Star chain laser communication signal transmission method of space small mammal feeding equipment

Abstract

The invention discloses a star chain laser communication signal transmission method for space small mammalian feeding equipment, and relates to the technical field of data transmission of in-orbit life science loads. A prediction-scheduling-fault-tolerant-feedback integrated transmission mechanism is constructed at a terminal side: priori of a visible window and a time delay interval is obtained through ephemeris and network measurement, dynamic topological coding is performed on a satellite-gateway station in combination with a graph neural network, and a spatio-temporal context is provided for reinforcement learning scheduling; sending opportunity, fragmentation, redundancy proportion, rate and queue selection are enabled to adaptively evolve along with window and link quality; a multi-queue weighting strategy and a priority label are adopted, so that the alarm and the vital signs obtain guaranteed bandwidth and time limit guarantee when resources are insufficient; the first packet time delay of window opening is shortened through 0-RTT handshake, and effective throughput is maintained under high packet loss through cascading of FEC and ARQ.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of data transmission of in-orbit life science payloads, and in particular to a space small mammal feeding device Starlink laser communication signal transmission method. BACKGROUND

[0002] The continuous monitoring and low-latency control requirements of manned / in-orbit life science experiments on the feeding device are high, and the load types such as vital signs, environmental parameters, high-definition videos and emergency instructions coexist, while under the low-orbit operating condition, the visible window is short, the switching is frequent, and the time delay is large. The existing method mainly depends on the sending / retransmission and simple queue of fixed threshold, and lacks the prediction and perception of the orbit and window, which is easy to cause congestion and packet loss when high-priority services burst or the window changes rapidly.

[0003] In recent years, LEO constellations with optical inter-satellite links (ISL) have been widely deployed, and users can access satellites through Ku / Ka / E-band radio, and the constellation can complete backhaul through ISL / ground stations, which can provide more stable coverage and shorter end-to-end paths in high-latitude areas. However, the terminal side cannot directly control the establishment and scheduling of the inter-satellite laser link, and the end-to-end performance is still affected by the short window and the dynamic internal routing of the network, so it is necessary to build a service classification and buffer scheduling mechanism for intermittent accessibility on the terminal / payload side, and to realize reliable transmission and priority protection across windows in cooperation with standards such as DTN (BPv7).

[0004] However, in the prior art, the terminal side scheduling relies on fixed thresholds and does not combine orbit and visible window prediction, resulting in uncontrollable time delay and packet loss during window switching; there is a lack of differentiated protection and multi-path redundancy strategy for high-priority streams such as vital signs / alarm; there is a lack of DTN / storage-forwarding and adaptive redundancy coding support for intermittent connections, making it difficult to meet the continuity requirements in the polar region and high-maneuvering attitude. SUMMARY

[0005] In view of the above existing problems, the present application is proposed.

[0006] The present application provides a space small mammal feeding device Starlink laser communication signal transmission method to solve the problems of rapid change of polar region topology, slow static threshold link establishment, and key services vulnerable to time delay and interruption.

[0007] To solve the above technical problems, the present application provides the following technical solutions:

[0008] The present application provides a space small mammal feeding device Starlink laser communication signal transmission method, the feeding device includes a communication module as a user side terminal, the method is applicable to a low-orbit satellite constellation with optical inter-satellite links, and includes:

[0009] a) classifying the data generated by the feeding device into monitoring alarms, vital signs, environmental parameters, audio and video, and batch scientific data, and writing a priority label corresponding to the category in the data packet;

[0010] b) predicting the visible time window and end-to-end path delay interval of the terminal in the preset time domain according to the public ephemeris and the available link index provided by the constellation operator;

[0011] c) performing reinforcement learning scheduling on the terminal side, selecting transmission timing, fragmentation length, redundancy coding ratio, transmission rate and queue mapping based on the current and predicted link state;

[0012] d) storing and forwarding the intermittently reachable data on the terminal side according to the priority, initiating a session and completing data transmission when the window is opened;

[0013] e) triggering an alternative path or delayed transmission strategy when the window is unavailable or congestion exceeds the threshold;

[0014] The method does not require the terminal to control the acquisition, tracking and pointing of the inter-satellite laser link.

