Routing method for supporting differential service requirements in satellite internet

Abstract

A routing method for supporting differential service requirements in a satellite internet comprises the following steps: step 1, constructing multi-dimensional network state perception as a base layer which is responsible for collecting and quantifying physical states and logic states of a satellite network in real time, and obtaining objective network states and subjective QoS (Quality of Service) requirements of services; 2, generating a nonlinear adaptive weight driven by a service as a strategy layer, mapping an objective network state and a subjective QoS (Quality of Service) demand of the service, dynamically adjusting the value orientation of routing judgment, and generating a dynamic weight; and step 3, taking the dynamic cost routing decision as an execution layer, calculating the comprehensive cost based on the dynamic weight, and executing next hop selection and path maintenance, and through cooperative work of the dynamic cost routing decision, the comprehensive cost and the next hop selection and path maintenance, realizing intelligent path selection and high-reliability transmission for different service flows in a high-dynamic and resource-limited environment of the low-orbit satellite. According to the invention, while the routing computation overhead is effectively reduced, multiple requirements of differentiated services on low time delay and high reliability can be considered.

Description

Technical Field

[0001] This invention belongs to the field of satellite communication technology, and specifically relates to a routing method that supports differentiated service requirements in satellite internet. Background Technology

[0002] In low-Earth orbit (LEO) constellation networks, routing decisions are crucial for ensuring efficient data transmission. As a key infrastructure for global communication coverage, LEO constellations offer significant advantages over traditional geostationary orbit (GEO) satellites, including lower signal propagation latency, less path loss, and stronger global coverage. However, LEO constellation networks operate under unique conditions: high topology dynamics, limited onboard resources, and high node mobility. These characteristics lead to significant shortcomings in traditional routing strategies in practical applications, making it difficult to meet the diverse latency, reliability, and throughput requirements of complex services. Existing technologies employ some routing schemes based on virtual topology concepts for pre-calculation of routes. While this approach reduces the frequency of route updates to some extent, its core flaw lies in the granularity of the topology snapshot. Too coarse granularity fails to reflect the dynamics of LEO constellation networks, while too fine granularity leads to a sharp increase in routing computation and storage overhead. Therefore, static routing decisions based on pre-calculation struggle to reflect real-time fluctuations in link quality, sudden changes in services, and timely avoidance of local congestion, exhibiting significant hysteresis and insufficient robustness. Furthermore, greedy routing strategies based solely on geometric location (such as GPSR) also present challenges. This method leverages the spatial relationship between nodes and destinations to quickly determine the next hop, making it well-suited to the constantly changing topology of LEO constellations. However, such methods consider only "geographical distance" as a single decision criterion, completely ignoring crucial factors such as service type, link quality fluctuations, and node load. Since a large volume of data flows are concentrated and forwarded along the geometrically optimal path, this can easily lead to long-term congestion on local hotspot links, increased latency, and even routing holes. Furthermore, different services have vastly different priorities regarding network performance. For example, URLLC services emphasize extremely low latency and high reliability, while eMBB services prioritize stable throughput. Traditional greedy strategies cannot identify or differentiate service requirements, resulting in a poor user experience.

[0003] In their classic work, "GPSR: Greedy Perimeter Stateless Routing for Wireless Networks," Karp and Kung proposed a greedy stateless routing method based on node and destination geographic locations. This method faces the challenge of high-frequency topology changes in LEO constellations compared to terrestrial AdHoc networks, resulting in rapid link establishment and disconnection, making decisions based on geometric location prone to failure. Furthermore, GPSR selects the next hop solely based on geographic location, completely ignoring link quality, congestion levels, and service type, easily leading to a large concentration of traffic along the "shortest path." Additionally, this type of method does not support service differentiation and cannot meet the needs of multi-service concurrent scenarios.

[0004] In the research on multi-factor routing based on link quality and load, attempts were made to integrate link quality and node load for next-hop selection, such as "Link Quality Guided Routing (LQGR)" and various "Load-aware routing" schemes. The typical schemes are as follows: (1) Estimate link stability based on SNR, BER, transmission success rate or predicted RTT. (2) Each node periodically broadcasts queue length / load level to its neighbors to achieve node load notification. (3) Combine factors such as link quality, load, and distance into a cost function using static or semi-static weights. (4) Select the neighbor with the lowest cost as the next hop based on the multi-factor weighted decision implemented by the cost function. Although this method introduces link layer and load layer information, its weights are usually static or semi-static, and the method cannot automatically adjust the weights. At the same time, frequent broadcasting of load information is expensive and prone to lag in low-Earth orbit constellations. It cannot distinguish between "congestion related to the current candidate next hop" and "congestion unrelated to the direction". In addition, this method ignores the impact of service type on routing strategy and lacks fine-grained control over services.

[0005] The paper "Reinforcement Learning based Routing for Satellite Mega-Constellations" proposes using machine learning methods (Q-learning, deep reinforcement learning, and graph neural networks) to construct intelligent routing decisions. However, this method suffers from significant training costs in practical low-Earth orbit constellation deployments, requiring large amounts of simulation data, massive computing power, and extensive storage, making it unsuitable for large-scale simultaneous satellite deployments. Furthermore, service characteristics are difficult to directly map to neural network inputs, leading to performance instability across different service types, and the "black box" nature of reinforcement learning hinders the interpretability and controllability of critical communication tasks.

[0006] In summary, most existing multi-factor routing schemes employ a "static linear weighting" approach to combine various indicators. This linear mechanism suffers from severe response lag in the highly dynamic environment of low-Earth orbit constellations: when network load approaches saturation, the adjustment of linear weights is often insufficient to trigger path switching, causing data packets to remain queued in congested links; and when a link is about to break, traditional threshold-based decisions often only trigger rerouting after the link is disconnected, leading to drastic fluctuations in the TCP congestion window. Furthermore, existing link lifetime prediction methods largely rely on centralized ground-based computing or complex machine learning inference, making it difficult to achieve millisecond-level real-time analysis and calculation under limited onboard computing power. Therefore, the key to solving these problems lies in constructing a non-linear, low-overhead routing decision mechanism that accurately matches service QoS characteristics under limited computing power. Summary of the Invention

[0007] To overcome the shortcomings of the existing technology, the present invention aims to provide a routing method that supports differentiated service requirements in satellite internet. This method constructs a multi-dimensional constraint model that integrates propagation distance, link status, and node load, and quickly filters out failed nodes from the neighbor set to form a high-confidence candidate subset, thereby breaking the limitations of traditional single-index routing. The present invention can effectively reduce the routing calculation overhead while taking into account the multiple requirements of differentiated services for low latency and high reliability.

