A double-layer satellite network anonymous routing method for latency-sensitive services

Abstract

The application discloses a kind of anonymous routing methods of double-layer satellite network for delay-sensitive service, it is applied to VLEO / LEO double-layer satellite network, this network is based on software-defined network SDN architecture implementation, including data plane, control plane and application plane;Data plane includes VLEO layer, VLEO / LEO cross layer and LEO layer;Application plane is deployed with the routing method of the application upper;The method comprises: constructing layered decoupling routing calculation framework, to divide routing decision into including LEO layer inner routing, VLEO / LEO cross layer routing and VLEO layer inner routing multilayer path selection;Using the collaborative computing capability of SDN controller and satellite node, multilayer path selection is reconfigured and weight disturbance in real time, generates optimal routing path.The application can effectively reduce end-to-end delay, balance network load, improve the real-time performance, robustness and security anonymity of double-layer satellite network.

Description

Technical Field

[0001] This invention belongs to the field of satellite network routing technology, specifically relating to an anonymous routing method for two-layer satellite networks oriented towards latency-sensitive services. Background Technology

[0002] With the rapid development of 6G communication technology and the construction of an integrated space-air-ground network system, the service capabilities of traditional terrestrial communication networks are extending into the depths of Low Earth Orbit (LEO) and Very Low Earth Orbit (VLEO) space. LEO and VLEO satellites, with their significant advantages such as wide coverage, low latency, and controllable deployment costs, have already begun to emerge in fields such as commercial communications, military missions, emergency disaster relief, and remote sensing. Especially in areas with weak ground infrastructure, such as oceans, mountains, and deserts, the dual-layer constellation architecture effectively overcomes geographical limitations and achieves full-area network coverage. However, the explosive growth of latency-sensitive services such as high-definition video, telemedicine, and remote industrial control poses stringent challenges to the real-time performance and reliability of communication systems.

[0003] While VLEO satellites (orbital altitude 250km-450km) possess the inherent advantage of short end-to-end propagation latency, their short orbital periods, frequent link switching, and susceptibility to cloud scattering and turbulence pose significant challenges to their stability. In contrast, LEO satellites (orbital altitude 500km-1200km) benefit from a thinner atmospheric environment, resulting in less interference and higher bandwidth for laser communication. Furthermore, their long on-orbit lifespan of 5-7 years makes them ideal relays and backups for VLEO systems. Therefore, constructing a two-layer collaborative architecture combining VLEO link microwave communication and LEO link laser communication has become a key technological approach to meeting the demands for low latency and high capacity.

[0004] In this architecture, routing design becomes the core bottleneck restricting system performance. On the one hand, satellite network topology exhibits highly dynamic characteristics, making it difficult to directly migrate traditional terrestrial routing algorithms, which can easily lead to path congestion and service interruptions. On the other hand, satellite links are subject to security risks such as being monitored and hijacked, and traditional static routing path mechanisms suffer from predictable paths and exposed links, failing to meet the anonymous transmission requirements in high-security scenarios.

[0005] Currently, existing routing technologies are mainly divided into two categories: traditional routing and intelligent routing. Traditional routing algorithms often employ topology virtualization or node virtualization strategies to transform dynamic satellite networks into static periodic networks to simplify computation. Although improved schemes based on SDN (Software-Defined Networking) architecture, such as multi-path routing and QoS-aware routing, have made progress in throughput and load balancing, they are essentially still static planning. These algorithms lack flexibility, struggle to perceive real-time conditions in satellite networks, and cannot effectively cope with complex network environments. They also find it difficult to dynamically adjust routing paths to fully utilize network resources and avoid congestion. Moreover, traditional routing algorithms do not adequately consider anonymity. Current research focuses primarily on optimizing QoS performance indicators such as latency, jitter, and bandwidth, lacking protection mechanisms for path predictability and link exposure risks, making it difficult to meet the needs of high-security services. To adapt to dynamic topologies, artificial intelligence technologies such as reinforcement learning and deep learning have been introduced into the routing field. These intelligent routing algorithms construct intelligent agent models, using network topology changes and link state information as input, to dynamically adjust and optimize routing paths in real time. However, existing intelligent routing algorithms typically rely on black-box models such as deep reinforcement learning, whose decision-making processes lack interpretability and are difficult to verify intuitively, making them unsuitable for troubleshooting, security auditing, and trusted deployment in high-security or mission-critical scenarios.

[0006] In summary, existing traditional routing algorithms are statically rigid and poorly adaptable dynamically; existing intelligent routing systems are mostly black-box models, with decisions that are uninterpretable and difficult to apply to high-security scenarios. Therefore, there is an urgent need for a two-layer satellite network routing method that balances low-latency transmission with high anonymity protection to overcome existing technological bottlenecks and ensure the efficient and secure operation of future integrated air-space-ground networks. Summary of the Invention

