neighbor detection on the node. If the detection is successful, the data packet is forwarded. If the next hop is temporarily unavailable, and the backup next hop meets the requirements of not revisiting the visited node, contact window, and time limit, the backup next hop performs local fast rerouting to avoid sending control messages to the medium-Earth orbit satellite.
[0163] Specifically, the source routing header includes at least: PATH_VER, PATH_LEN, PATH_IDX, PATH_TTL, PATH_VEC, PATH_VEC, ALT_HOP, DEST_GS, PlanID, MEO_ID_List of medium-Earth orbit satellites participating in planning or repair, and K, the maximum number of in-domain repairs. repmax HOP_CLOCK per hop soft timeout, N number of I-frame replicas dup And an optional integrity check (SIG). Low-Earth orbit (LEO) satellites only need to read this header to forward the message; there's no need to execute the learning model hop-by-hop. Current LEO satellite L... cur The expected next hop is read as PATH_VEC[PATH_IDX+1]. If the node is still a valid neighbor and passes the security check, the packet is forwarded and PATH_IDX is incremented by 1, while PATH_TTL is decremented by the hop count or actual time. If the expected next hop is temporarily unavailable, and ALT_HOP meets the requirements of not revisiting visited nodes, contact window, and timeout, then ALT_HOP is used to perform local fast rerouting to avoid sending control messages to medium Earth orbit satellites.
[0164] As an alternative implementation, if the expected next hop and backup next hop are both unavailable, or the path lifetime expires, or the cursor goes out of bounds, or the security check rejects the current path, the current low-Earth orbit satellite sends a repair request to the medium-Earth orbit satellite it manages.
[0165] Medium-Earth orbit (MEO) satellites use the current low-Earth orbit (LEO) satellites as a new starting point, planning a new suffix from the current LEO satellites to the end of the original path segment and covering the list of incomplete path nodes; the number of repairs is accumulated to K. repmax Then, it triggers the reselection of the path endpoint and the optimal intra-domain path from the current management domain.
[0166] Specifically, if neither the expected next hop nor the backup next hop is available; or the path time-to-live (TTL) expires; or the cursor goes out of bounds; or the security check rejects the current path, a repair request is sent if any of the above conditions are not met. When the expected next hop and ALT_HOP are both unavailable, PATH_TTL expires, the cursor goes out of bounds, or the security check rejects the current path, L... cur Send a repair request to the medium Earth orbit satellite it manages. The medium Earth orbit satellite uses an L-shaped orbit. cur As a new starting point, a new suffix is planned to reach the end of the original path segment, overwriting any incomplete PATH_VEC; the number of repairs is accumulated to K. repmax Afterwards, a new path segment endpoint and complete path are selected from the current management domain. This includes: each newly planned or repaired PATH_VEC must not contain duplicate low-Earth orbit satellites; the repair suffix must not contain nodes already visited by the data packet, which can be recorded through lightweight access sets or Bloom filters; PATH_IDX can only increase monotonically and is only reset during repair or cross-domain handover; forwarding with the original header is prohibited when PATH_TTL is not positive; cross-domain handover must occur at dual-homed edge nodes. If repairs fail consecutively within a short period, the exponential backoff time is calculated using the following formula:
[0167] (13)
[0168] Where c represents the number of consecutive failures. Initial repair retreat.
[0169] The main fields and their functions in the source routing header are shown in the table below:
[0170] field group Example fields effect Path description PATH_LEN, PATH_IDX, PATH_VEC Provide the ordered low-Earth orbit satellite sequence and relay cursor within the current management domain. Timeliness and tolerance for error PATH_TTL, HOP_CLOCK, ALT_HOP Restrict the continued use of outdated paths and support fast local rerouting for short-term failures. Consistency Identifier PATH_VER, PlanID, MEO_ID_List Identify the planning version and record the medium-Earth orbit satellites involved in the planning and repair. Video and Purpose Information <![CDATA[DEST_GS、N dup Image group / frame markers]]> Supports cross-domain destination resolution, I-frame copy control, and video priority forwarding. Security Information <![CDATA[K repmax SIG, Access Summary Limit the number of repair attempts, verify path integrity, and prevent node revisits.