[0008] To achieve the above objectives, the technical solution adopted by the present invention is as follows: A routing method supporting differentiated service requirements in satellite internet includes the following steps: Step 1: Construct a foundational layer for multidimensional network state awareness, responsible for real-time collection and quantification of the physical state (location, speed, link lifetime) and logical state (load, congestion) of the satellite network, to obtain objective network state and subjective QoS requirements of services; Step 2: Using a policy layer that generates non-linear adaptive weights based on business needs, the objective network state obtained in Step 1 is mapped to the subjective QoS requirements of the business (latency-sensitive or throughput-sensitive), and the value orientation of routing decisions is dynamically adjusted to generate dynamic weights. Step 3: Utilizing the execution layer based on dynamic cost routing decision, the comprehensive cost is calculated based on the dynamic weights generated in Step 2, and specific next-hop selection and path maintenance are performed; through the collaborative work of the basic layer, strategy layer and execution layer, intelligent path selection and high-reliability transmission for different service flows are realized in the highly dynamic and resource-constrained environment of low-Earth orbit satellites.

[0009] Step 1 specifically involves: Step 1.1: Based on the hybrid network state awareness strategy, the original motion vector information and load data of neighbors are obtained with low overhead by combining piggybacking and adaptive detection. Step 1.2: Convert the original motion vector information and load data into normalized routing metrics (propagation potential, link instability, congestion potential) to obtain key indicators; Step 1.3: Threshold-based neighbor node filtering mechanism (neighbor node filtering) uses the key indicators (such as Link Remaining Lifetime LRT) calculated in Step 1.2 to set hard thresholds, and pre-removes unusable nodes before entering the complex weighted calculation, so as to obtain the objective network status and the subjective QoS requirements of the service.

[0010] Step 1 not only provides an accurate view of the network status, but also significantly reduces the search space for subsequent route optimization through pre-screening, thereby reducing the on-board computing burden.

[0011] Step 1.1 specifically includes: Step 1.1.1. In-path piggyback mechanism: When satellite node with neighbors When business data transmission is in progress, the node does not send independent data. Instead of signaling, it uses the extended header of the data packet to include the node's own location vector. Velocity vector and load status Encapsulated in the data packet header; receiving node While parsing the business data, extract this header information to update the neighbor status table; Step 1.1.2. Adaptive Active Detection: Each node maintains a last-update timestamp for each neighbor. The node periodically checks the timestamp, only checking if the current time... Only then does the node switch to active mode and send lightweight data. Detection packet.

[0012] The specific steps of step 1.1.2 are as follows: Step (1): Define the state interaction cycle Satellite nodes The set of neighboring nodes within its one-hop communication range Broadcast probe message (Hello Packet); Step (2): The message payload contains the following key state vectors : Kinematic state: Three-dimensional position vector based on the geocentric inertial coordinate system With velocity vector ; Resource load status: CPU utilization after exponentially weighted moving average smoothing. and cache queue utilization ; Step (3): The satellite node receives and maintains the "neighbor status table" and uses the SGP4 orbit prediction model to calculate the neighbor's position in real time to eliminate the position error caused by the propagation delay; The neighbor maintenance strategy is as follows: satellite nodes maintain a dynamic neighbor list and set an aging timer. If a neighbor's Hello packet is not received before the timer expires, the link is considered to be interrupted, the neighbor is deleted from the list, and a local route update event is triggered.

[0013] Step 1.2 specifically includes: Current satellite node Upon receiving neighbor nodes After obtaining the status information, for the current node to the destination node Calculate the following three core metrics: i. Normalized propagation potential ( Introducing propagation potential based on spatial geometry to guide data packets toward the destination node. Approaching, setting up satellite nodes Neighbor nodes and destination node The position vectors are respectively Satellite nodes arrive The Euclidean distance is ,Neighbor arrive Euclidean distance Construct a nonlinear distance cost based on the Sigmoid function: in Here, δ is the distance sensitivity coefficient, and δ is the bias term. This represents whether the distance relative to the current node has decreased (negative value) or increased (positive value), when the neighbor is significantly closer to the destination. The exponential term increases. When the neighbor is far from the destination ( )hour ; ii. Cost of Link Instability The model is based on the Earth's core. As the origin, the satellite and position vector and The only certainty; the geocentric angular size between the two stars Through vector dot product Quickly obtain; when the relative motion of the two stars causes The link is interrupted when the distance increases to the Earth's blocking boundary; the relative angular velocity is directly calculated using vector cross product. Thus, without iteration, through the formula Accurately predict the remaining time of the link interruption; Let the Earth's radius be... The orbital altitude is The minimum elevation angle is Maximum geocentric angle for maintaining connection between two stars for: Using satellite nodes Calculate the current geocentric angle using the dot product of position vectors. Combined with relative angular velocity Calculate the remaining link time : in , It is the angular velocity vector; Using an inverse proportional mapping, an extremely high penalty is imposed on links that are about to be disconnected: in To switch protection time, The time normalization constant is used. When the link is about to break, the cost is directly set to 1 (maximum penalty) to force route avoidance; when the link lifetime is very long, the cost approaches 0. iii. Node congestion potential cost We introduce the M / M / 1 queuing theory model, assuming that the arrival of messages at satellite nodes follows a Poisson distribution (parameter λ), the service time follows an exponential distribution (parameter μ), and the average delay of the nodes is... With system utilization Nonlinear relationship Simply using queue length cannot reflect the trend of nonlinear latency deterioration under high load; this invention defines node congestion potential. As a kind of "congestion risk probability": in This is the overall load factor; the formula ensures that when the load is close to 1, the cost is... A steep increase approaching 1 creates a "soft rejection" mechanism. This method gives the routing algorithm a strong "repulsive force" against high-load nodes, which is crucial for latency-sensitive services. Step 1.3 specifically includes: Construct a threshold-based filtering mechanism, where nodes traverse their neighbor lists. Remove neighboring nodes that do not meet the following constraints to form a valid candidate set. ; 1. Link Remaining Lifetime Constraint: If the remaining lifetime of a neighboring link is limited... Below the system's set protection threshold If the link is found to be on the edge of unavailability, it will be removed from the candidate set. 2. Node load saturation constraint: If the overall load rate of a neighboring node is... Exceeding the saturation threshold This indicates that the node's cache or CPU has no available resources, and forcibly forwarding will result in packet loss; in this case, the node is directly excluded and will no longer participate in the subsequent weighted cost calculation.