[0007] To address the aforementioned problems in existing technologies, this invention provides a two-layer anonymous routing method for time-sensitive services in satellite networks. The technical problem to be solved by this invention is achieved through the following technical solution: This invention proposes an anonymous routing method for a two-layer satellite network oriented towards latency-sensitive services. The method is applied to a VLEO / LEO two-layer satellite network, which is implemented based on a Software-Defined Networking (SDN) architecture, including a data plane, a control plane, and an application plane. The data plane includes the VLEO layer, VLEO / LEO cross-layer, and the LEO layer. The anonymous routing method of this invention is deployed on the application plane. Routing decisions are generated by running the anonymous routing method and forwarded to the data plane for execution via the control plane. The anonymous routing method includes: Based on the hierarchical structure of the VLEO / LEO dual-layer satellite network, a hierarchical decoupled routing computation framework is constructed. The hierarchical decoupled routing computation framework divides routing decisions into multi-layer path selection, including LEO intra-layer routing, VLEO / LEO cross-layer routing, and VLEO intra-layer routing. By leveraging the collaborative computing capabilities of the SDN controller and satellite nodes, multi-layer path selection is reconfigured and weighted in real time to generate the optimal routing path; Specifically, LEO layer routing is implemented using a K-shortest path algorithm and a weighted random selection routing decision mechanism, and communicates via laser inter-satellite links; VLEO / LEO cross-layer routing is implemented using a two-stage routing decision mechanism of weighted random selection and load-aware optimization, and communicates via VLISL cross-layer links that satisfy constraints of no Earth obstruction and maximum communication distance; VLEO layer routing is implemented using a hierarchical optimization, latency-aware, and load-balancing routing decision mechanism, and communicates via microwave inter-satellite links.

[0008] The beneficial effects of this invention are: This invention proposes an anonymous routing method for VLEO / LEO dual-layer satellite networks, implemented under a Software-Defined Networking (SDN) architecture. First, based on the layered structure of the VLEO / LEO dual-layer satellite network, a layered and decoupled routing computation framework is constructed to divide routing decisions into multi-level path selection, including intra-LEO layer routing, VLEO / LEO cross-layer routing, and intra-VLEO layer routing. Then, utilizing the collaborative computing capabilities of the SDN controller and satellite nodes, the multi-level path selection is reconfigured and weighted in real time to generate the optimal routing path. Specifically, intra-LEO layer routing employs a K-shortest path algorithm and a weighted random selection routing decision mechanism, communicated via laser inter-satellite links; VLEO / LEO cross-layer routing employs a two-stage routing decision mechanism of weighted random selection and load-aware optimization, communicated via VLISL cross-layer links that satisfy constraints of no Earth obstruction and maximum communication distance; intra-VLEO layer routing employs a layered optimization, latency-aware, and load-balancing routing decision mechanism, communicated via microwave inter-satellite links. Then, leveraging the collaborative computing capabilities of the SDN controller and satellite nodes, multi-layer path selection is reconfigured and weighted in real time to generate the optimal routing path. This method significantly improves routing performance and anonymity through differentiated decision-making mechanisms at each layer; moreover, the layered decoupled collaborative optimization, lightweight load balancing, and dynamic clustering topology construction mechanisms ensure large-scale network scalability and robustness while maintaining extremely low computational complexity. It successfully achieves a flexible trade-off between low-latency transmission and high anonymity protection for latency-sensitive services, making it suitable for high-security scenarios and highly dynamic satellite environments.

[0009] The present invention will be further described in detail below with reference to the accompanying drawings and embodiments. Attached Figure Description

[0010] Figure 1 A diagram of a VLEO / LEO dual-layer satellite network architecture based on a software-defined networking (SDN) architecture provided for embodiments of the present invention; Figure 2 A schematic diagram illustrating the maximum VLISL range provided in this embodiment of the invention; Figure 3 A flowchart illustrating an anonymous routing method for delay-sensitive services in a two-layer satellite network provided by an embodiment of the present invention; Figure 4 A flowchart illustrating the VLEO layer routing algorithm provided in an embodiment of the present invention; Figure 5 A schematic diagram illustrating the calculation process of VLEO layer latency-aware routing provided in an embodiment of the present invention; Figure 6 The theoretical experimental results of entropy and time delay using the diversified hierarchical algorithm of this invention are presented. Figure 7 The results are based on experimental results using the entropy and time delay theory of random stratification; Figure 8 These are experimental results of entropy and time delay theory under extreme conditions; Figure 9 The simulation results of entropy and time delay using the diversified hierarchical algorithm of this invention are presented. Figure 10 Theoretical experimental results on entropy and time delay under different relay satellite network scales; Figure 11 Entropy / delay with respect to parameters in CLAPS and AR-DLSN algorithms τ A diagram illustrating the changing relationships; Figure 12 A comparison of the running times of the CLAPS and AR-DLSN algorithms under different relay satellite network scales; Figure 13 This is a comparison chart showing the runtime ratio of CLAPS compared to two equalization strategies under different relay satellite network sizes. Detailed Implementation

[0011] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the embodiments described below are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0012] The first aspect of this invention provides an anonymous routing method for a two-layer satellite network oriented towards latency-sensitive services. This method is applied to a VLEO / LEO two-layer satellite network, wherein the VLEO / LEO two-layer satellite network is implemented based on a software-defined networking (SDN) architecture, including a data plane, a control plane, and an application plane. The data plane includes a VLEO layer, a VLEO / LEO cross-layer, and an LEO layer. The anonymous routing method of this invention is deployed on the application plane. Routing decisions are generated by running this anonymous routing method and forwarded to the data plane for execution via the control plane.

