AI-BASED STORE-AND-FORWARD METHOD FOR NTN NETWORKS

Abstract

An operating method for a non-terrestrial network (10) satellite (100) in store-and-forward mode is described. Data packets are received from user equipment (200); packet attributes (priority, size, due date) are extracted; a prioritization model (111) assigns a priority score. Buffer space is allocated in a hierarchical buffer (120) according to the scores. A feedline prediction model (112) predicts the time and duration of feedline availability from historical link data, satellite position, and environmental factors. A pre-trained AI engine (110) determines which of the forward, store, or drop operations to perform for each packet, based on buffer status and packet information. A scheduler (130) defines transmission windows and forwards high-priority packets first to a ground station (300), followed by delay-tolerant packets.Coordinated prioritization, prediction, and action selection improve delivery reliability under intermittent connectivity while reducing buffer overflows, missed deadlines, and latency. Figure 1

Description

1 TARIFF AI-BASED STORE-AND-FORWARD METHOD FOR NTN NETWORKS TECHNICAL FIELD The invention involves a method of operating a satellite belonging to a non-terrestrial network for storage and communication. It is related to the method of receiving data packets from user equipment, and the data packets. temporarily storing the data in a buffer and releasing it when a feeder line becomes available. This involves transmitting packets to a ground station. PREVIOUS TECHNIQUE Non-terrestrial networks (NTNs) connect users and devices via satellites. It is increasingly used for connecting. In many existing applications, uninterrupted communication from a satellite to the ground, either directly or via other satellites. It is assumed that the service follows a certain path. When this assumption fails, service quality rapidly deteriorates. It falls. The following overview shows that modern NTN communications typically use a store-and-forward method. (S&F) how it works without it and why S&F is designed to shut it down He explains that it creates gaps. In the simplest case, a satellite provides a direct feedline to a ground station and User equipment (UE) transmits traffic to the ground in real time. This includes audio or video. Although it supports interactive services, it is fragile: it can go out of the satellite's field of view or be affected by terrain. or the supply line was temporarily interrupted due to buildings blocking the road or interference. If it becomes unusable, the data will not be delivered. As a result, 25 Sessions are interrupted and information is lost. A second approach relies on inter-satellite links (ISL), where a satellite transmits data. Currently, traffic is routed through one or more satellites until it reaches a satellite with a feeder line. It transmits through multiple neighboring satellites. This reduces reliance on a single satellite for ground connectivity. 30 However, for ISL chains to be effective, they need to remain continuously available. It is necessary. In sparse constellations or during intercellular transitions, a continuous line It may not be available, and long or unstable chains can cause additional delays and packet loss. It is possible. 35 2 Some systems operate in transparent or "twisted-line transmission" mode. These satellites transmit data. They neither store nor process; they simply transmit what they receive to its destination. All routing and processing is done by them. It remains on Earth. This architecture is simple, but it's entirely focused on being right next to a supply line. It depends on its existence. If this connection is broken, transmission stops. Also, It is also inefficient for delay-tolerant traffic, because the satellite is later to be delivered. 5 It cannot store data in the buffer. Some NTN systems use positioning and timing for global navigation satellites. It relies on the GNSS (Global Signal System). In such cases, UEs (Universal Electronics) use ground or satellite systems to manage traffic. It sends GNSS-based location data used by the network. This is from 10 GNSS receivers. It excludes devices lacking this capability and limits the flexibility of operation independent of GNSS. Data Without S&F to separate the transmission from the continuous positioning input, these devices are active. cannot participate in any way. Real-time services such as video calling and gaming are typically available on satellite networks (15). This is a priority. Service quality controls minimize latency for these applications. It dynamically allocates bandwidth to maintain capacity. In the absence of S&F, this Prioritization requires very little effort to process delay-tolerant traffic in an orderly manner. They can monopolize resources by releasing capacity or not