[0015] As a preferred scheme of the space small mammal feeding device star chain laser communication signal transmission method, the reinforcement learning scheduling adopts a policy gradient type algorithm, and the state space of the terminal includes:

[0016] The remaining time of the visible window, the last session round-trip delay, the packet loss rate estimate, the uplink signal-to-noise ratio estimate, the queue length and the data priority distribution; the action set includes: transmission timing, fragmentation length, redundancy coding ratio, target transmission rate and queue selection.

[0017] As a preferred scheme of the space small mammal feeding device star chain laser communication signal transmission method, model updating is performed among multiple on-orbit feeding devices using federated learning, and only the gradient or model difference is uploaded to the ground aggregation party and the updated model is issued after aggregation.

[0018] As a preferred scheme of the space small mammal feeding device star chain laser communication signal transmission method, the constellation reachable satellite and ground gateway relationship in the prediction time domain is topologically coded using a graph neural network, and the coding result is used as an auxiliary feature of the reinforcement learning scheduling.

[0019] As a preferred scheme of the space small mammal feeding device star chain laser communication signal transmission method, wherein: the priority label is bound to the scheduling strategy as a multi-queue weighted scheduling mechanism, high weight and minimum reserved bandwidth are configured for monitoring alarm and vital sign data, adaptive code rate and frame level discard strategy are used for video stream, and low priority background transmission is used for environmental parameters and batch scientific data.

[0020] As a preferred scheme of the space small mammal feeding device star chain laser communication signal transmission method, wherein: the terminal uses a session protocol based on zero round-trip time handshake at the transmission layer, combined with connection migration and path detection in an intermittent reachable environment.

[0021] As a preferred scheme of the space small mammal feeding device star chain laser communication signal transmission method, wherein: forward error correction and automatic retransmission cascade coding are applied to data fragments at the terminal side, and the redundancy coding ratio and retransmission threshold are adaptively determined by reinforcement learning scheduling according to link quality and service priority.

[0022] As a preferred scheme of the space small mammal feeding device star chain laser communication signal transmission method, wherein: the terminal has multi-path capability, and when the constellation path is unavailable or the estimated delay exceeds the threshold, switches or uses the task mother ship relay, ground direct connection backup link or on-orbit storage to wait for the next window in parallel, and the switching trigger condition and switching strategy are given by the scheduling decision.

[0023] As a preferred scheme of the space small mammal feeding device star chain laser communication signal transmission method, wherein: the ground aggregation party performs de-duplication reorganization, out-of-order recovery and checking according to priority at the receiving end, combines the out-of-order and repeated packets caused by intermittent reachability, and feeds the end-to-end delay and reachability rate of the key alarm stream to the terminal as an online learning signal.

[0024] As a preferred scheme of the space small mammal feeding device star chain laser communication signal transmission method, wherein: end-to-end security is realized by session key rolling and data packet level integrity checking, and the key rolling timing and replay protection window are aligned with the visible time window.

[0025] The application has the advantages that: under the premise of not controlling the inter-satellite laser link, the application constructs a prediction-scheduling-fault-tolerant-feedback integrated transmission mechanism on the terminal side: the prior knowledge of the visibility window and the time delay interval is obtained through ephemeris and network measurement, the satellite-gateway station dynamic topology is encoded by combining the graph neural network, the space-time context is provided to the reinforcement learning scheduling, the transmission opportunity, fragmentation, redundancy ratio, rate and queue selection are adapted to evolve with the window and link quality, the multi-queue weighted strategy and priority label are used to ensure the minimum bandwidth and time limit guarantee of the alarm and vital signs in the case of resource shortage, the 0-RTT handshake is used to shorten the first packet delay of the window opening, the FEC and ARQ are cascaded to maintain the effective throughput under high packet loss, when the constellation path is unavailable or the jitter exceeds the threshold, the task mother ship relay, ground backup chain or on-orbit storage is automatically switched to wait for the next window to improve the service continuity, the federal learning aggregates the strategy among multiple devices to reduce the on-orbit training bandwidth and privacy risk, the deduplication and closed-loop feedback of the receiving end further stabilize the end-to-end delay and reachability, and the security side aligns the window by using the key rolling and packet-level verification to ensure the data integrity and confidentiality in the intermittent connection scenario. BRIEF DESCRIPTION OF DRAWINGS

[0026] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the following will briefly introduce the drawings needed to be used in the embodiments. It should be understood that the following drawings only show some embodiments of the present application, and therefore should not be regarded as a limitation on the scope of the present application.