[0014] Through a pre-screening mechanism, the algorithm automatically filters out "bad nodes" in the network, ensuring that subsequent nonlinear weight calculations are performed only in healthy subsets of the topology.

[0015] Step 2 specifically involves: By parsing the service type identifier in the packet header, it can be determined whether the current data flow is latency-sensitive or throughput-sensitive; at the same time, the routing cost function is defined as: The weights are adjusted in real time based on the current network congestion level; the weight vector... Define network background congestion The average load of neighboring nodes: .

[0016] Weights that meet the QoS requirements of different service types are generated separately. The specific generation logic is as follows: Step 2.1: URLLC business weight generation function strategy logic: Dynamic weights are constructed using the hyperbolic tangent function: congestion weight γ (for...) Distance weight (against and stability weights (against ) in The congestion trigger threshold, This is the sensitivity coefficient. When At that time, γ remained at a low level; Set congestion trigger threshold Set the sensitivity coefficient The weight γ is set to 10 to ensure that when the load exceeds the threshold, the weight γ can quickly saturate to the maximum value within 0.1 seconds; when the network load is lower than this value, the congestion weight γ remains low (about 0.1), and the routing strategy degenerates to shortest path priority; once the load exceeds 0.6, the curve shows a steep upward trend and quickly saturates to 0.9, forming a strong rejection of congested nodes; Step 2.2: eMBB Business Weight Generation Function Strategy Logic eMBB (throughput-sensitive) services prioritize transmission continuity and are sensitive to the TCP window halving effect caused by routing changes; their weighting is determined by the average remaining lifetime of local links. Driven by; Formula definition: Dynamic weights are constructed using an exponential decay function: When local links are generally unstable ( When the β weight is relatively small, the β weight increases significantly, forcing the algorithm to select only those "long-lived" links to ensure transmission stability.

[0017] Step 3 specifically involves: Based on the above perception and strategy, the routing decision-making process is divided into three stages: cost calculation, next hop selection and stability maintenance. Step 3.1: Route cost calculation: Step (1): The routing module parses the service type of the data packet to be forwarded; Step (2): Collect the current neighbor set The state of the local network background index is calculated. or ); Step (3): Adjust the corresponding policy function in Section 6.2 to generate the dynamic weight vector at the current time step. ; Step (4): Traverse the neighbor list and calculate the comprehensive routing cost for each neighbor using the formula. ; Step 3.2: Next-hop selection mechanism: i. Greedy forwarding judgment mechanism The greedy forwarding decision mechanism searches for the neighbor node that minimizes the overall cost. : ii. Local extremum recovery Compare The virtual cost of the current node itself ; (1) If , execute forwarding.

[0018] (2) If the cost of all neighbors is higher than that of itself (entering a routing hole), trigger the surrounding forwarding mode. Use the right-hand rule to circumvent the hole boundary in the planar local topology until a node that satisfies the greedy forwarding condition is found.

[0019] Step 3.3: Routing stability maintenance mechanism i. The principle of preemptive soft handover for URLLC: Step (1): The node continuously monitors the LRT of the current next-hop link; Step (2): When (Switch preparation time, such as 500ms before disconnection) The router does not need to wait for the link to disconnect and immediately calculates the alternative next hop. At this point, the IP layer copies the packet, adding two different routing headers to the same data packet, and sends them to different destinations. and ; Step (3): The destination node maintains a "received sequence sliding window". It checks whether a data packet is a duplicate based on its sequence number. To support accurate deduplication at the receiving end, a unique "stream sequence number" is encapsulated in the extended field of the data packet header. Step (4): When the old node The connection was completely interrupted or At that time, stop sending Send, seamlessly restored to Single-path transmission reduces packet loss during handover and ensures transmission stability. ii. Hysteresis offloading strategy for eMBB: For throughput-sensitive services, it is necessary to avoid the routing ping-pong effect caused by slight parameter fluctuations, and at the same time, utilize multipath to improve throughput; This is achieved through two methods: hysteresis switching and multipath shunting. (1) Delayed shunting: Only when the cost advantage of the new path exceeds the threshold. Time (i.e.) Only after this process is complete will the routing table be updated; (2) Multipath diversion: If there exists a suboptimal node k, its cost is close to that of the optimal node ( If so, a traffic splitting group is established. The traffic allocation ratio is inversely proportional to the cost: This enables bandwidth aggregation and load balancing.

[0020] The beneficial effects of this invention are: This invention achieves precise mapping and adaptive adaptation between service QoS requirements and dynamic network states. By executing the dynamic weight generation based on nonlinear functions in step 2, this invention breaks through the limitations of traditional routing metric criteria being singular and static. It introduces the Tanh function and exponential decay function to establish a nonlinear mapping model between objective network states (such as latency, load, and stability) and subjective service requirements (such as URLLC and eMBB). This enables routing decisions to dynamically adjust the priority of evaluation dimensions according to service type. In situations of network congestion or highly dynamic topology fluctuations, it sensitively captures marginal changes in metrics to provide customized optimal paths for differentiated service flows.

[0021] This invention significantly improves link reliability and transmission continuity under highly dynamic topologies. Through quantitative modeling of link instability in step 1.2 and the routing maintenance mechanism in step 3, this invention leverages cross-layer awareness of satellite motion vectors and remaining link lifetime (LRT) to predict link failure risks. Combined with the preemptive soft handover logic in step 3, the system can guide smooth migration of service flows in advance during the critical period before the physical link actually disconnects. Simulation results show that, compared to traditional post-disconnection readdressing mechanisms, this invention significantly reduces frequent link handovers and resulting packet loss caused by high-speed satellite motion, effectively ensuring the throughput stability of long-running services.

[0022] This invention effectively suppresses local congestion and optimizes the load distribution and resilience of network resources. In step 1.2, the invention introduces congestion potential energy modeling, which, combined with the hysteresis-based traffic splitting strategy in step 3, enables routing decisions to possess "proactive disaster avoidance" capabilities. Through hysteresis-based decision logic, this invention effectively solves the "routing oscillation" and "ping-pong effect" problems commonly found in traditional load balancing algorithms. In high-load network scenarios (such as load rates exceeding 80%), this mechanism can guide traffic to avoid hotspot areas and reasonably distribute pressure, maintaining end-to-end latency at a low level and significantly improving the satellite network's capacity to handle sudden surges in traffic.