releasing any at all. Non-essential data. It is removed from the priority list, which leads to inefficient use of spectrum and power, and sometimes to 20 This results in the complete loss of data. These models describe systems lacking S&F, which can experience interruptions in connectivity or resource scarcity. It explains why it's difficult to share fairly when it should be shared. S&F Without them, networks only operate continuously when feeder lines or ISLs are available. It can support communication; but only seamlessly under a narrow set of conditions. It works but is inadequate at managing workloads that are tolerant of delay. Store and transmit, buffering data when satellites have no connection to the ground. By allowing it to receive and transmit this data when the connection is restored, these shortcomings are addressed. 30 It resolves. Real-time, high-priority traffic can be sent immediately, while lower-priority traffic can be sent quickly. Priority, delay-tolerant data is stored. This eliminates the need for continuous end-to-end connectivity. It reduces bandwidth usage and increases overall efficiency in energy consumption. 3 A particular problem that arises in NTN systems, especially in 6G scenarios, is power supply. Maintaining reliable communication during line disruptions. Orbital dynamics, Geographical or atmospheric conditions can intermittently disrupt the path between a satellite and its ground station. It can interrupt communication. During these periods, end-to-end communication is only possible via real-time relay. This cannot be achieved. When disaster relief is evaluated: a satellite emergency 5 over the disaster area. It collects status alerts and sensor readings, but the nearest ground station is also affected. and goes offline temporarily. Service continuity fails without S&F. With S&F satellite, without user intervention and without additional ground infrastructure or permanent alternative connections. By delivering the data locally without the cost and complexity of setup, the incoming data is transferred locally. It stores and transmits when the supply line is restored. 10 A second problem is remote areas like polar regions and oceans where geostationary connectivity is inconsistent. and emerges in isolated regions. Maritime IoT exemplifies this: a ship, existing It could be in the line of sight of a satellite lacking a feed line. S&F, the satellite's position near the ship. accepting and storing operational data and then connection 15 by enabling it to transmit information when it is provided, such as monitoring and maintenance scheduling. It ensures that delay-tolerant functions continue uninterrupted despite long gaps. A third challenge involves multi-hop satellite communications in sparse networks. Some satellites, The ground may not have a direct path—either directly or indirectly. Wilderness 20 Monitoring its life is one example: when sensors pass over the hills of a satellite It transmits under suitable conditions, but currently there is neither a supply line nor a suitable one. It could be an ISL route. S&F ensures that the collected data is preserved throughout the transition between cells. and by ensuring that it is transmitted when a route becomes available, during satellite transits. It prevents losses. 25 Finally, many applications require final delivery, not immediate results. Environmental and agricultural. Monitoring is typical: soil moisture sensors in remote locations are currently monitoring the ground. It can reach a satellite that is otherwise inaccessible. S&F holds the data until a suitable link is found. while guaranteeing delivery, it also eliminates unnecessary interference in a real-time relay. By avoiding these factors, it preserves the spectrum and power. In short, S&F directly addresses situations where real-time relays are impractical: the power supply. line interruptions, infrequent or changing network routes, remote geographies, and tolerance for latency. workloads. S&F, which separates data capture from real-time primary carrier availability, is the final delivery 35 4 It provides an efficient and reliable mechanism for this. In these scenarios, no other approach is suitable. It does not offer the same combination of robustness, resource efficiency, and ease of implementation, which This makes S&F a core capability for NTN in the 6G era. S&F delivers over interrupted connections by holding onto data until the connection is restored. While this makes it possible, current practices reveal three recurring problem areas. Firstly, satellite storage space is limited; under heavy traffic or prolonged outages Buffers fill up and force a drop or aggressive interrupt. The latter involves a lot of system traffic. It roughly prioritizes, therefore mission-critical packets are determined by conditions. When it changes, it can ineffectively compete