[0171] Example 2:
[0172] In one embodiment, the invention may be deployed with the following parameters, but the listed parameters do not constitute a limitation on the scope of protection:
[0173] Management domain update cycle T update Take 30 seconds; integrate scoring weights (w) T ,w R ,w S ,w L The edge node scoring weights (γ1, γ2, γ3) are set to (0.4, 0.3, 0.3); the number of candidate paths K is set to 6; and the maximum number of hops per segment H is set to (0.3, 0.3, 0.2, 0.2). max Set the weight to 10; I / P / B frame weights are set to (1.0, 0.6, 0.3); minimum segment timeout margin Δ min Take 80 milliseconds, path stability threshold p stab,minTake 0.85, minimum contact time T for a single jump hop,min Take 1 second, I-frame redundancy trigger threshold Take 0.05, redundant budget B dup Take 10 tokens, and set the initial repair backoff time T0 to 100 milliseconds.
[0174] In one end-to-end simulation embodiment, the source and destination ground stations are located in a long-distance region, utilizing a low-Earth orbit constellation and a medium-Earth orbit management node to transmit 1080p, 30 frames per second live video. Compared to schemes that only route based on distance or latency and do not distinguish between video frames, this invention maintains a high delivery rate even at high packet generation rates and significantly reduces path repair frequency and video stuttering; specifically, accessing satellite scoring reduces long stutters during handover, and dual I-frame replicas increase the probability of keyframe arrival with minimal additional latency. This embodiment illustrates the synergistic effect between domain-specific routing, video prioritization, and reinforcement learning path decision-making.
[0175] Example 3:
[0176] This embodiment provides a live video end-to-end transmission device for satellite-ground converged networks, used to implement a live video end-to-end transmission method for satellite-ground converged networks as described in any of the above embodiments, including:
[0177] The management domain partitioning module is used to divide the low-Earth orbit satellite set into multiple management domains. Each management domain consists of multiple low-Earth orbit satellites and is assigned a medium-Earth orbit satellite as a management node.
[0178] The node selection module is used to select low-orbit satellites that can maintain control communication with two management nodes at the boundary of adjacent management domains as dual-homed edge nodes.
[0179] The access module is used to calculate the set of visible low-Earth orbit satellites in real time for the source ground station and the destination ground station, and to select and access the optimal low-Earth orbit satellites as the source low-Earth orbit satellites and the destination low-Earth orbit satellites.
[0180] The task partitioning module is used by the source ground station to divide the encoded video into continuous task units according to the group of pictures (GOP). Each GOP contains multiple frames. According to the video coding dependency, the frames are divided into intra-coded I-frames, predictive coded P-frames, and bidirectional predictive B-frames.
[0181] The segmentation module is used to segment frames into multiple network layer data packets for transport. Each data packet contains identification information, including GOP sequence number, frame sequence number, data packet sequence number, and frame type information.
[0182] The path generation module is used to generate the final path from the source ground station to the destination ground station and send the data packet from the source ground station to the destination ground station.
[0183] The receiving module is used to receive data packets at the destination ground station and rearrange them according to the identification information to recover the encoded video.
[0184] Example 4:
[0185] This embodiment provides an electronic device, including a memory, a processor, a RAM, and a computer program stored in the memory and executable on the processor. The device is characterized in that, when the processor executes the computer program, it implements the steps of a live video end-to-end transmission method for a satellite-ground converged network as described in any of the preceding embodiments.
[0186] Those skilled in the art will understand that embodiments of this application can be provided as methods, systems, or computer program products. Therefore, this application can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this application can take the form of a computer program product embodied on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.
[0187] This application is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this application. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart... Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.
[0188] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.
[0189] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.
[0190] The foregoing description of specific exemplary embodiments of the invention is for illustrative and explanatory purposes. These descriptions are not intended to limit the invention to the precise forms disclosed, and it will be apparent that many changes and variations can be made in accordance with the foregoing teachings. The exemplary embodiments were chosen and described in order to explain the specific principles of the invention and its practical application, thereby enabling those skilled in the art to implement and utilize various different exemplary embodiments of the invention, as well as various different choices and variations. The scope of the invention is intended to be defined by the claims and their equivalents.