[0027] Figure 1 The flowchart of the star chain laser communication signal transmission method of the space small mammal feeding device in the embodiments. DETAILED DESCRIPTION

[0028] In order to make the purpose, technical solutions and advantages of the present application more clear, the following will further illustrate the present application by combining with the drawings and embodiments. It should be understood that the specific embodiments described here are only used to explain the present application, and should not be regarded as a limitation on the present application.

[0029] All terms used in the present application (including technical and scientific terms) have the meanings generally understood by those skilled in the art, unless otherwise defined. It should be noted that the terms used herein should be interpreted to have meanings consistent with the context of the present specification, and should not be interpreted in an idealized or overly formal manner.

[0030] For example, the terms "first", "second" and the like used in the present application are only used to distinguish similar objects, to distinguish the first object from another object, and are not used to describe a specific order or sequence, nor can be understood as indicating or implying relative importance.

[0031] The so-called star chain laser communication refers to the backhaul and routing capability provided by a low-orbit satellite constellation with optical inter-satellite links, so as to realize data transmission between a user terminal and the ground / relay, such as a commercial constellation like Starlink. Unless otherwise explicitly stated, the present application does not involve direct control of the acquisition, tracking, pointing, transmission power or routing rules of the inter-satellite laser link.

[0032] The present application is directed to terminal-side transmission adaptation in a star chain laser communication environment: using window prediction, priority scheduling, intermittent reachability buffer and redundant coding, etc., to ensure the latency and continuity of critical flows such as alarms / vital signs; the present application does not require the terminal to control the establishment and pointing of the inter-satellite laser link.

[0033] The present application proposes a star chain laser communication signal transmission method for a small space mammal feeding device, as shown in Figure 1 The method comprises the following steps:

[0034] The feeding device comprises a communication module as a user terminal, and the method is applicable to a low-orbit satellite constellation with optical inter-satellite links. The method comprises the following steps:

[0035] a) classifying the data generated by the feeding device into monitoring alarms, vital signs, environmental parameters, audio / video and batch scientific data, and writing a priority label corresponding to the category in the data packet;

[0036] b) predicting the visible time window and end-to-end path delay interval of the terminal in a preset time domain according to the public ephemeris and the available link indicators provided by the constellation operator;

[0037] c) performing reinforcement learning scheduling at the terminal side, selecting the transmission time, fragmentation length, redundancy coding ratio, transmission rate and queue mapping based on the current and predicted link state;

[0038] d) caching and forwarding the data intermittently according to the priority at the terminal side, initiating a session and completing data transmission when the window is opened;

[0039] e) triggering an alternative path or a delayed transmission strategy when the window is not available or the congestion threshold is exceeded;

[0040] Wherein, the method does not require the terminal to control the acquisition, tracking and pointing of the inter-satellite laser link;

[0041] In this embodiment, the communication module as a user-side terminal includes a radio frequency unit for interacting with low-orbit constellation user downlink / uplink and an on-orbit computing unit. The visible time window refers to the continuous reachable period when the terminal meets the minimum elevation angle and link budget threshold, which is obtained based on the public orbital elements and the online measured link quality estimation. The priority label is the fixed enumeration mapping of five types of alarm, vital signs, audio and video, environmental parameters, and batch scientific data. The alternative path refers to the path selection of the task mother ship relay, ground direct backup, or on-orbit storage waiting for the next window. The ground aggregation party is the receiving and processing end of the task control center or ground station. The minimum determination elevation angle of the visible window is 10° by default, and it can be adjusted to 5°-20°. The link budget threshold is above the uplink signal-to-noise ratio or packet success rate conversion, which corresponds to the equivalent point of 10% packet loss by default, and can be adjusted to 5%-20%. The setting basis is the tolerance of the task to the key business and the on-orbit test experience. The congestion threshold is defined as one of the following conditions: the end-to-end delay exceeds the upper limit of the prediction interval by 20%, or the packet loss rate exceeds 10%. The adjustment range is 10%-40% and 5%-20%. Optionally, when the operator does not provide link indicators, the previous window historical measurement and current probe packet statistics are used instead. If necessary, when the orbital elements are temporarily missing, the latest valid elements are extrapolated for no more than one window, and the prediction bias is corrected by real-time measurement after the window is opened.