[0013] Specifically, for a two-layer satellite routing scenario consisting of a VLEO satellite constellation and a LEO satellite constellation, VLEO satellites transmit data directly or via satellite network relay. Therefore, the goal in this routing scenario is to find the optimal path in the VLEO network to ensure high anonymity and low latency in data transmission while saving satellite resources.

[0014] Furthermore, this invention implements a VLEO / LEO dual-layer satellite network system through a Software-Defined Networking (SDN) architecture, thereby decoupling the data plane and control plane, making the routing algorithm more flexible and programmable. See also... Figure 1 , Figure 1 This diagram illustrates a VLEO / LEO two-layer satellite network architecture implemented based on a Software-Defined Networking (SDN) architecture, as provided in this embodiment of the invention. Under this architecture, the satellite network system is mainly divided into a data plane, a control plane, and an application plane.

[0015] The data plane consists of the VLEO layer, the VLEO / LEO cross-layer, and the LEO layer, responsible for the transmission and forwarding of actual data. The LEO layer uses laser inter-satellite links for communication, the VLEO / LEO cross-layer uses VLISL (VELO-LEO Inter-Satellite Link) cross-layer link communication to meet the constraints of no Earth obstruction and maximum communication distance, and the VLEO layer uses microwave inter-satellite links. When a direct link exists between two VLEO satellites, data can be transmitted directly; otherwise, it is relayed via LEO satellites.

[0016] The control plane operates between the ground control center and medium-to-high orbit satellites, acting as a bridge between the application plane and the data plane. The control plane continuously collects network status and service requirements from the data plane and feeds the results back to the application plane. Simultaneously, based on the routing decisions of the application plane, it issues and updates the forwarding rules of the data plane in real time.

[0017] The application plane deploys the Anonymous Routing Algorithm for Dual-Layer VLEO / LEO Satellite Networks (AR-DLSN) proposed in this invention. This algorithm calculates the optimal routing path through comprehensive analysis of real-time link status, service flow characteristics, and satellite distribution. When a VLEO satellite needs to transmit data, the switch on the source VLEO satellite initiates a routing request to the ground control center or the remote controller of the MEO / GEO. After synchronizing with the application plane, the controller maintains a global topology snapshot, continuously monitors link changes, and calls the AR-DLSN algorithm to generate the optimal path. Subsequently, the control plane writes the calculation results into the forwarding tables of the VLEO and LEO satellites. Finally, the source VLEO satellite selects a direct path or a relay path via an LEO satellite based on the forwarding table information, reliably delivering the data packet to the target satellite.

[0018] The VLEO / LEO dual-layer satellite network of this invention will be described in detail below.

[0019] In this embodiment, both VLEO and LEO satellites adopt a polar orbit constellation configuration. The VLEO satellite constellation has 25 polar orbital planes, each accommodating 25 VLEO satellites operating at an altitude of approximately 410 kilometers in ultra-low Earth orbit. The orbital inclination is 85°. The LEO satellite constellation has 8 polar orbital planes, each accommodating 9 LEO satellites operating at an altitude of approximately 550 kilometers in low Earth orbit. The orbital inclination is 90.5°.

[0020] Optionally, in this embodiment, the VLEO / LEO satellite constellation adopts the classic Walker constellation configuration, characterized by all satellites operating in near-circular orbits with the same altitude and inclination. Specifically, the orbital planes are uniformly distributed along the equator, and the satellites in each plane are arranged at equal intervals. The Walker constellation can be formally represented as... ,in , , and These represent the total number of satellites, the number of orbital planes, the orbital altitude, and the orbital inclination, respectively. For phase factor, The number of satellites in each orbital plane is The phase difference between adjacent orbital planes is calculated as follows: ; Suppose that the VLEO / LEO two-layer satellite network is composed of an undirected graph. The system represents the set of satellite nodes and the set of edges, respectively. VLEO satellites and LEO satellites, thus satisfying the complete set of satellite nodes. and For ease of representation, VLEO / LEO satellites will be numbered as follows: ,make Therefore, the set of satellite nodes Based on this, the VLEO layer and LEO layer satellite networks can be represented as two independent topological subgraphs: the VLEO layer network topology is represented as a subgraph. ,in This represents the set of microwave inter-satellite links between all VLEO satellites. The LEO layer network topology is represented as a subgraph. ,in This represents the set of laser inter-satellite links between all LEO satellites. Additionally, the set of VLEO-LEO inter-satellite links (VLISL) is denoted as... It consists of connections between VLEO and LEO satellites that satisfy visibility and distance constraints. In summary, the entire network communication link set satisfies... Because satellites within the same orbital plane maintain the same relative velocity and direction, their intra-orbital inter-satellite links (ISLs) exhibit extremely high stability. In contrast, inter-orbital ISLs are affected by the relative motion between satellites, exhibiting significant spatiotemporal dynamic characteristics.

[0021] Furthermore, in this embodiment, the inter-satellite link design follows the "four-link" principle of the Iridium system: each satellite establishes inter-satellite links with four adjacent satellites, two of which are intra-orbit links between adjacent satellites in the same orbit, and the other two are inter-orbit links between adjacent orbital planes. On both sides of the polar orbit constellation, the satellites move in directions from south to north (ascending) and from north to south (descending), respectively, resulting in a "reverse gap" between adjacent orbits with opposite directions of motion. Satellites located on both sides of the reverse gap cannot establish cross-gap links, therefore, the maximum number of usable inter-satellite links is three.