[0023] This invention achieves a highly efficient balance between decision-making accuracy and computational overhead, and is well-suited to the constrained computing environment on satellites. Through the threshold-based neighbor node filtering mechanism in step 1.3, this invention employs a cascaded algorithm architecture of "pre-screening-fine computation." By using physical indicators such as link lifetime to set hard entry thresholds, potentially unavailable or unstable neighbor nodes are pre-emptively eliminated before entering the high-order, complex weighted calculations. This mechanism ensures high path availability while significantly reducing the search space for route optimization, substantially decreasing the algorithm's time complexity and computational entropy, and alleviating the computational burden on low-Earth orbit satellite processors. Attached Figure Description

[0024] Figure 1 This is a schematic diagram of a link lifetime prediction model based on spherical geometry.

[0025] Figure 2 This is a congestion weight response curve for URLLC (URL-sensitive communication) services.

[0026] Figure 3 This is a stability weight response curve for throughput-sensitive (eMBB) services.

[0027] Figure 4 This is a diagram illustrating the overall architecture and logical flow of satellite internet routing methods.

[0028] Figure 5This is a logic diagram for dynamic weight generation and mapping based on business needs.

[0029] Figure 6 Flowchart for next-hop selection mechanism and local routing hole recovery.

[0030] Figure 7 This is a performance simulation comparison chart of the embodiments of the present invention and existing routing technologies in terms of end-to-end latency, packet delivery rate, and throughput jitter.

[0031] Figure 8 This is a timing interaction diagram of the routing stability maintenance mechanism for differentiated services. Detailed Implementation

[0032] The present invention will now be described in further detail with reference to the accompanying drawings.

[0033] like Figure 1 As shown, this invention discloses a routing method for supporting differentiated service requirements in satellite internet, comprising the following steps; Step 1: Multidimensional Network State Awareness Model: Step 1.1: Hybrid Network State Awareness Strategy: 1. Path-based piggybacking mechanism: When a node... with neighbors When business data transmission is in progress, the node does not send independent data. Signaling. Instead, it uses the extended header of the data packet to include the node's own position vector. Velocity vector and load status Encapsulated in the data packet header. Receiving node. While parsing the business data, extract this header information to update the neighbor status table.

[0034] 2. Adaptive Proactive Detection: Nodes maintain a last-update timestamp for each neighbor. The node periodically checks this timestamp, only checking if the current time... Only then does the node switch to active mode and send lightweight data. The detection packet, the specific steps are as follows: Step (1): Define the state interaction cycle Satellite nodes Periodically communicate with the set of neighboring nodes within its one-hop communication range. Broadcast probe message (Hello Packet). The message structure is shown in Table 1.

[0035] Step (2): The message payload contains the following key state vectors :: (1) Kinematic state: Three-dimensional position vector based on the geocentric inertial coordinate system With velocity vector ; (2) Resource load status: CPU utilization after exponential weighted moving average smoothing. and cache queue utilization .

[0036] Step (3): The node receives and maintains the "neighbor status table" and uses the SGP4 orbit prediction model to calculate the location of the neighbor in real time to eliminate the location error caused by the propagation delay.

[0037] Neighbor maintenance strategy: Nodes maintain a dynamic list of neighbors and set an aging timer. If a neighbor's Hello packet is not received before the timer expires, the link is considered to be broken, the neighbor is deleted from the list, and a local route update event is triggered.

[0038] Table 1 Hello Message Design message header motion status field Load status field Link prediction field Step 1.2: Routing Metrics Current node Upon receiving neighbor nodes After obtaining the status information, for the current node to the destination node Calculate the following three core metrics: i. Normalized propagation potential (1) Normalized propagation potential energy ( Unlike traditional hop count metrics, this invention introduces propagation potential based on spatial geometry. This is to guide data packets towards the destination node. Approximation, set source node Neighbor nodes and destination node The position vectors are respectively .node arrive The Euclidean distance is ,Neighbor arrive Euclidean distance In order to effectively suppress reverse routing (i.e. Furthermore, to improve sensitivity to distance reduction, a nonlinear distance cost based on the Sigmoid function is constructed: in δ is the distance sensitivity coefficient, and δ is the bias term. This indicates whether the distance relative to the current node has decreased (negative value) or increased (positive value). When a neighbor is significantly closer to the destination ( The exponential term increases. When the neighbor is far from the destination ( )hour Compared to traditional linear normalization This formula has higher discriminative power when approaching the destination and can effectively avoid route oscillations.

[0039] ii. Cost of Link Instability (2) Cost of link instability To address the frequent link disconnections caused by the relative motion of low-Earth orbit satellites, this invention utilizes spherical geometry to derive the Link Residual Time (LRT). Figure 1 The figure shows the geometric model for link lifetime prediction, with the model centered on the Earth. As the origin, the satellite and position vector and The only certainty. The geocentric angular size between the two stars. Through vector dot product Quickly obtain. When the relative motion of the two stars causes... The link breaks when the distance increases to the Earth's blocking boundary. The relative angular velocity is directly calculated using the vector cross product. Thus, without iteration, through the formula It accurately predicts the remaining time of the link interruption.

[0040] Let the Earth's radius be... The orbital altitude is The minimum elevation angle is Maximum geocentric angle for maintaining connection between two stars for: Utilizing nodes Calculate the current geocentric angle using the dot product of position vectors. Combined with relative angular velocity Calculate the remaining link time : in , This is the angular velocity vector.

[0041] To convert the time dimension to the cost in the [0, 1] interval and to reflect the characteristic that "the shorter the time, the greater the risk," a "cost" function was constructed, employing an inverse proportional mapping to impose an extremely high penalty on links that are about to be disconnected: in To switch protection time, The time normalization constant is used. When the link is about to break, the cost is directly set to 1 (maximum penalty) to force route avoidance; when the link has a long lifespan, the cost approaches 0.

[0042] Through formula This method accurately predicts the remaining time of link interruption. Utilizing the geometric analytical properties of spherical trigonometry, it eliminates the need for cyclic simulations, unlike traditional time-step-based iterative extrapolation methods. The constant time complexity directly calculates the link connection and disconnection time, which greatly reduces the computational burden on the on-board processor and meets the real-time requirement of millisecond-level route updates.