[0042] In one embodiment, the reinforcement learning scheduling adopts a policy gradient type algorithm, which includes the following in the state space of the terminal:

[0043] The visible window remaining time, the latest session round-trip delay, the packet loss rate estimation, the uplink signal-to-noise ratio estimation, the queue length, and the data priority distribution; the action set includes: sending opportunity, fragmentation length, redundancy coding ratio, target sending rate, and queue selection;

[0044] Specifically, the state variables are obtained as follows: the remaining window time is obtained by subtracting the current time from the predicted window end time; the round-trip delay is estimated by the time difference between the handshake message or data packet acknowledgment; the packet loss rate is the failure rate of the most recent valid transmitted samples; the uplink signal-to-noise ratio is mapped from the physical layer estimated reading; and the queue length and priority distribution are provided by the terminal's transmission queue statistics. The state refresh cycle is 200 milliseconds by default and adjustable from 100 to 500 milliseconds; the action update cycle is synchronized with the state refresh or is an integer multiple thereof not exceeding 1 second; the fragment length is 1024 bytes by default and adjustable from 256 to 4096 bytes; the target transmission rate upper limit is 90% of the estimated available uplink bandwidth by default and adjustable from 70% to 95%; the adjustable ranges of redundancy ratio and queue selection are constrained by the predetermined intervals and label mappings in the aforementioned embodiments. The round-trip delay statistics and packet loss rate are consistent with the quantile scaling in the aforementioned embodiments, using the sliding window caliber. For example, when computing and storage resources are limited, the action update cycle can be set to 500 milliseconds to reduce inference load; if the physical layer does not provide signal-to-noise ratio estimation, the link quality can be approximated by a monotonic mapping of the recent packet loss rate.

[0045] The reward function for reinforcement learning scheduling is defined as follows:

[0046] Step c1: The end-to-end latency, packet loss rate, jitter, and energy consumption per unit bit are statistically analyzed using a sliding time window, and the service priority is read from the queue label. This statistical analysis serves as the input for subsequent normalization and reward.

[0047] Step c2 involves scaling and truncation at each decision point to ensure comparability across different flight segments and within a window:

[0048] ,

[0049] in, The normalized latency, jitter, and power consumption per bit are: For observation purposes, and For the first and Sub-anchor points, The scaling matrix, and These are vectors composed of all 1s and all 0s, respectively. , According to the function of elements, Indicates the current decision-making moment;

[0050] Step c3, classify the business categories Priority labels are mapped to scheduling weight ranges within and put As a contextual feature, it participates in strategy evaluation and learning;

[0051] Step c4 generates an immediate reward after each fragmentation / send decision to encourage low latency, low packet loss, low jitter, and low power consumption, and deducts for overdue transactions.

[0052] ,

[0053] in, For a moment Instant rewards For category The weight, Weighting each indicator and , The normalized value obtained in step c2, For packet loss rate, For overdue amount, For indicator functions, For category The time limit threshold Indicates the business category superscript. Let be the end-to-end delay at time t;

[0054] Step c5 employs a dual-threshold criterion of stagnant sliding return gain and policy divergence constraint:

[0055] ,

[0056] in, The length of the sliding window (in terms of the visible window). For the first Average instant reward of the next iteration To reward the convergence threshold, For parameters The distribution of strategies The relative entropy is a strategy that uses state-based access distribution weighting. For the policy stability threshold, To train iterative indices, This means that both of the above inequalities are satisfied simultaneously;

[0057] Here is a suggestion , , , , Use 3–5 visible windows; temporarily relax this limit when there is a link mutation. And reduce the learning rate, then revoke the threshold after stabilization;

[0058] Specifically, firstly, scale alignment is achieved for different flight segments and windows using fractional scaling; then, latency, packet loss, jitter, and energy consumption are comprehensively considered using a weighted negative cost approach; finally, the time-limited attributes of critical services are highlighted using overdue quotas; priority is reflected through weighted mapping to show service differences, ensuring that alarms and vital signs remain dominant even when the link is strained; during the training phase, sliding reward gain and policy divergence are used as dual thresholds for shutdown and stabilization, making it easier to maintain the controllability of the learning process in scenarios with intermittent accessibility and rapid topology changes;