[0022] In addition, to reduce communication complexity and avoid antenna tracking loss, this embodiment defines the region with latitude exceeding a preset value as the polar region. When the satellite enters this region, all inter-orbit links are closed due to the excessive relative speed. After leaving the polar region, the inter-orbit links are automatically re-established.

[0023] Optionally, this implementation may define areas exceeding 70°N and 70°S as polar regions.

[0024] Furthermore, considering the time-varying characteristics of satellite constellations, this embodiment employs a time spread graph method to discretize the network topology, dividing the continuous time axis into a fixed-interval time slice sequence, and in each time slice... The network topology and resource configuration remain static, and subsequent analysis and research are carried out within a single time slice.

[0025] It should be noted that in the VLEO / LEO dual-layer satellite network architecture, the core of topology construction lies in establishing the VLISL cross-layer link between the VLEO and LEO layers. In this embodiment, the maximum communication distance of the VLISL cross-layer link is calculated jointly by the Earth's radius, VLEO orbital altitude, LEO orbital altitude, and atmospheric altitude, and both communicating satellites must be on the same side of the Earth with no line of sight obstructed by the Earth.

[0026] Please see Figure 2 , Figure 2 This is a schematic diagram illustrating the maximum VLISL range provided in an embodiment of the present invention. Figure 2 As shown, the effective establishment of VLISL requires the simultaneous fulfillment of two key conditions: (1) No Earth obstruction condition. The satellites of both communicating parties must be located on the same side of the Earth, and the line-of-sight (LoS) between them must not be blocked by the Earth. For this purpose, visibility can be accurately determined based on a spherical geometric model. By analyzing the Earth's radius, the orbital altitudes of the two satellites, and their relative geometric positions, it can be determined whether the LoS passes through the Earth's interior. (2) Distance constraint condition. The relative distance between the communicating satellites should be less than the maximum communicable distance. x Otherwise, link attenuation and latency will cause the signal to be unable to maintain a stable connection.

[0027] The maximum communicable distance of VLISL can theoretically be derived from visibility constraints, and its expression is as follows: ; in, r For the Earth's radius, , These are the orbital altitudes of the VLEO and LEO satellites, respectively. a This represents the atmospheric height corresponding to the visible boundary (usually taken as 80 km). If we take... , , , Therefore, the maximum VLISL range is approximately 4599 km.

[0028] The main objective of this invention is to develop a latency-aware routing strategy for VLEO / LEO dual-layer satellite networks that provides a good trade-off between anonymity and end-to-end latency, defining a parameter. τ (0 ≤ τ≤ 1) to adjust this trade-off, where τ = 0 indicates that latency has been optimized (reduced) to the greatest extent. τ = 1 indicates that no route optimization is performed to limit latency, which corresponds to a route selection strategy of random uniformity.

[0029] When assessing anonymity, the primary consideration is the attacker's visibility across all satellite network links. Specifically, an attacker can observe all messages transmitted between entities within the network and infer the probabilistic relationships between input and output messages. To quantify message anonymity, the concept of entropy is used, measuring anonymity by calculating the entropy value of the probability distribution of a message at one end of the communication with all possible corresponding messages at the other end.

[0030] The end-to-end latency of a VLEO / LEO two-layer satellite network typically consists of five parts: hybrid latency, processing latency, transmission latency, propagation latency, and queuing latency. Assume the message path passes through... L ( L 2) If there are 2 relay satellites, then the average end-to-end delay is expressed as: ; ; in, This is the average time (hybrid delay) for each relay satellite to reorder messages, introduced for anonymity. This is the time (processing delay) for the relay satellite and the destination satellite to process the message; It is the time required for the source satellite and relay satellite to send messages (transmission delay); This is the average time (propagation delay) for a message to travel one hop between adjacent satellites in the network. Since the AR-DLSN algorithm explicitly introduces load balancing in the VLEO layer routing decision, queuing delay is negligible in this scenario. As the formula shows, the components of end-to-end delay are coupled to varying degrees with communication anonymity, resulting in a mixed delay. It directly affects anonymity. The increased size allows relay satellites to perform more complex message sorting for each relay, thereby disrupting message transmission characteristics, preventing fingerprint attacks, and providing stronger anonymity; and by using efficient encryption algorithms, processing latency is compressed as much as possible. To minimize its impact on overall latency; in contrast, transmission latency and propagation delay It is not directly related to anonymity, therefore it can be shortened. and Composition To reduce average end-to-end latency.

[0031] This can be further broken down into the latency from the message-sending satellite to the relay satellite network. Total latency within ASN Delay from ASN to the final receiving satellite ,Right now: ; Among them, the latency of sending satellites to the relay satellite network and the latency from the relay satellite network to the receiving satellite The latency is primarily affected by the physical locations of the transmitting and receiving satellites within the global network, while the total latency within the relay satellite network is... This primarily reflects the structure and link characteristics of the relay satellite network itself, which are, on average, the same for all clients. Specifically, the propagation delay within the relay satellite network depends on the routing scheme used and whether propagation delay factors are fully considered in path selection; the transmission delay also depends on the routing scheme used, link bandwidth, message size, and other factors, directly affecting the final transmission efficiency.