[0043] iii. Node congestion potential (3) Node congestion potential cost ( To accurately assess node processing latency, an M / M / 1 queuing theory model is introduced. This model assumes that message arrival at satellite nodes follows a Poisson distribution (parameter λ), service times follow an exponential distribution (parameter μ), and the average latency of the nodes is... With system utilization Nonlinear relationship Simply using queue length cannot reflect the nonlinear deterioration trend of latency under high load. This invention defines node congestion potential. As a kind of "congestion risk probability": in This is the overall load factor. This formula ensures that when the load is close to 1, the cost... The steep increase in load approaches 1, creating a "soft rejection" mechanism. This method gives the routing algorithm a strong "repulsive force" against high-load nodes, which is crucial for latency-sensitive services.

[0044] Step 1.3: Threshold-based neighbor node filtering mechanism: Before calculating the overall routing cost, to reduce routing computation overhead and ensure basic link reliability, this invention constructs a threshold-based filtering mechanism. Nodes traverse their neighbor lists. Remove neighboring nodes that do not meet the following constraints to form a valid candidate set. : 1. Link Remaining Lifetime Constraint: If the remaining lifetime of a neighboring link is limited... Below the system's set protection threshold If the link is found to be on the edge of unavailability, it will be removed from the candidate set.

[0045] 2. Node load saturation constraint: If the overall load rate of a neighboring node is... Exceeding the saturation threshold This indicates that the node's cache or CPU has no available resources, and forcibly forwarding will result in packet loss. In this case, the node is directly excluded and will no longer participate in the subsequent weighted cost calculation.

[0046] Through a pre-screening mechanism, the algorithm automatically filters out "bad nodes" in the network, ensuring that subsequent nonlinear weight calculations are performed only in healthy subsets of the topology.

[0047] Step 2: Business-driven nonlinear adaptive weight generation mechanism; Existing technologies mostly use fixed weights, which cannot adapt to the time-varying nature of network congestion. This invention identifies whether the current data flow is latency-sensitive or throughput-sensitive by parsing the service type identifier in the packet header. Simultaneously, the routing cost function is defined as: The weights are adjusted in real time based on the current network congestion level; the weight vector... Define network background congestion. The average load of neighboring nodes: .

[0048] Weights that meet the QoS requirements of different service types are generated separately. The specific generation logic is as follows: Step 2.1: URLLC business weight generation function strategy logic: URLLC (Latency-Sensitive Communication) services are extremely sensitive to queuing latency. The core challenge lies in the fact that when network background congestion is high... When the (local neighbor average load) is low, it is mainly dominated by the distance factor; when When network congestion escalates, the congestion factor weight γ needs to increase non-linearly, at which point queuing delay becomes the dominant factor in total delay. Therefore, the weight should not be a constant, but rather... The function. Formula definition: Dynamic weights are constructed using the hyperbolic tangent function: congestion weight γ (for...) Distance weight (against and stability weights (against ) in The congestion trigger threshold, This is the sensitivity coefficient. When When the background load exceeds the threshold, γ remains low, and routing primarily considers distance. Once the background load exceeds the threshold, γ rapidly and non-linearly increases, forcing the algorithm to prioritize congestion avoidance. This strategy achieves a smooth switch from "distance-first" to "congestion avoidance-first".

[0049] exist Figure 2 In the Tanh-type weighted response for URLLC services, a congestion trigger threshold is set. Set the sensitivity coefficient The value is set to 10 to ensure that when the load exceeds the threshold, the weight γ can quickly saturate to its maximum value within 0.1 seconds. When the network load is below this value, the congestion weight γ remains low (approximately 0.1), and the routing strategy degenerates to shortest path priority; once the load exceeds 0.6, the curve shows a steep upward trend, quickly saturating to 0.9, forming a strong rejection of congested nodes.

[0050] Step 2.2: eMBB business weight generation function strategy logic: eMBB (throughput-sensitive) services prioritize transmission continuity and are sensitive to the TCP window halving effect caused by route switching. Their weighting is determined by the average remaining lifetime of local links. Driven by: Formula definition: Dynamic weights are constructed using an exponential decay function. When local links are generally unstable ( When the β weight is relatively small, the β weight increases significantly, forcing the algorithm to select only those "long-lived" links to ensure transmission stability.

[0051] Figure 3 This demonstrates the exponentially decaying weighted response for eMBB services. A time constant is set. benchmark weight While ensuring link stability, routing decisions still possess a certain load balancing capability. This applies when the link has remaining lifetime. When sufficient, stability weight Maintain the baseline value; when When entering the 5-second protection window, It grows exponentially, forming a high-cost barrier and forcibly guiding traffic to switch to long-life links.

[0052] Figure 5 The invention details the dynamic weight generation logic for different service types. For latency-sensitive services (right branch), the system uses network background congestion as a criterion. Using this as input, and through mapping with the Tanh nonlinear function, when the congestion exceeds a threshold, the algorithm automatically increases the congestion weight (γ), thereby forcing routing decisions to avoid congested nodes; conversely, under low load, it increases the distance weight (γ). The system prioritizes selecting the shortest path. For throughput-sensitive services (left branch), the system uses the remaining link lifetime (LRT) as input and, through an exponential decay function, significantly increases the link weight when the link lifetime falls below the protection threshold. This triggers a preemptive switching mechanism; if the lifetime is sufficient, the baseline weight is maintained to preserve the stability of the current high-bandwidth link.

[0053] Step 3: Dynamic Cost Routing Decision Scheme Based on the above perception and strategy, the routing decision-making process is divided into three stages: cost calculation, next hop selection, and stability maintenance.

[0054] Step 3.1: Route cost calculation: Step 1: The routing module parses the service type of the data packet to be forwarded.

[0055] Step 2: Collect the current neighbor set The state of the local network background index is calculated. or ).

[0056] Step 3: Adjust the corresponding policy function to generate the dynamic weight vector at the current time step. .

[0057] Step 4: Traverse the neighbor list and calculate the comprehensive routing cost for each neighbor using the formula. .

[0058] like Figure 4 As shown, step 3.2: Next-hop selection mechanism: i. Greedy forwarding judgment mechanism The greedy forwarding decision mechanism searches for the neighbor node that minimizes the overall cost. : ii. Local extremum recovery Compare The virtual cost of the current node itself .

[0059] (1) If , execute forwarding.

[0060] (2) If the cost of all neighbors is higher than that of itself (entering a routing hole), the surrounding forwarding mode is triggered. The right-hand rule is used to circumvent the hole boundary in the planar local topology until a node that satisfies the greedy forwarding condition is found.