with routine telemetry. Thirdly, delayed 10 delivery, especially when transmitted in large batches over restricted connections, This can further increase delays and energy consumption. Methods from the previous technique alleviate certain aspects of this situation, but it is fragmented. It continues to be so. Dynamic programming algorithms allocate resources in a realistic way. It adjusts according to timely traffic and supply line status, and in some cases, the packet ends. It takes into account the dates. These approaches normally consider data volume and termination. It improves latency, but typically only works in the scheduler and buffer. acceptance / discharge policies or any connection availability estimation model It is not coordinated. Priority-sensitive buffer management schemes organize traffic into classes or 20 It divides into queues and reduces loss for high-priority packets during congestion. It offers basic QoS guarantees. However, priorities are generally static; aging, Without reclassification or foresight regarding future connectivity statuses, As conditions evolve, queues may become empty or overfilled. Machine learning based programmers—using supervised or reinforcement learning—for the past 25 It is adapted from models and can handle complex operating regimes. In practice, They generally focus on a single function (e.g., programming) and buffer management. By leaving prioritization to fixed rules, they limit end-to-end gains. Bundle Latency-tolerant network protocols, such as the protocol, are by design to handle interruptions well. It is manageable and widely used in environments with long or frequent interruptions, but 30 Detailed prioritization, QoS, or when and how backlog data is transmitted. It provides limited local mechanisms for proactive, predictive control. These limitations share a common root: the control plane is separate and highly reactive. It is divided into modules. Programmers are based on the current load, buffers on instantaneous occupancy, and 35 The protocol logic also reacts later to connection availability. As a result, Current solutions preserve QoS across changing traffic mixes and connection patterns. It does not adequately support an integrated, adaptable system. Accordingly, prioritization involves predictive modeling of supply line availability and 5 a unified and scalable framework that addresses hierarchical buffer management together It is necessary. This type of approach coordinates end-to-end adoption, queuing, and transmission decisions. This way, critical data is protected, storage is used economically, and deferred payments are avoided. Transmissions minimize delay and energy overhead when the connection is re-established. All the problems mentioned above ultimately necessitate an innovation in the relevant technical field. This has made it necessary to do so. A BRIEF DESCRIPTION OF THE INVENTION The present invention will eliminate the aforementioned disadvantages and will contribute to the relevant technical field. It is related to a method that will provide new advantages. The aim of the invention is to provide dynamic solutions for highly variable conditions in non-terrestrial (NTN) networks. The aim is to provide a method that adapts accordingly. 20 Another aim of the invention is to reduce the satellite communication load by minimizing unnecessary transmissions. To prolong life. Another aim of the invention is to optimize buffer usage in satellites, thereby improving QoS 25. to provide. All the objectives mentioned above and those that will emerge from the detailed explanation below To achieve this, invent, store, and transmit, operate a non-terrestrial network satellite. It is related to the method, the method involves receiving data packets from the user equipment, data 30 the packets are temporarily stored in a buffer and a feed line is available This involves transmitting data packets to a ground station when necessary. Accordingly, the method It also includes the following: 6 - a package consisting of nominal priority, size, and due date for each package extraction of its characteristics; - a priority score assigned to each package by a prioritization model appointment; - A hierarchical buffer with priority scores and multiple priority levels 5 Buffer space allocation for incoming packets based on their structure; - historical connection data from satellites by a feedline prediction model the location and environmental factors that make the supply line available The use of time and a period of time; - by a pre-trained AI engine, buffer state and packet 10 Based on the information, a transmission action and a storage action are performed for each data packet. the action and the selection of an act of letting go; - determining transmission windows from prediction and high-priority packets Packets within windows to be delivered before delay-tolerant packets. Programming of transmissions. The invention provides static queuing for non-terrestrial networks. 