[0059] For example, jitter can be calculated as the median of the absolute value of the first difference of the end-to-end delay sequence within the same sliding time window. Energy consumption per unit bit is obtained by dividing the power consumption metering of the communication module by the number of effective data bits within the same window. Quantity anchors are estimated on the union of the same window or adjacent windows to improve robustness. The sliding time window defaults to covering 30-120 seconds or a complete visible window (whichever is smaller), and the minimum number of samples for quantile estimation defaults to 50. When samples are insufficient, an exponentially weighted average is used instead of quantile updates to avoid instability. The priority weight mapping interval and the overdue limit range are based on the recommendations of the aforementioned embodiments. The policy divergence threshold and reward convergence threshold are selected from the middle of the recommended interval according to the stability requirements under intermittently reachable environments. Optionally, if energy consumption metering is temporarily unavailable, the energy consumption per unit bit is approximated by dividing the product of the transmit duty cycle and rated power consumption by the number of data bits. If necessary, the policy is frozen in advance under the boundary condition that either of the two thresholds is not met and the window will close, and the previous stable policy is used until the next window.

[0060] In one embodiment, federated learning is used for model updates among multiple on-orbit feeding devices. Only gradients or model differences are uploaded to the ground aggregator, and the updated model is distributed after aggregation, so as to reduce inter-satellite bandwidth consumption and protect mission data privacy.

[0061] Similarly, federated aggregation uses the window end event as its natural beat, triggering an aggregation round by default every 1-3 visible windows. Uploaded content is limited to compressed gradients or model differences along with necessary training statistics to support robust aggregation. The upload size per round defaults to a few megabytes and can be adjusted based on task throughput. The local training cycle on the terminal side defaults to covering the data from the most recent window. The aggregation timeout defaults to no more than one window duration. Aggregation weights are set according to the proportion of effective samples from each terminal or its square root. Optionally, when critical business operations within a window are heavily utilized, resulting in insufficient uplink throughput, the current upload round can be skipped and merged in the next round. If necessary, when multiple upload failures occur, the terminal maintains its existing strategy and records the failed rounds for subsequent compensatory aggregation.

[0062] In one embodiment, a graph neural network is used to perform topological coding on the relationship between constellation reachable satellites and ground gateway stations in the predicted time domain. The coding result is used as an auxiliary feature for reinforcement learning scheduling to improve decision stability when the topology changes rapidly.

[0063] Optionally, the node characteristics of the topology coding include the start and end times of the window for each reachable satellite in the prediction time domain, the visibility elevation angle, the estimated round-trip time, and the available bandwidth range. An edge is defined as a simultaneous visibility or transferable relationship existing at a given time. Each update retains only the top 3-6 nodes sorted by visibility and link quality to limit inference overhead; the topology refresh cycle defaults to being consistent with the window prediction cycle and does not exceed 1 second. Auxiliary features are incorporated after being dimensionally consistent with the state space in the aforementioned embodiments, without changing the original state definitions. If necessary, when the predictor temporarily fails to perform accurately or the input is missing, it degenerates to using the topology coding from the previous time period or only using scalar features, without interrupting the scheduling decision.

[0064] In one embodiment, priority labels and scheduling strategies are bound to a multi-queue weighted scheduling mechanism, which configures high weight and minimum reserved bandwidth for monitoring alarms and vital signs data, and adopts an adaptive bitrate and frame-level discarding strategy for video streams, and uses low-priority background transmission for environmental parameters and batch scientific data.

[0065] Furthermore, multi-queue weighted scheduling is implemented using a triplet of weight, reserved bandwidth, and preemption order, with a default of five levels: high, relatively high, medium, relatively low, and low. Alarms and vital signs correspond to the two highest levels and are configured with a minimum reserved bandwidth. The minimum reserved bandwidth defaults to 10% and 5% of the estimated available uplink bandwidth, respectively, and is adjustable from 5% to 15% and 3% to 10%. The adaptive bitrate adjustment step size defaults to ±10% of the bitrate and is adjustable from 5% to 20%. Frame-level dropping prioritizes dropping low-importance frames based on frame importance to control congestion. The bandwidth estimation and priority weights are consistent with the aforementioned embodiments. Optionally, the scheduler can operate under either a weighted fair or strict priority strategy with consistent input and output; when computational resources are limited, the weights are fixed and only the minimum reserved bandwidth and dropping strategy are executed to ensure critical services.