[0061] like Figure 6This document demonstrates the specific algorithm flow for next-hop selection and local extremum (routing hole) recovery in this embodiment. After route calculation begins, the minimum initial cost is selected first, and the neighbor set is traversed. For each neighbor node, its comprehensive routing cost is calculated. The system updates the current best node. After traversal, the system performs local extremum judgment: comparing the cost of the best neighbor with the virtual cost of the current node itself. If a neighbor with a lower cost exists (…), the system updates the current best node. If the cost of all neighbors is higher than the cost of the current neighbor (i.e., it is trapped in a local optimum), it will automatically switch to peripheral forwarding and use the right-hand rule to circumvent the hole boundary until it finds a node that meets the conditions for greedy forwarding, thus effectively solving the routing hole problem in dynamic topology.

[0062] Step 3.3: Routing stability maintenance mechanism i. Preemptive soft handover principle for URLLC Since latency-sensitive services cannot tolerate reconnection latency after link interruption, preemptive soft handover is used to reduce reconnection latency.

[0063] Step (1): The node continuously monitors the LRT of the current next-hop link.

[0064] Step (2): When (Switch preparation time, such as 500ms before disconnection) The router does not need to wait for the link to disconnect and immediately calculates the alternative next hop. At this point, the IP layer copies the packet, adding two different routing headers to the same data packet, and sends them to different destinations. and .

[0065] Step (3): The destination node maintains a "received sequence sliding window". It checks whether a data packet is a duplicate based on its sequence number. To support accurate deduplication at the receiving end, a unique "stream sequence number" is encapsulated in the extended field of the data packet header. Step (4): When the old node The connection was completely interrupted or At that time, stop sending Send, seamlessly restored to Single-path transmission reduces packet loss during handover and ensures transmission stability.

[0066] ii. Hysteresis-based offloading strategy for eMBB For throughput-sensitive services, it is necessary to avoid the routing ping-pong effect caused by minor parameter fluctuations, while utilizing multipath propagation to improve throughput. This is achieved through two methods: hysteresis handover and multipath routing. (1) Delayed shunting: Only when the cost advantage of the new path exceeds the threshold. Time (i.e.) Only after this process is complete will the routing table be updated.

[0067] (2) Multipath diversion: If there exists a suboptimal node k, its cost is close to that of the optimal node ( If so, a traffic splitting group is established. The traffic allocation ratio is inversely proportional to the cost: This enables bandwidth aggregation and load balancing.

[0068] Figure 8 The timing diagram illustrates the differentiated routing maintenance mechanisms for different services. For latency-sensitive services, the "preemptive soft handover" process involves the current node continuously monitoring the link LRT (Local Time Tolerance). Once the LRT is detected to be below the protection time, the node will initiate a handover. If a new link is established before the old one is lost, a dual-transmission mechanism is enabled, and the receiving end performs deduplication to ensure zero packet loss during the handover. For throughput-sensitive services, a "hysteresis-based traffic offloading strategy" is implemented: after calculating the new path cost... Subsequently, the system introduces delayed decision-making logic. The main path is switched only when the advantages of the new path are obvious; if the costs are similar, traffic is distributed proportionally to the old and new paths to achieve bandwidth aggregation and load balancing.

[0069] Network status awareness based on service packets: During network operation, satellite network nodes use neighbor information carried in packet headers to analyze the network status in real time, and only initiate active probing during idle periods without service interaction. This allows them to obtain multi-dimensional network status information such as propagation distance, link stability, and node load. This method significantly reduces the overhead of periodic signaling interactions while ensuring real-time network status updates.

[0070] A cost quantization model based on nonlinear mapping is proposed: a nonlinear mapping function is constructed to transform multidimensional network states into normalized routing costs. An exponential nonlinear penalty mechanism is adopted for link stability and node load to ensure that the model can keenly perceive network state deterioration, effectively solving the problem of insufficient sensitivity of linear weighted algorithms and preventing data from entering high-risk links that are about to be interrupted or severely congested.

[0071] Threshold-based pre-screening path selection: A threshold-based pre-screening mechanism is established to eliminate nodes with link stability below the threshold and those with overload, forming a high-confidence candidate subset, narrowing the search space for route optimization, and significantly reducing computational overhead. For the screened candidate subset, the weighting coefficients of routing decisions are configured differently based on service QoS requirements: For latency-sensitive services, the weight of propagation distance is emphasized to drive the routing algorithm to converge towards the geometric shortest path, minimizing end-to-end latency; for reliability-sensitive services, the weights of link stability and node load are increased to prioritize avoiding link failure risks and congestion; finally, the locally optimal next hop that best meets the current service requirements is selected from the candidate subset.

[0072] Network state awareness based on service messages directly forms the technical foundation for "Step 1: Multidimensional Network State Awareness Model". This part uses a hybrid sensing mechanism to transform the highly dynamic physical characteristics and logical load status of low-orbit satellites into quantified raw feature vectors, providing a real-time, high-dimensional network situation view for the entire routing system.

[0073] The "cost quantization model based on nonlinear mapping" is the core algorithmic logic of "Step 2: Business-driven nonlinear adaptive weight generation mechanism". This part utilizes an exponential nonlinear penalty mechanism to overcome the shortcomings of traditional linear algorithms in perceiving link deterioration. It acts as a "sensitivity regulator" for system decision-making, dynamically calculating adaptive weights reflecting network risks based on the raw data obtained in Step 1, thus achieving an intelligent leap from objective state to evaluation criteria.

[0074] "Threshold-based pre-screening path selection" serves as the execution and optimization criterion for "Step 3: Dynamic Cost Routing Decision Scheme." This part employs a "pruning first, optimization later" strategy. First, it uses hard thresholds to eliminate inferior nodes to compress the search space. Then, it performs differentiated cost calculations based on specific service QoS requirements (such as latency-sensitive or reliability-sensitive). This step ultimately transforms the perceived data from Step 1 and the nonlinear weights from Step 2 into specific next-hop selections, ensuring the efficiency and service adaptability of routing decisions under limited on-board resources.

[0075] In summary, the above methods provide a complete logical architecture that supports the progressive routing control process of this method, from "multi-dimensional state perception" to "adaptive weight generation" and then to "dynamic path decision-making".

[0076] Simulation conditions This embodiment uses the MATLAB network simulation platform to build a low-Earth orbit satellite constellation communication scenario. Specific parameters are shown in Table 2. Table 2 Simulation Parameters Simulation content and results The routing method of this invention is compared with static shortest route, greedy stateless route, and load-aware route in terms of end-to-end latency, delivery rate, and throughput jitter. The simulation results are as follows: Figure 7 As shown, where Figure 7 .1 compares end-to-end latency under different network loads. Figure 7 .2 compares packet delivery rates under varying link handover frequencies. Figure 7 .3 represents the throughput changes of each routing decision during link switching.