15 and replacing rule-based programming with a unified, learning-oriented framework. It provides an adaptive store-and-forward method that changes packet prioritization. A system that handles programming and buffer management together, with intermittent feeding. It responds to lines, fluctuating traffic, and limited on-site resources. Machine Learning-based prioritization, packages by size, due date and nominal priority 20 predictive modeling, while ranking in real-time from metadata such as these, Estimating supply line availability to proactively plan transmissions It does so in a hierarchical manner where low-priority or older packets are preventively dropped. A buffer prevents overflow and conserves space for urgent data. Status reporting, location and provides transparent delivery estimates to device endpoints. 25 This provides tangible advantages: Critical data is accessible first, even under limited storage conditions. They are delivered first; avoiding useless transmissions and grouping them when useful. Energy usage is reduced; delay time for high-priority traffic is reduced; better results are achieved. Service quality improves thanks to historical accuracy and data integrity; and the approach differs by 30 Scaling across constellations and traffic regimes, disruptions and variability. It maintains robust performance under load. A possible arrangement of the invention would be through a control interface to ground stations or It is characterized by the transmission status to the devices and the provision of the expected delivery time. 35 7 Another possible arrangement of the invention is that a pre-trained AI engine uses synthetic traffic tracks. and / or by obtaining it through reinforcement learning using historical satellite records It is characterized, and the motor is trained with: (i) buffer occupancy relative to Bmax and The priority distribution consists of nominal priority (πᵢ), dimension (sᵢ), deadline (tᵢ), and remaining time (tᵢ-t), forming a 5 Package including supply line availability indicator and historical performance metrics. (i) a state containing attributes at the beginning (Sₜ); (ii) a set of actions including transmit, store and release; and (iii) weighted around target occupancy (β) by packet priority and buffer stability Increased on-time delivery and missed deadlines, buffer overflow, and higher A reward that imposes penalties for downgrading priority packages, reduced cumulative 10 A policy optimized to maximize the reward. Another possible arrangement of the invention is that the act of releasing occurs when the retention thresholds are exceeded. or has a priority lower than a priority threshold when it is predicted to be exceeded It is characterized by involving the dropping off of packages. 15 BRIEF DESCRIPTION OF THE FIGURES Figure 1 is a diagram showing a schematic view of the system. Figure 2 is a diagram showing the satellite components that made the invention possible. Figure 3 shows the simulated buffer level during a 12-hour store-and-forward operation. It is a graph that shows the way. REFERENCE NUMBERS GIVEN IN THE FIGURES NTN network 100 Satellites 110 AI engines 30 111 Prioritization model 112 Feedline prediction model 120 Buffer 121 Buffer management system 130 Programmers 35 8 200 User equipment 300 Ground Stations DETAILED DESCRIPTION OF THE INVENTION This detailed explanation is intended solely to make the subject more understandable. This is explained with reference to examples without creating any limiting effects. Figure 1 shows satellites (100), numerous user equipment devices (200) and at least one ground Shows an example non-terrestrial network (NTN) (10) containing stations (300). Each satellite (100) 10 via a service connection with user equipment (200) and via a supply line It is arranged to communicate with the ground station (300). In some arrangements, two an inter-satellite link for the exchange of data or control information between satellites (100) It is available. Referring to Figure 2, each satellite (100) is designed to perform store and forward functions. is configured. The configuration is a prioritization model (111), a feedline estimation model (112), an AI engine (110), a buffer (120), a buffer management system (121) includes a programmer (130) and a control interface (not shown). Prioritization model (111) assigns nominal priority, package size, due date and for each packet Optionally, prioritize one of the features such as package age or measured channel quality. The feed line prediction model (112) is configured to calculate the score. inputs that may include connectivity status, satellite position, and environmental indicators using this to estimate future supply line availability and the