[0032] Based on the above analysis, the design goal of the AR-DLSN algorithm proposed in this invention is to define a routing strategy biased towards low latency, collaboratively optimize the propagation latency and transmission latency within the two-layer satellite network, and minimize the average total latency within the relay satellite network while ensuring anonymity. This reduces end-to-end latency and improves overall transmission performance.

[0033] To achieve an effective trade-off between communication anonymity and latency performance, this invention starts from the hierarchical structure of a two-layer satellite network and divides the routing decision-making process into three mutually cooperating components: LEO layer intra-layer routing, VLEO / LEO cross-layer routing, and VLEO layer intra-layer routing. By performing path selection and collaborative optimization at each level of the multi-orbit satellite network, an anonymity routing algorithm, AR-DLSN, suitable for VLEO / LEO two-layer satellite network architecture, is designed.

[0034] Please see Figure 3 , Figure 3 This is a flowchart illustrating a two-layer satellite network anonymous routing method for latency-sensitive services provided in an embodiment of the present invention. The two-layer satellite network anonymous routing method mainly includes the following two steps: Step 1: Based on the hierarchical structure of the VLEO / LEO dual-layer satellite network, construct a hierarchical decoupled routing calculation framework. The hierarchical decoupled routing calculation framework divides routing decisions into multi-layer path selection, including LEO intra-layer routing, VLEO / LEO cross-layer routing, and VLEO intra-layer routing. Step 2: Utilize the collaborative computing capabilities of the SDN controller and satellite nodes to reconfigure and weight the multi-layer path selection in real time, generating the optimal routing path; Specifically, LEO layer routing is implemented using a K-shortest path algorithm and a weighted random selection routing decision mechanism, and communicates via laser inter-satellite links; VLEO / LEO cross-layer routing is implemented using a two-stage routing decision mechanism of weighted random selection and load-aware optimization, and communicates via VLISL cross-layer links that satisfy constraints of no Earth obstruction and maximum communication distance; VLEO layer routing is implemented using a hierarchical optimization, latency-aware, and load-balancing routing decision mechanism, and communicates via microwave inter-satellite links.

[0035] The routing decisions for each layer will be described in detail below.

[0036] First, for routing within the LEO layer, it primarily employs the K-Shortest Paths (KSP) algorithm and a weighted random selection routing decision mechanism, specifically including: S101: Calculate multiple optimal candidate paths for the source and destination satellite nodes based on the KSP algorithm.

[0037] Specifically, the existing KSP algorithm can be used to calculate the source satellite node and destination satellite node before... The optimal candidate path is determined. Detailed algorithm implementation can be found in existing technologies; this embodiment will not provide a detailed description.

[0038] S102: Assign weights to each optimal candidate path and calculate the selection probability for each optimal candidate path based on a weighted random mechanism; wherein, the weight of each optimal candidate path is negatively correlated with the path length.

[0039] Here, this embodiment proposes an adaptive probability allocation mechanism that comprehensively considers multi-dimensional network state indicators. Specifically, for each candidate path... k ( The probability of selection is calculated as follows: ; in, Indicates the first k The probability that the shortest path is selected. Indicates the first k The weights corresponding to the shortest paths are negatively correlated with the path length; that is, the longer the path, the smaller its corresponding weight.

[0040] Secondly, for VLEO / LEO cross-layer routing, a two-stage routing decision mechanism of weighted random selection and load-aware optimization is mainly adopted. This mechanism achieves intelligent selection of cross-layer satellites through mathematical modeling, and comprehensively optimizes link diversity, load balancing performance, and transmission anonymity in the process, specifically including: S201: Weighted random selection: The initial forwarding probability is calculated based on the Euclidean distance between VLEO satellites and LEO satellites, and the forwarding probability is used to pre-select the next-hop LEO satellite for VLEO satellites, thereby realizing weighted random selection of satellites across layers.

[0041] Specifically, let's set VLEO satellite u With LEO satellite v Let the Euclidean distance between them be... VLEO satellite u If a set of LEO satellites for VLISL can be established, then for VLEO satellites... u To LEO satellite The initial forwarding probability is defined as: ; According to this definition, the closer two satellites are, the higher the probability of relaying. The AR-DLSN algorithm relies on probability. Pre-selecting the next-hop LEO satellite for VLEO satellites enhances the stealth of the path by introducing randomness.

[0042] S202: Load Awareness Optimization: Based on the weighted random selection results of cross-layer satellites, the node degree of LEO satellites is statistically analyzed to sense the load of LEO satellites. The load of LEO satellites and the Euclidean distance between VLEO satellites and LEO satellites are optimized to obtain the final probability of VLEO satellites selecting LEO satellites as cross-layer relay nodes.

[0043] In this embodiment, a binary indicator variable is introduced. l uv VLEO satellite u With LEO satellite v The connectivity, when l uv When =1, it indicates that the two are within communication range and a VLISL can be established; when l uv When =0, VLISL cannot be established.