[0066] In one embodiment, the terminal adopts a session protocol based on zero round-trip time handshake at the transport layer, combined with connection migration and path detection in intermittent reachable environments, in order to shorten the first packet latency when the window is opened and reduce reconnection overhead.

[0067] In this embodiment, the session protocol caches reusable tickets before the window closes and prioritizes 0-RTT recovery in the next window. The session identifier remains unchanged during satellite handover to enable connection migration. Path probing uses low-overhead probe messages to update available bandwidth and round-trip time estimates. The ticket validity period defaults to covering 1-3 orbital periods and does not cross the mission day boundary. The path probing interval defaults to 200-500 milliseconds and adaptively increases or decreases according to congestion levels. Replay protection and key rolling are consistent with the security policies of the aforementioned embodiments. Optionally, when 0-RTT verification fails or the server policy does not allow it, it automatically degenerates to a 1-RTT handshake; if necessary, 0-RTT is temporarily disabled to reduce replay risk when the terminal clock drift exceeds the allowable range.

[0068] In one embodiment, concatenated coding of forward error correction and automatic retransmission is applied to data fragments at the end, and the redundancy coding ratio and retransmission threshold are adaptively determined by reinforcement learning scheduling based on link quality and service priority.

[0069] The adaptive calculation steps for the redundancy coding ratio and retransmission threshold include:

[0070] Step d1: The terminal calculates the packet loss rate estimate at each decision time. The remaining time of the current visible window The weight of the queue's category Planned transmission rate (bit / s), slice length (bit) and FEC data fragment number (Number of data fragments per FEC block); these quantities are used for subsequent calculations. and Input;

[0071] Step d2: Increase redundancy when the window becomes tight or packet loss increases, and adjust it according to the business weight:

[0072] ,

[0073] in, For a moment category The redundancy coding ratio (the ratio of the number of odd and even fragments to the number of data fragments). This indicates that the scalar Cut off to interval , For packet loss sensitivity, For category weights, For packet loss rate estimation, To avoid robust terms with excessively small denominators, As a weight for window urgency, The degree of urgency (increases as the remaining time decreases). and These represent the redundancy ratio at the lower / upper limits, respectively. Indicates the time of decision-making. Superscript indicates business category. Indicates the remainder;

[0074] Step d3 embeds the single success probability after FEC into the threshold design, so as to give a minimum retransmission upper limit under a given reliability objective:

[0075] ,

[0076] in, For category At any moment The maximum number of retransmissions. This represents the maximum allowed retransmission limit for this category. To round up, For category The minimum success probability requirement at the block level It is the cumulative distribution function of the standard normal distribution (used as a normal approximation of the binomial tail). The number of data fragments contained in each FEC block, used to characterize the impact of block length on decoding success rate;

[0077] Step d4: Within the visible window budget, FEC transmission and worst-case retransmission must not time out.

[0078] ,

[0079] The first term in the numerator is the encoding and transmission time of one FEC block, and the second term is the extra time for worst-case retransmission of each block. For the current planned sending rate, This is the fragment length; when this inequality is not satisfied, the fragment length is adjusted downwards first according to priority. (Low-priority queues shrink first), then decrease. (Reduce the time required for a single transmission), and increase it again. Use redundant methods to replace transmissions until the budget is met or an alternative path is switched;

[0080] suggestion , , , , , Follow Stratification (e.g., vital signs / alarms) ), Desirable , exist Inner piecewise mapping (When packet loss is high, use smaller fragments). These represent the minimum and maximum fragment lengths, respectively. The unit for the above intervals is bits. Based on queue backlog and encoding / decoding overhead Choose from;

[0081] Specifically, the redundancy and retransmission limits are written as computable functions that directly depend on packet loss, window urgency, and service weight, thus maintaining a stable strategy under conditions of link changes and intermittent reachability. The redundancy ratio is calculated by superimposing and truncating the packet loss and urgency terms to avoid over-encoding. The retransmission threshold incorporates the single success probability after FEC into a closed-form rounding equation, facilitating online fine-tuning and compatibility with federated updates. The time budget constraint incorporates the sending rate and fragment length into the same inequality, forming a linked control path through priority-ordered backoff. When budget convergence is difficult, it can be combined with multi-path or on-orbit storage strategies to reduce dependence on a single path and maintain the reachability of critical flows.