estimated 25 It is configured to exclude availability features or delivery windows. The AI ​​engine (110) takes priority scores from the prioritization model (111) as input, Feedline availability estimated from the feedline estimation model (112) and The packet 30 retrieves the current buffer state and selects between forward, store, or drop. It is a decision module that provides an action instruction at the beginning. Buffer (120) is a finite built-in buffer for packets held during store-and-forward operations. It provides storage. The buffer management system (121) manages buffer (120) usage. A hierarchical buffer structure in which packets are classified according to their priority levels 35 9 It implements and enforces preventive release, for example, when retention thresholds are exceeded. In particular, intermediate The memory management system (121) responds to the action directives given by the AI ​​engine (110). configured to preferentially remove lower priority or older packages. It is done. The programmer (130) feeds the action directives from the AI ​​engine (110). Using the availability estimated from the line prediction model (112) for the supply line They prepare transmission plans and account for latency, buffer usage, and meeting deadlines. It can implement balancing policies. Control interface with ground station (300). It supports signaling and in some configurations queue levels, expected delivery 10 indicators and ground systems or selected user equipment (200) It provides status information such as configuration or versioning data. Ground station (300) provides gateway functions toward a terrestrial core or cloud network. In some arrangements, ground station (300) prioritization model (111) and feeder line 15 Distributes ephemeris, configuration data and updates for the prediction model (112) and It collects operational records suitable for offline analysis or retraining. Satellites (100) in low Earth orbit, medium Earth orbit or geostationary It can be implemented on various platforms. Examples include 20 electronically operated multi-beam systems. small satellite data buses equipped with steerable phased array antennas or high Includes larger platforms with high-gain antennas. Processing resources. prioritization model (111), feedline prediction model (112) and AI engine (110) may include general purpose processors to run. Platform power and thermal Restrictions may affect supply line configurations. 25 User equipment (200), handheld terminals with NTN-enabled modems, environmental monitoring, Fixed or mobile IoT nodes used for agriculture or public services, on ships and marine terminals on buoys, vehicle telematics units and unmanned aerial vehicles, etc. This includes devices capable of establishing a direct NTN link, such as airborne platforms. Capabilities 30 among them omnidirectional or directional antennas, for narrowband or broadband waveforms. Optional GNSS receivers for support and location and time may be available. Some users equipment (200) operating under tight energy budgets with intermittent transmission capability, Others offer higher data rates and continuous performance depending on antenna gain and available power. It supports sessions. 35 The method in question is carried out using the system mentioned above. Operation The process is as follows: Packets from user equipment (200) are sent to a satellite (100) Upon arrival, the system retrieves the metadata including nominal priority (π), dimension (s), and due date (t). extracts the data. The prioritization model (111) calculates a priority score and assigns 5 to each package. It issues a priority code (bit sequence). In parallel, the feed line prediction model (112) towards the ground station (300) It continuously estimates supply line availability and the start time and duration. It generates estimated connection windows with the AI ​​engine (110), priority outputs from (111), 10 It retrieves the estimated windows and current buffer status from (112) and A single action instruction per packet to be executed by the programmer (130) (forward, store or (leave) gives. The buffer management system uses instructions received from the AI ​​engine (110). (121) buffers packets according to a hierarchical structure with multiple levels (120) accepts; if retention thresholds are approached, lower priority or 15 Aged packets are pre-selected to conserve space for higher-priority traffic. It is left behind. The programmer (130) determines the transmission windows from these predictions and the AI ​​engine (110) According to their outputs, they prioritize high-priority packages and those close to their deadlines first, followed by 20 When capacity allows, it prepares a transmission plan that prioritizes delay-tolerant packets. Throughout the study, the control interface queue levels, transmission results, and estimated