[0044] Firstly, based on binary indicator variables l uv Based on the selection results of the first phase, the node degree of LEO satellites is calculated: ; in, Indicates LEO satellite v The degree of nodality, that is, LEO satellites v The number of VLEO satellites currently pre-connected.

[0045] The load on the LEO satellite is then optimized to obtain the final probability that the VLEO satellite will select an LEO satellite as a cross-layer relay node, specifically including: a) Normalize the node degree of LEO satellites and the Euclidean distance between VLEO satellites and LEO satellites to obtain normalized node degree and Euclidean distance.

[0046] To eliminate the influence of dimensions, the Euclidean distance is... and node degree Normalization is performed: ; ; In the formula, This represents the normalized VLEO satellite. u With LEO satellite v The Euclidean distance between them and These represent the maximum and minimum Euclidean distances, respectively. These represent the normalized LEO satellites. v degree of nodes, and These represent the maximum node degree and the minimum node degree, respectively.

[0047] b) Calculate the distance weight and load weight based on the normalized node degree and Euclidean distance, respectively. The calculation formula is as follows: ; ; In the formula, Indicates distance weight, This indicates the load weight.

[0048] It can be seen that the closer the distance between VELO and LEO satellites, the better. The higher the weight, the fewer LEO satellites are connected to VLEO satellites. The higher the weight.

[0049] c) Introduce adjustment factors based on distance weight and load weight. To balance distance and load, by increasing The value is used to increase the distance weight, thereby prioritizing LEO satellites with closer physical distances; by decreasing... The value is used to reduce the distance weight, thereby prioritizing the selection of LEO satellites with lower loads, and thus obtaining the final probability of a VLEO satellite selecting an LEO satellite as a cross-layer relay node.

[0050] Specifically, after introducing the adjustment factor, the total weight is expressed as: ; It can be seen that when When the algorithm is close, it tends to select LEO satellites that are physically closer; when At that time, the focus is more on selecting LEO satellites with lower payloads.

[0051] Ultimately, the VLEO satellite u Select LEO satellite v The probability of performing cross-layer message transmission is defined as follows: ; in, This indicates that, after comprehensively considering factors such as distance and payload, VLEO satellites... u Select LEO satellite v The final probability of being a cross-layer forwarding node.

[0052] Finally, for routing within the VLEO layer, it primarily employs a routing decision mechanism that combines hierarchical optimization, latency awareness, and load balancing. Please refer to [link to relevant documentation]. Figure 4 , Figure 4 The flowchart of the VLEO layer routing algorithm provided in this embodiment of the invention mainly includes three parts: hierarchical satellite network topology construction, latency-aware routing, and network load balancing, corresponding to the following steps S301 to S303.

[0053] S301: Hierarchical Network Topology Construction: The K-medoids clustering algorithm is used to complete the clustering based on the satellite's geographical location, and the relay satellites are divided into a multi-layer structure according to the distribution of cluster centers to construct a hierarchical network topology with geographical diversity.

[0054] First, based on the random selection N The geographic locations of each VLEO relay satellite are clustered, and the geographic locations are represented by Cartesian coordinates, which are then used as input to provide... K- The medoids clustering algorithm is used to obtain... K Each cluster and the location of its cluster center.

[0055] in, K- The medoids algorithm iteratively optimizes the cluster center positions to minimize the total distance from each internal node to the cluster center, ultimately resulting in cluster centers and node assignments that guarantee good clustering performance. First, random selection... KOne relay satellite is used as the initial cluster center to begin the clustering process. For each relay satellite, the distance from that node to each cluster center is calculated. The node is then assigned to the cluster containing the nearest cluster center, and its cluster label is recorded. For each cluster, each node within that cluster is considered as a new candidate cluster center. For each candidate cluster center, the sum of the distances from all nodes within that cluster to the candidate cluster center is calculated, i.e., the cost. The node with the minimum cost is selected as the new cluster center, minimizing the distances from all nodes within the cluster to the cluster center, thus optimizing the clustering results. This process is repeated until the currently selected node is reached. K The cluster centers remain unchanged. Once the algorithm converges, the final output is... K The coordinates of each cluster center and the cluster assignment label for each relay satellite.

[0056] Then, a diversified hierarchical algorithm is used to arrange the relay satellite layers, and each relay satellite is assigned to a different layer according to the cluster information.

[0057] Assuming the message path passes through L One relay satellite, therefore N The network composed of relay satellites is divided into... L Layers, each layer includes W There are several relay satellites to choose from, and the different geographical locations of these relay satellites result in geographical diversity (i.e., cluster diversity) at each layer. The specific algorithm depends on... K and W Divided into W<K、W> K, W=K Three scenarios.

[0058] Based on the cluster center distribution, relay satellites are divided into a multi-layered structure, including: A) When W < K At that time, clusters are randomly selected for each layer according to the principle of the farthest cluster center, and relay satellites are randomly selected from the clusters; the above process is repeated until all relay satellites are assigned to the corresponding layers.