[0082] Specifically, packet loss rate estimation employs sliding statistics within the same window, which can be superimposed with exponential weighting to smooth out bursts. Window urgency is calculated by taking the predicted remaining window time and the ratio of each category's time limit threshold. The planned transmission rate is jointly estimated by recent effective throughput and path probing results. The packet loss rate statistical window defaults to 1-3 seconds, and may be aligned with the status refresh cycle or other timeframes if necessary. The number of data fragments contained in each FEC block is selected based on a trade-off between processing capacity and latency constraints, defaulting to a range of 8-32. Reliability targets, maximum retransmission limits, and redundancy ratio upper and lower limits are configured according to the ranges already defined in the aforementioned embodiments. Optionally, when the window is about to close and critical services are not yet completed, priority is given to increasing the redundancy ratio rather than increasing the number of retransmissions to reduce tail risk. If necessary, when the estimated throughput suddenly drops and the budget inequality is continuously not satisfied, alternative path evaluation is triggered, and a one-time switchover is performed before the current window ends.

[0083] In one embodiment, the terminal has multi-path capability. When the constellation path is unavailable or the estimated latency exceeds the threshold, it switches or uses the mission mothership relay, ground direct backup link or on-orbit storage to wait for the next window in parallel. The switching trigger condition and the back-off strategy are given by the scheduling decision.

[0084] For example, multi-path selection is based on a comprehensive ranking of path health scores and service priorities. Health scores are derived from a dimensionless normalized combination of round-trip latency, packet loss, and available bandwidth estimates. To avoid frequent jitter, a hysteresis strategy is used for switching, and a minimum dwell time is set. The minimum dwell time defaults to 5-15 seconds, and the default switchback condition is that the target path's score improves by more than a set percentage relative to the current path and remains so for more than one status refresh cycle. The dimensions of the thresholds and scores are consistent with the status and reward normalization in the aforementioned embodiment. Optionally, mirroring of the highest priority service across two paths is allowed to improve success rate; if necessary, when only a single path is available, critical services are stored in-orbit and delayed until the next window for transmission, while low-priority services are suspended from queuing.

[0085] In one embodiment, the ground aggregator performs priority-based deduplication and reassembly, out-of-order recovery and verification at the receiving end, merges out-of-order and duplicate packets caused by intermittent reachability, and feeds back the end-to-end latency and reachability of key alarm streams to the terminal as online learning signals.

[0086] Similarly, the receiving end performs deduplication and reassembly based on packet sequence number and timestamp. The out-of-order recovery window adaptively adjusts according to the predicted round-trip time and its jitter margin, and periodically returns the latency distribution and reachability summarized by category to the terminal in low-overhead feedback messages. The out-of-order recovery window defaults to 2-4 times the predicted round-trip time, and the reachability statistics period defaults to covering one visible window; the deduplication buffer retention period defaults to being consistent with the statistics period. The feedback only includes necessary statistics and is consistent with the learning signal caliber of the aforementioned embodiments. Optionally, when the feedback channel is congested, the feedback frequency is reduced and key service statistics are prioritized; if necessary, suspicious packets are discarded and event counts are recorded for the security module's reference when abnormal duplication or replay is detected.

[0087] In one embodiment, end-to-end security is achieved by session key rolling and packet-level integrity verification. The timing of key rolling is aligned with the replay protection window and the visible time window to ensure secure continuity under intermittent connection conditions.

[0088] Furthermore, key rolling is completed within a few seconds before the window opens and confirmation is completed before the window closes. The replay protection window covers the most recent statistical period and is consistent with the validity period of the session ticket. The time base is provided by the onboard high-stability clock and corrected with network time when available. The advance rolling margin is 2-5 seconds by default, and the replay protection window is consistent with the feedback statistical period by default. Integrity verification is uniformly enabled for all categories. Key rolling and session recovery are consistent with the session and reward process in the aforementioned embodiments. Optionally, for the highest priority service, a brief delay in rolling to the beginning of the window is allowed under the boundary condition that the key is about to roll and the window is about to close; if necessary, when time synchronization mismatch is detected and exceeds the allowable error, 0-RTT is temporarily prohibited and a strict handshake path is switched to ensure security.

[0089] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit them. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features therein. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of this application.