delivery were analyzed. Status indicators such as durations are transmitted to ground systems and, where appropriate, to the user. It provides equipment (200). In this way, the learned priorities influence all subsequent decisions —acceptance, 25 Release and transmission—guide, thus managing latency, buffer usage and deadlines. Meeting the intermittent supply line conditions is optimized together. Referring to Figure 3, a store and forward application on a low Earth orbit satellite (100). The curve shows the evolution of buffer utilization over time for user 30. Packages from equipment (200) during the period when the supply line is unavailable It rises while being stored and the scheduler (130) places the buffered packets when transmitting them. It falls during the predicted connection windows to the station (300). Buffer memory (120), The occupancy rate is managed in such a way that it returns to a target level rather than reaching saturation; In particular, lower priority or older packets, buffer management system (121) 35 11 can be released preventively before reaching harsh limits and is time-sensitive. The area can be preserved for traffic. The training of the AI ​​engine (110) is given in more detail below: The AI ​​engine (110) maps a state of Sₜ to an action of Aₜ ∈ {send, store, drop}. Reinforcement learning is trained as a decision module and uses finite buffer and discontinuous feed-line. Maximizing long-term delivery performance under constraints objectives At each decision time (t), the environment provides Sₜ which is a fraction of Bmax as follows: 10 current buffer occupancy B(t) and priority between buffered packets Distribution of levels; attributes per packet (priority (πᵢ), dimension (sᵢ), due date (tᵢ)); feedline availability aᵢ(t) ∈ {0,1}; deadline for packets in the queue remaining time (tᵢ - t); and dropped packet numbers, average latency and from previous actions Historical performance summaries, including the volume of data generated. 15 The representative chooses from among the following: a packet (Pᵢ) selected when aᵢ(t)=1. transmission; storage of a packet in the buffer; or higher priority or less Dropping a packet to free up capacity for time-sensitive data. Action Masking prevents transmission when the supply line is unavailable. 20 The reward per step (Rₜ) accumulates: ∑ successful delivery: R_delivery = Σᵢ δᵢ-πᵢ, where δᵢ=1 if Pᵢ is delivered on time, Otherwise it is 0; 25 ∑ buffer stability: R_buffer = −((B(t) / Bmax) − β)², where β is a target occupancy (for example, 0.7); ∑ Priority-based retention: R_priority = Σ_{i∈buffer} πᵢ·yᵢ, where Pᵢ is preserved, yᵢ=1. Punishments deter unwanted behavior: ∑ Missed deadlines: P_miss = −Σᵢ (1−δᵢ)·πᵢ; ∑ Buffer overflow: If B(t) > Bmax, then P_overflow = -∞ (or in practice a very large one) negative constant); 35 12 ∑ Dropping high-priority packets: P_drop = −Σ_{i∈dropped} πᵢ. Objective and learning framework. Representative, with discount factor γ, G = Σ_{t=0}^{T} γᵗ Rₜ It maximizes cumulative discounted returns. A compact neural policy and value. function, environment (satellite buffer dynamics, feedline availability aᵢ(t) and 5 reinforcement learning (e.g., deep Q-learning) within a simulator that generates (incoming traffic) or proximal policy optimization) is used for training. Training data sources are varied. Synthetic tracks and satellites with buffer sizes, priority mixes, and deadlines. It includes real-world records from communication systems. Traffic density, Domain randomization improves robustness in terms of downtime and connection quality. 10 The policy being trained includes deadline-missed rate, dropout rate by priority class, average, and Synthetic scenarios and history maintained using termination delay and data volume. This is verified over repetitions. The final network is quantified and a deterministic, limited network is tested on the satellite. Compiled for delayed inference; health 15 to ensure safe operation on the anomaly. Controls and rule-based feedback are provided. Prioritization model (111) packet metadata (nominal priority, size and final delivery (historical) space and buffer allocation and programming guidance for each package It is a machine learning classifier that assigns a priority score (or class). 20 Feedline prediction model (112), historical link availability, satellite It is a time series forecasting model trained on location / ephemeris and environmental factors. (e.g., LSTM or ARIMA) and including the start and duration of the upcoming window. It provides an output of the estimated supply line availability over time. 25 The scope of protection of the invention is specified in the attached claims, and this detailed description includes... It cannot be limited to what is explained for illustrative purposes. Technical expert. a person, without deviating from the main theme of the invention, in light of the facts mentioned above It is clear that it may exhibit similar arrangements. 