[0059] Specifically, when W < K For the first relay satellite selection, a cluster is randomly selected, and a relay satellite from that cluster is randomly selected as the first relay satellite in the first layer. For the second relay satellite selection, the distance between the selected cluster center and other cluster center locations is calculated, the cluster with the farthest distance is determined, and a relay satellite from this cluster is randomly selected as the second relay satellite in the second layer. For the third relay satellite selection, the third cluster with the farthest distance from the first two cluster centers is determined, and a relay satellite from this cluster is randomly selected as the third relay satellite in the third layer. This process continues until... WAll relay satellites are selected, and the above process is repeated until all relay satellites are assigned to the corresponding layer (the selection of relay satellites is not repeated during this process).

[0060] B) When W>K At that time, according to the cluster size ratio, one relay satellite is randomly selected from each cluster as part of the members of that layer, and the remaining relay satellites are randomly selected from each cluster according to the cluster size ratio; repeat the above process until all relay satellites are assigned to the corresponding layer.

[0061] Understandably, the relay satellite selection process does not involve duplication.

[0062] C) When W=K At that time, a relay satellite is randomly selected from each cluster as a member of that layer; the above process is repeated until all relay satellites are assigned to the corresponding layer, during which the selection of relay satellites is not repeated.

[0063] S302: Delay-Aware Routing: Based on a hierarchical network topology, a balancing parameter τ, τ∈[0,1], is introduced to balance end-to-end delay and anonymity in the relay satellite selection process. When τ=0, it means that the goal is to minimize the average total delay of the relay satellite network, and the relay satellite with the lowest delay is selected first, so that the average total delay of the network is optimized to the greatest extent. When τ=1, it means that the goal is to enhance anonymity, and the relay satellite is randomly selected to maximize the anonymity of the network.

[0064] Specifically, in a two-layer satellite network, the average end-to-end latency of messages is affected not only by the location of relay satellites in the transmission link, but also by the relay satellite selection strategy, which directly impacts network anonymity. To optimize transmission performance, routing typically tends to select relay satellites with lower latency, making message paths more predictable and thus reducing anonymity. Conversely, increasing the randomness of relay satellite selection, while helping to improve anonymity, may lead to unmet link bandwidth constraints, resulting in network congestion and deteriorating overall transmission performance. Therefore, we introduce parameters... τ (0 ≤ τ ≤1), used to balance latency and anonymity in the relay satellite selection process. When τ When =0, it indicates that the goal is to minimize the average total latency of the relay satellite network, prioritizing the relay satellite with the lowest latency to achieve the desired network average total latency. Optimized to the maximum extent; when τ When =1, it indicates that the goal is entirely to enhance anonymity, and relay satellites are randomly selected to maximize the anonymity of the network.

[0065] The same strategy will be followed when selecting relay satellites for all paths in the satellite network, when randomly selected... After the relay satellite of the layer, subsequent The selection of relay satellites for a layer is determined by parameters. τ The degree to which the decision leans towards selecting a low-latency link.

[0066] Please see Figure 5 , Figure 5 This is a schematic diagram illustrating the calculation process of VLEO layer latency-aware routing provided in an embodiment of the present invention.

[0067] set up It was randomly selected as the first relay satellite. for Relay satellites in the next layer The total latency between. (Through) For relay satellites Sort the data and define a function. To indicate the sorted relay satellites The relay satellites with the lowest total latency, ranked by location, are: Conversely, the relay satellites with the highest total latency are ranked as follows: Select the next layer relay satellite. probability The calculation is as follows: .

[0068] S301: Network Load Balancing: At the VLEO layer, for scenarios where τ<1, a load balancing mechanism is adopted, which achieves traffic balance by adjusting the relay satellite routing weight.

[0069] Specifically, for τ In the case where <1, according to probability... Selecting relay satellites can overload some relay satellites in layers other than the first layer, meaning that the ability to reach the first layer cannot be guaranteed. Total traffic of routing in the layer A persistent imbalance in network load can easily lead to bottleneck relay satellites, which can severely degrade network performance. Therefore, it is necessary to adjust the selected relay satellites to balance network traffic.

[0070] For ease of representation, the probability of selecting the next-layer relay satellite will be... Represented as , Composition size is scattering matrix : ; The core objective of load balancing is to optimize the scattering matrix. In the experiments of this invention, all VLEO relay satellites had the same bandwidth, therefore bandwidth normalization was performed. This was done while prioritizing low-latency routing paths and simultaneously satisfying a balance condition. ,as well as and This achieves balanced traffic distribution across relay satellites. Nevertheless, the scattering matrix still prioritizes low-latency paths to ensure the optimization objective is achieved.

[0071] Optionally, as an implementation method, this invention proposes two lightweight load balancing algorithms: one is naive load balancing, and the other is greedy load balancing.

[0072] The naive load balancing algorithm primarily achieves load balancing through a single weight adjustment and load redistribution operation, mainly including: Based on the scattering matrix columns and the calculation of the load of each relay satellite, the relay satellites are divided into underloaded relay satellites and overloaded relay satellites according to a preset threshold. Reduce the routing weight of overloaded relay satellites and calculate the remaining load released; Based on the shortfall ratio of underloaded relay satellites, the remaining load will be redistributed to underloaded relay satellites.