[0077] Figure 7 Section 1 illustrates the end-to-end latency comparison under congestion scenarios. As the network load increases from 10% to 95%, the static shortest route strategy, lacking any congestion awareness, experiences an exponential increase in queuing latency when the load exceeds 50%, as a large number of data packets flood into hotspot nodes on the shortest path. The greedy stateless route strategy is prone to getting trapped in local optima (routing holes) and taking detours, resulting in high base propagation latency that increases linearly with load. While the load-aware route strategy introduces a load factor, its fixed weights prevent it from effectively sacrificing distance for low latency under high load, leading to limited performance improvement. This invention employs a Tanh-type dynamic weight mechanism for latency-sensitive services. When network congestion is detected (load exceeds 60%), the congestion weight γ increases rapidly and non-linearly, forcing the routing algorithm to actively avoid congested nodes. Simulation results show that under high load, the average latency of this invention is significantly lower than the aforementioned routing strategies, with a smooth curve change, effectively ensuring low latency requirements.

[0078] Figure 7 .2 illustrates a comparison of packet delivery rates in a highly dynamic scenario, simulating a scenario where increased relative satellite motion leads to frequent link disconnections. As the link switching frequency increases, the delivery rates of static shortest routes and greedy stateless routes rapidly decrease. This is because they lack the ability to predict link states, continuing to send data even at the moment of link disconnection, resulting in significant packet loss. This invention introduces a link remaining lifetime prediction mechanism based on spherical geometry derivation, predicting the remaining lifetime of a link during the protection time before it is about to disconnect (…). Within the link cost, the instability factor increases rapidly, triggering the "build first, then disconnect" mechanism. The results show that even with frequent link switching, the present invention can still maintain a high delivery rate, demonstrating extremely high robustness.

[0079] Figure 7Section 3.3 demonstrates a comparison of throughput stability during link switching. At the instant of link switching, due to the disconnection of the old link and the lack of connection to the new link, the static shortest route strategy and the greedy stateless route strategy exhibited a significant zero-throughput phenomenon, which subsequently recovered slowly relying on the TCP slow start mechanism, resulting in extremely poor stability. While load-aware routing can find a new path relatively quickly, the lack of hysteresis switching and smooth migration mechanisms still leads to TCP out-of-order delivery and a halving of the congestion window at the moment of switching between the old and new paths, causing a significant instantaneous drop in throughput. This invention benefits from the hysteresis traffic splitting strategy and LRT (Link Remaining Life) dynamic weights designed for throughput-sensitive services, which completes the smooth migration of traffic before the link is disconnected, keeping the throughput fluctuation during link switching within 10%, thus verifying the invention's ability to guarantee continuous transmission with improved throughput.

Claims

1. A routing method supporting differentiated service requirements in satellite internet, characterized in that, Includes the following steps: Step 1: Construct a foundational layer for multidimensional network state awareness, responsible for real-time collection and quantification of the physical and logical states of the satellite network, and obtaining objective network state and subjective QoS requirements of services; Step 2: Using a policy layer that generates non-linear adaptive weights based on business needs, the objective network state obtained in Step 1 is mapped to the subjective QoS requirements of the business, and the value orientation of routing decisions is dynamically adjusted to generate dynamic weights. Step 3: Utilize the execution layer based on dynamic cost routing decision to calculate the comprehensive cost based on the dynamic weights generated in Step 2, and perform next-hop selection and path maintenance. Through the collaborative work of the basic layer, strategy layer and execution layer, intelligent path selection and high-reliability transmission for different service flows are achieved in the highly dynamic and resource-constrained environment of low-Earth orbit satellites.

2. The routing method for supporting differentiated service requirements in satellite internet according to claim 1, characterized in that, Step 1 specifically involves: Step 1.1: Based on the hybrid network state awareness strategy, the original motion vector information and load data of neighbors are obtained with low overhead by combining piggybacking and adaptive detection. Step 1.2: Convert the raw motion vector information and load data into normalized routing metrics to obtain key indicators; Step 1.3: Threshold-based neighbor node filtering mechanism. Using the key indicators calculated in Step 1.2, a hard threshold is set to pre-eliminate unusable nodes before entering the complex weighted calculation, thereby obtaining the objective network status and the subjective QoS requirements of the service.

3. The routing method for supporting differentiated service requirements in satellite internet according to claim 2, characterized in that, Step 1.1 specifically includes: Step 1.1.1: When the satellite node with neighbors When business data transmission is in progress, the node does not send independent data. Instead of signaling, it uses the extended header of the data packet to include the node's own location vector. Velocity vector and load status Encapsulated in the data packet header; receiving node While parsing the business data, extract this header information to update the neighbor status table; Step 1.1.2: Each node maintains a last update timestamp for each neighbor. The node periodically checks the timestamp, only checking if the current time... Only then does the node switch to active mode and send lightweight data. Detection packet.

4. The routing method for supporting differentiated service requirements in satellite internet according to claim 3, characterized in that, The specific steps of step 1.1.2 are as follows: Step (1): Define the state interaction cycle Satellite nodes The set of neighboring nodes within its one-hop communication range Broadcast detection messages; Step (2): The message payload contains the following key state vectors : Kinematic state: Three-dimensional position vector based on the geocentric inertial coordinate system With velocity vector ; Resource load status: CPU utilization after exponentially weighted moving average smoothing. and cache queue utilization ; Step (3): The satellite node receives and maintains the "neighbor status table" and uses the SGP4 orbit prediction model to calculate the neighbor's position in real time to eliminate the position error caused by the propagation delay; The neighbor maintenance strategy is as follows: satellite nodes maintain a dynamic neighbor list and set an aging timer. If a neighbor's Hello packet is not received before the timer expires, the link is considered to be interrupted, the neighbor is deleted from the list, and a local route update event is triggered.