Claims
1. A method for end-to-end transmission of live video in a space-ground integrated network, applied to a live video end-to-end transmission system for a space-ground integrated network, the system comprising a medium-Earth orbit satellite, a low-Earth orbit satellite, a source ground station, a destination ground station, a source low-Earth orbit satellite, and a destination low-Earth orbit satellite, characterized in that, The method specifically includes the following steps: S1: Divide the low-Earth orbit satellite set into multiple management domains. Each management domain consists of multiple low-Earth orbit satellites, and assign a medium-Earth orbit satellite as a management node. S2: At the boundary of adjacent management domains, select a low-orbit satellite that can maintain control contact with two management nodes at the same time as a dual-homed edge node; S3: The source ground station and the destination ground station calculate the set of visible low-Earth orbit satellites in real time, select and access the optimal low-Earth orbit satellites as the source low-Earth orbit satellites and the destination low-Earth orbit satellites; S4: The source ground station divides the encoded video into continuous task units according to the group of pictures (GOP). Each GOP contains multiple frames. Based on the video coding dependency, the frames are divided into intra-coded I-frames, predictive coded P-frames, and bidirectional predictive B-frames. S5: The frame is divided into multiple network layer data packets for transport. Each data packet contains identification information, including GOP sequence number, frame sequence number, data packet sequence number, and frame type information. S6: Generate the final path from the source ground station to the destination ground station, and send the data packet from the source ground station to the destination ground station; S7: The destination ground station receives and rearranges the data packets according to the identification information to restore the encoded video.
2. The end-to-end live video transmission method for satellite-ground converged networks as described in claim 1, characterized in that, The process of generating the final path from the source ground station to the destination ground station specifically includes the following steps: S61: Medium-Earth orbit satellites in the management domain of the source low-Earth orbit satellite will take the source low-Earth orbit satellite as the starting point of their path and the management domain of the source low-Earth orbit satellite as their current management domain. S62: Determine the affiliation relationship between the target low-orbit satellite and the current management domain; S63: If the target low-Earth orbit satellite is located in the current management domain, then the target low-Earth orbit satellite is taken as the path endpoint in the current management domain, and an optimal intra-domain path from the path start point to the path endpoint is generated. If the target low-Earth orbit satellite is not located within the current management domain, then select an optimal dual-homed edge node within the current management domain as the path endpoint within the current management domain, and use the source low-Earth orbit satellite as the path starting point to generate the optimal intra-domain path within the current management domain. S64: Take the management domain of the other management node that manages the optimal dual-homed edge node as the new current management domain, and take the optimal dual-homed edge node as the new path starting point; S65: Repeat steps S62-S64 until the target low-Earth orbit satellite is within the current management domain; S66: Output all optimal intra-domain paths and synthesize the final path.
3. The end-to-end live video transmission method for satellite-ground converged networks as described in claim 2, characterized in that, The step of selecting an optimal dual-homed edge node from the current management domain includes: S631: Retrieve all dual-homed edge nodes in the current management domain; Select the k-th station whose geographical location or estimated arrival time is closest to the destination. edge A set of dual-homed edge nodes as candidate nodes; Based on the normalized queue load, the predicted delay of the current low-Earth orbit satellite to each dual-homed edge node, and the remaining dwell time of each dual-homed node, all dual-homed edge nodes in the candidate node set are scored. The edge node with the highest score is selected as the optimal dual-homed edge node in the current management domain.
4. The end-to-end live video transmission method for satellite-ground converged networks as described in claim 1, characterized in that, The process of generating the final path from the source ground station to the destination ground station, sending the data packet from the source ground station to the destination ground station, and also includes an I-frame packet loss handling mechanism, including: Packet loss risk prediction: Medium-Earth orbit satellites estimate the packet loss probability of the optimal intra-domain path based on information such as historical packet loss rate, current queue occupancy, link contact window, repair rate, and recent jitter. ; When the packet loss probability Greater than the threshold When the number of redundant tokens is not less than the preset value c2, a backup intra-domain path is configured for the I-frame. The backup intra-domain path does not intersect with any node of the optimal intra-domain path. When there is no path with completely non-intersecting nodes, the path with the fewest shared links is selected as the backup intra-domain path. The backup intra-domain path and the optimal intra-domain path have a transmission time slot of Δt.