[0090] Furthermore, those skilled in the art will understand that although some embodiments herein include certain features included in other embodiments but not others, combinations of features from different embodiments are meant to be within the scope of this application and form different embodiments. For example, all the embodiments above can be used in any combination. The information disclosed in this background section is intended only to enhance the understanding of the general background of this application and should not be construed as an admission or in any way implying that such information constitutes prior art known to those skilled in the art.

Claims

1. A method for transmitting Starlink laser communication signals in a space-based small mammal rearing device, wherein the rearing device includes a communication module as a user-side terminal, and the method is applicable to low-Earth orbit satellite constellations equipped with optical inter-satellite links, characterized in that... include: a) Classify the data generated by the feeding equipment into monitoring alarms, vital signs, environmental parameters, audio and video, and batch scientific data, and write priority tags corresponding to the categories into the data packets; b) Based on publicly available ephemeris and available link metrics provided by constellation operators, predict the terminal’s visible time window and end-to-end path delay range within a preset time domain. c) Perform reinforcement learning scheduling on the terminal side, and select the sending timing, fragment length, redundancy coding ratio, sending rate and queue mapping based on the current and predicted link state; d) Cache the data on the terminal side according to the stated priority and perform intermittently reachable data forwarding. Initiate a session and complete data transmission when the window opens. e) When the window is unavailable or congestion exceeds the threshold, trigger the alternative path or delayed transmission strategy; The method does not require the terminal to control the acquisition, tracking, and pointing of the inter-satellite laser link.

2. The Starlink laser communication signal transmission method for a space-based small mammal rearing device as described in claim 1, characterized in that, The reinforcement learning scheduling employs a policy gradient algorithm, and the terminal's state space includes: The remaining time of the visual window, the round-trip time of the most recent session, packet loss rate estimation, uplink signal-to-noise ratio estimation, queue length and data priority distribution; the action set includes: sending timing, fragment length, redundancy coding ratio, target sending rate and queue selection.

3. The Starlink laser communication signal transmission method for a space-based small mammal rearing device as described in claim 2, characterized in that, Federated learning is used to update the model among multiple in-orbit feeding devices. Only gradients or model differences are uploaded to the ground aggregator, and the updated model is distributed after aggregation.

4. The Starlink laser communication signal transmission method for a space-based small mammal rearing device as described in claim 3, characterized in that, A graph neural network is used to perform topological coding on the relationship between constellation reachable satellites and ground gateway stations in the predicted time domain. The coding result serves as an auxiliary feature for reinforcement learning scheduling.

5. The Starlink laser communication signal transmission method for a space-based small mammal rearing device as described in claim 4, characterized in that, The priority label and scheduling strategy are bound to a multi-queue weighted scheduling mechanism. High weight and minimum reserved bandwidth are configured for monitoring alarms and vital signs data. Adaptive bitrate and frame-level discarding strategies are adopted for video streams. Environmental parameters and batch scientific data are sent to the background with low priority.

6. The Starlink laser communication signal transmission method for a space-based small mammal rearing device as described in claim 5, characterized in that, The terminal uses a session protocol based on zero round-trip time handshake at the transport layer, combined with connection migration and path detection in intermittent reachable environments.

7. The Starlink laser communication signal transmission method for a space-based small mammal rearing device as described in claim 6, characterized in that, At the endpoint, concatenated coding of forward error correction and automatic retransmission is applied to data fragments. The redundancy coding ratio and retransmission threshold are adaptively determined by reinforcement learning scheduling based on link quality and service priority.

8. The Starlink laser communication signal transmission method for a space-based small mammal rearing device as described in claim 7, characterized in that, The terminal has multi-path capability. When the constellation path is unavailable or the estimated latency exceeds the threshold, it can switch to or use the mission mothership relay, ground direct backup link, or on-orbit storage in parallel to wait for the next window. The switching trigger conditions and back-off strategy are given by the scheduling decision.

9. The Starlink laser communication signal transmission method for a space-based small mammal rearing device as described in claim 8, characterized in that, The ground aggregator performs priority-based deduplication and reassembly, out-of-order recovery, and verification at the receiving end. It merges out-of-order and duplicate packets caused by intermittent reachability and feeds back the end-to-end latency and reachability of key alarm streams to the terminal as online learning signals.

10. The Starlink laser communication signal transmission method for a space-based small mammal rearing device as described in claim 9, characterized in that, End-to-end security is achieved through session key rolling and packet-level integrity verification, with the key rolling timing and replay protection window aligned with the visible time window.

Images