5. A routing method for supporting differentiated service requirements in satellite internet according to claim 4, characterized in that, Step 1.2 specifically includes: Current satellite node Upon receiving neighbor nodes After obtaining the status information, for the current satellite node to the destination node Calculate the following three core metrics: i. Normalized propagation potential Introducing propagation potential energy based on spatial geometry, and setting satellite nodes... Neighbor nodes and destination node The position vectors are respectively Satellite nodes arrive The Euclidean distance is ,Neighbor arrive Euclidean distance Guide data packets to the destination node Approximation, constructing a nonlinear distance cost based on the Sigmoid function: in Here, δ is the distance sensitivity coefficient, and δ is the bias term. This indicates whether the distance relative to the current node has decreased or increased, when the neighbor is significantly closer to the destination. The exponential term increases. When the neighbor is far from the destination, that is... hour, ; ii. Cost of Link Instability The model is based on the Earth's core. As the origin, the satellite and position vector and The only certainty; the geocentric angular size between the two stars Through vector dot product Quickly obtain; when the relative motion of the two stars causes The link is interrupted when the distance increases to the Earth's blocking boundary; the relative angular velocity is directly calculated using vector cross product. Thus, without iteration, through the formula Accurately predict the remaining time of the link interruption; Let the Earth's radius be... The orbital altitude is The minimum elevation angle is Maximum geocentric angle for maintaining connection between two stars for: Using satellite nodes Calculate the current geocentric angle using the dot product of position vectors. Combined with relative angular velocity Calculate the remaining link time : in , It is the angular velocity vector; Using an inverse proportional mapping, an extremely high penalty is imposed on links that are about to be disconnected: in To switch protection time, The cost is a time normalization constant. When the link is about to break, the cost is directly set to 1 to force route avoidance; when the link has a long lifespan, the cost approaches 0. iii. Node congestion potential cost We introduce the M / M / 1 queuing theory model, assuming that the arrival of messages at satellite nodes follows a Poisson distribution λ, the service time follows an exponential distribution μ, and the average delay of the nodes is... With system utilization Nonlinear relationship Simply using queue length cannot reflect the nonlinear deterioration trend of latency under high load; define node congestion potential. As a kind of "congestion risk probability": in This is the overall load factor; the formula ensures that when the load is close to 1, the cost is... A steep rise approaches 1, creating a "soft rejection" mechanism.

6. A routing method for supporting differentiated service requirements in satellite internet according to claim 5, characterized in that, Step 1.3 specifically includes: Construct a threshold-based filtering mechanism, where nodes traverse their neighbor lists. Remove neighboring nodes that do not meet the following constraints to form a valid candidate set. ; Link remaining lifetime constraint: If the remaining lifetime of a neighboring link is... Below the system's set protection threshold If the link is found to be on the edge of unavailability, it will be removed from the candidate set. Node load saturation constraint: If the overall load rate of a neighboring node is... Exceeding the saturation threshold This indicates that the node's cache or CPU has no available resources, and forcibly forwarding will result in packet loss; in this case, the node is directly excluded and will no longer participate in the subsequent weighted cost calculation.

7. A routing method for supporting differentiated service requirements in satellite internet according to claim 6, characterized in that, Step 2 specifically involves: By parsing the service type identifier in the packet header, it can be determined whether the current data flow is latency-sensitive or throughput-sensitive; at the same time, the routing cost function is defined as: The weights are adjusted in real time based on the current network congestion level; the weight vector... Define network background congestion The average load of neighboring nodes: .

8. A routing method for supporting differentiated service requirements in satellite internet according to claim 7, characterized in that, Weights that meet the QoS requirements of different service types are generated separately. The specific generation logic is as follows: Step 2.1: URLLC business weight generation function strategy logic: Dynamic weights are constructed using the hyperbolic tangent function: congestion weight γ and distance weight. and stability weights : in The congestion trigger threshold, This is the sensitivity coefficient. When At that time, γ remained at a low level; Set congestion trigger threshold Set the sensitivity coefficient This ensures that when the load exceeds the threshold, the weight γ can quickly saturate to its maximum value within 0.1 seconds; when the network load is below this value, the congestion weight γ remains low, and the routing strategy degenerates to shortest path priority; once the load exceeds the congestion trigger threshold, the curve shows a steep upward trend, quickly saturates, and forms a strong rejection of congested nodes. Step 2.2: eMBB business weight generation function strategy logic: eMBB services prioritize transmission continuity and are sensitive to the TCP window halving effect caused by route switching; their weighting is determined by the average remaining lifetime of local links. Driven by; Formula definition: Dynamic weights are constructed using an exponential decay function: When local links are generally unstable, the β weight increases significantly.

9. A routing method for supporting differentiated service requirements in satellite internet according to claim 8, characterized in that, Step 3 specifically involves: Based on the above perception and strategy, the routing decision-making process is divided into three stages: cost calculation, next hop selection and stability maintenance. Step 3.1: Route cost calculation: Step (1): The routing module parses the service type of the data packet to be forwarded; Step (2): Collect the current neighbor set The state of the local network background index is calculated. or ); Step (3): Adjust the corresponding policy function to generate the dynamic weight vector at the current time. ; Step (4): Traverse the neighbor list and calculate the comprehensive routing cost for each neighbor using the formula. ; Step 3.2: Next-hop selection mechanism: i. Search for the neighbor node that minimizes the overall cost using a greedy forwarding decision mechanism. : ii. Comparison The virtual cost of the current node itself ; (1) If Execute forwarding; (2) If the cost of all neighbors is higher than that of itself, the surrounding forwarding mode is triggered; the right-hand rule is used to circumvent the hole boundary in the planar local topology until a node that satisfies the greedy forwarding condition is found. Step 3.3: i. The principle of preemptive soft handover for URLLC: Step (1): The node continuously monitors the LRT of the current next-hop link; Step (2): When Routing does not require waiting for the link to break; it immediately calculates alternative next hops. At this point, the IP layer copies the packet, adding two different routing headers to the same data packet, and sends them to different destinations. and ; Step (3): The destination node maintains a "received sequence sliding window", checks whether the data packet is a duplicate data packet based on the sequence number flag, and encapsulates a unique "stream sequence number" in the extended field of the data packet header; Step (4): When the old node The connection was completely interrupted or At that time, stop sending Send, seamlessly restored to Single-path transmission; ii. Hysteresis offloading strategy for eMBB: For throughput-sensitive services, it is necessary to avoid the routing ping-pong effect caused by slight parameter fluctuations, and at the same time, utilize multipath to improve throughput; This is achieved through two methods: hysteresis switching and multipath shunting. (1) Delayed shunting: Only when the cost advantage of the new path exceeds the threshold. The routing table is only updated when the time is right. (2) Multipath diversion: If there exists a suboptimal node k, its cost is close to that of the optimal node ( If so, a traffic splitting group is established; the traffic allocation ratio is inversely proportional to the cost: Achieve bandwidth aggregation and load balancing.

Images