30 AI-based Store and Forward (S&F) systems, especially for Non-Geostationary Orbit (NGSO) WG SA1 focuses on delay-tolerant communication services for constellations. Version 19 is compliant with the 3GPP standards specified in SI. S&F according to TR 22.865, satellite. Critical use cases such as IoT applications and UE-to-UE communication via relays 35 13 by addressing their situations under satellite coverage without a simultaneous active feedline. It provides communication. This is particularly valuable for GNSS-independent operations, and Positioning and communication for remote areas without direct GNSS access. It improves their abilities. In practice, as detailed in TR 23.700-29, directly or indirectly S&F is indispensable in scenarios where supply lines are unavailable. For example, in an active A satellite without a feeder line can store data locally and maintain a connection. It can transmit it when it becomes available, thus enabling IoT monitoring devices in remote areas. It provides continuous service. This approach is examined in WG SA5 Version 19 SI as well as NTN 10. It supports end-to-end management in various scenarios. Additionally, the system enables IoT devices to respond to disasters. Data in Mobile-Based (MO) communications, such as transmitting critical environmental data during transit. It ensures integrity and latency optimization. The S&F mechanism prioritizes matching. It effectively minimizes data storage costs while maximizing efficiency, and NTN It offers a scalable solution for communication challenges. 15

Claims

14 REQUESTS 1. A non-terrestrial network (10) satellite (100) for store and transmit operation a method for operation, the method is from user equipment (200) receiving data packets, data packets temporarily stored in a buffer (120) 5 storage and a place where data packets are located when a feed line is available It includes transmitting to the station (300), the feature of the method is also the following: It is characterized by its inclusion of: - Package 10, including nominal priority, size, and deadline for each package. extraction of its characteristics; - a priority to each packet by a prioritization model (111) assignment of the score; - a hierarchical intermediate with priority scores and multiple priority levels Buffer space allocation for incoming packets based on memory structure 15 being done; - past connection by a supply line prediction model (112) data, satellite positioning and environmental factors, the supply line using a time and a period when it is available; - by a pre-trained AI engine (110), to the buffer state and 20 A transmission action for each data packet based on packet information, a the act of keeping and the choice of an act of letting go; - Determining transmission windows from the forecast and high priority packets will be transmitted before delay-tolerant packets Scheduling packet transmissions within windows. 25 2. The method is based on claim 1 and its characteristic is that it is located via a control interface. transmission status and expected delivery time to stations (300) or devices It is characterized by its provision.

3. This is a method based on one of the previous requirements, and its feature is pre-trained AI. using synthetic traffic tracks and / or past satellite records of the engine (110) This is achieved through reinforcement learning, the motor is trained with: (i) according to Bmax buffer occupancy and priority allocation, nominal priority (πᵢ), size (sᵢ), deadline (tᵢ) and remaining time (tᵢ-t), an indicator of supply line availability and the past 35 (ii) a case (Sₜ) containing per-package attributes including performance metrics; (iii) a set of actions including transmit, store and release; and (iii) packing around target occupancy (β). Increased on-time delivery weighted by priority and buffer stability. and missed deadlines, buffer overflow, and higher priority packets A reward policy that imposes penalties for its reduction, a reduced cumulative reward 5 It is optimized to maximize efficiency.

4. According to one of the previous requests, the method is the act of abandoning the characteristic, preserving. when thresholds are exceeded or are expected to be exceeded, a priority threshold 10 characterized by involving the dropping of low-priority packets. It is done.

5. A non-terrestrial network (10) satellite (100) for store and transmit operation, satellite (100) to receive data packets from user equipment (200), temporarily store data packets to store in a buffer (120) and data when a feed line is available 15 It is configured to transmit packets to a ground station (300), its feature is NTN The satellite (100) is to carry out one of the methods according to request 1-4. It is the configuration.