5. The end-to-end live video transmission method for satellite-ground converged networks as described in claim 1, characterized in that, The generation of the optimal intra-domain path from the path's starting point to its ending point includes: A shared model is constructed, which is an Actor-Critic model, including an Actor network and a Critic network; the shared model adopts a CTDE mechanism of centralized training and decentralized execution. A candidate path set is generated based on ephemeris data and the latest link statistics. This candidate path set contains K... p Candidate paths; Using the current time as the decision time t, and based on the network environment and candidate path characteristics, construct the state vector of the medium-Earth orbit satellite at decision time t. , the state vector Input the Actor network to obtain K p Probability distribution of candidate paths ; Set constraints and a masking function for the candidate path set so that candidate paths that do not meet the constraints are not selected. If there are paths that meet the constraints, then select the appropriate path from among those paths based on the probability distribution. The final selected path P is obtained. sel If none of the paths meet the constraints, the candidate path set is regenerated until a selected path is obtained.
6. The end-to-end live video transmission method for satellite-ground converged networks as described in claim 5, characterized in that, This also includes updating the shared model; This includes establishing a segmented reward mechanism to calculate the immediate reward value obtained at decision time t. ; Batch collection The trajectory serves as the input to the model; When the cumulative number of collections reaches the preset batch size N batch Or the preset time T since the last update train The model is updated when the time is right.
7. The end-to-end live video transmission method for satellite-ground converged networks as described in claim 1, characterized in that, The source routing header of the data packet includes at least the low-Earth orbit satellite sequence, the next-hop cursor, and the backup next hop; After reading the expected next-hop satellite node from the low-Earth orbit satellite sequence, the current low-Earth orbit satellite performs a security check and effective neighbor detection on that node; If the detection passes, the data packet is forwarded. If the next hop is temporarily unavailable and the backup next hop meets the requirements of not revisiting visited nodes, contact window, and timeout, the backup next hop performs local fast rerouting to avoid sending control messages to medium-Earth orbit satellites.
8. The end-to-end live video transmission method for satellite-ground converged networks as described in claim 7, characterized in that, If the expected next hop and backup next hop are both unavailable, or the path lifespan expires, or the cursor goes out of bounds, or the security check rejects the current path, the current low-Earth orbit satellite sends a repair request to the medium-Earth orbit satellite it manages. The medium-Earth orbit satellite takes the current low-Earth orbit satellite as a new starting point, plans a new suffix from the current low-Earth orbit satellite to the end of the original path segment, and covers the list of unfinished path nodes; The number of repairs has accumulated to K. repmax Then, it triggers the reselection of the path endpoint and the optimal intra-domain path from the current management domain.
9. A live video end-to-end transmission apparatus for a satellite-ground converged network, used to implement the live video end-to-end transmission method for a satellite-ground converged network as described in any one of claims 1 to 8, characterized in that, include: The management domain partitioning module is used to divide the low-Earth orbit satellite set into multiple management domains. Each management domain consists of multiple low-Earth orbit satellites and is assigned a medium-Earth orbit satellite as a management node. The node selection module is used to select low-orbit satellites that can maintain control communication with two management nodes at the boundary of adjacent management domains as dual-homed edge nodes. The access module is used to calculate the set of visible low-Earth orbit satellites in real time for the source ground station and the destination ground station, and to select and access the optimal low-Earth orbit satellites as the source low-Earth orbit satellites and the destination low-Earth orbit satellites. The task partitioning module is used by the source ground station to divide the encoded video into continuous task units according to the group of pictures (GOP). Each GOP contains multiple frames. According to the video coding dependency, the frames are divided into intra-coded I-frames, predictive coded P-frames, and bidirectional predictive B-frames. The segmentation module is used to segment frames into multiple network layer data packets for transport. Each data packet contains identification information, including GOP sequence number, frame sequence number, data packet sequence number, and frame type information. The path generation module is used to generate the final path from the source ground station to the destination ground station and send the data packet from the source ground station to the destination ground station. The receiving module is used to receive data packets at the destination ground station and rearrange them according to the identification information to recover the encoded video.
10. An electronic device comprising a memory, a processor, a RAM, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the computer program, it implements the steps of the end-to-end live video transmission method for a satellite-ground converged network as described in any one of claims 1 to 8.