[0073] Specifically, firstly, based on the current scattering matrix, calculate the first... l The load status of each relay satellite in the layer. The load of relay satellites can be determined by analyzing the scattering matrix. The summation of the corresponding column elements yields the value, reflecting the amount of routing traffic handled by the relay satellite. Based on a set load threshold, the system categorizes relay satellites into two types: those with insufficient load. One type is the relay satellite, meaning the payload is below the target level; the other type is the overloaded relay satellite. The load balancing algorithm prioritizes overloaded relay satellites, reducing their load by appropriately decreasing their routing weights and calculating the released "remaining load." Then, based on the load gap of each underloaded relay satellite, the remaining load is proportionally redistributed to these satellites, achieving a more reasonable load distribution across the entire network. This process is known as the "naive load balancing" strategy, characterized by performing only one weight adjustment and load redistribution operation, resulting in high computational efficiency. It is suitable for scenarios with relatively smooth load fluctuations or where the initial load already has a certain degree of balance.

[0074] The greedy load balancing algorithm mainly identifies overloaded relay satellites and reduces their routing weights through multiple rounds of iteration, while increasing the weights of underloaded relay satellites. In each round, the weights are recalculated based on the latest load distribution, and the remaining load is distributed proportionally to gradually achieve load migration.

[0075] Specifically, the greedy load balancing strategy is similar in basic idea to the naive load balancing strategy described above, but its core difference lies in its more refined and progressive load adjustment strategy. Specifically, greedy load balancing continuously identifies overloaded relay satellites in the network and dynamically reduces their routing weights to proactively reduce their load. Simultaneously, for underloaded relay satellites, their weights are increased accordingly, gradually guiding more traffic to these satellites. In each iteration, greedy load balancing recalculates the relay satellite weights based on the latest relay satellite load distribution and proportionally redistributes the released remaining load to underloaded relay satellites. Through multiple iterations, this process effectively migrates load from overloaded relay satellites to underloaded relay satellites, ultimately achieving global load balancing across all layers of the network. In contrast, the naive load balancing method only performs one weight adjustment and load redistribution operation. Although computationally simple and convergent quickly, its effectiveness may be limited when facing scenarios with highly uneven load distribution or significant network dynamics. Greedy load balancing, due to its multi-iterative optimization mechanism, typically requires... n It may take several or more iterations to reach the ideal balance, but it can significantly improve the load balancing accuracy and overall stability of the network.

[0076] Furthermore, this load balancing algorithm can be easily extended to support VLEO relay satellites with varying bandwidths. The main change will be in the layout of the VLEO relay satellite network, ensuring that the diversity of satellite nodes does not lead to significant differences in bandwidth between different tiers. This can be achieved by selecting relay satellites from different clusters based on their bandwidth and introducing constraints on the total bandwidth of each tier of satellite nodes. Bandwidth variations will not affect the routing algorithm. The load balancing algorithm can also be conveniently applied to relay satellites supporting variable bandwidth; simply define their total load based on their bandwidth and determine accordingly whether a satellite is overloaded or underloaded. When extending the balancing strategies (including greedy balancing and naive balancing) to support general satellite bandwidth, we consider the individual bandwidth of the relay satellites and mark them as overloaded or underloaded based on this. In the balancing strategy, when marking overloaded relay satellites, for all nodes... j We will modify the conditions to Instead of the original Similarly, we will update the flags for underloaded and balanced relay satellites. When allocating load from overloaded nodes to underloaded nodes, we will normalize the load based on the relay satellite bandwidth.

[0077] In summary, this invention proposes an anonymous routing method for VLEO / LEO dual-layer satellite networks, implemented under a Software-Defined Networking (SDN) architecture. First, based on the layered structure of the VLEO / LEO dual-layer satellite network, a layered and decoupled routing computation framework is constructed to divide routing decisions into multi-level path selection, including intra-LEO layer routing, VLEO / LEO cross-layer routing, and intra-VLEO layer routing. Then, utilizing the collaborative computing capabilities of the SDN controller and satellite nodes, the multi-level path selection is reconfigured and weighted in real time to generate the optimal routing path. Specifically, intra-LEO layer routing employs a K-shortest path algorithm and a weighted random selection routing decision mechanism, communicated via laser inter-satellite links; VLEO / LEO cross-layer routing employs a two-stage routing decision mechanism of weighted random selection and load-aware optimization, communicated via VLISL cross-layer links that satisfy constraints of no Earth obstruction and maximum communication distance; intra-VLEO layer routing employs a layered optimization, latency-aware, and load-balancing routing decision mechanism, communicated via microwave inter-satellite links. Then, leveraging the collaborative computing capabilities of the SDN controller and satellite nodes, multi-layer path selection is reconfigured and weighted in real time to generate the optimal routing path. This method significantly improves routing performance and anonymity through differentiated decision-making mechanisms at each layer; moreover, the layered decoupled collaborative optimization, lightweight load balancing, and dynamic clustering topology construction mechanisms ensure large-scale network scalability and robustness while maintaining extremely low computational complexity. It successfully achieves a flexible trade-off between low-latency transmission and high anonymity protection for latency-sensitive services, making it suitable for high-security scenarios and highly dynamic satellite environments.

[0078] The effects of the present invention will be further illustrated below through simulation experiments.

[0079] I. Simulation Conditions This invention establishes a VLEO / LEO dual-layer satellite network scenario using STK and MATLAB, and employs a COM interface to achieve seamless integration between the two platforms. The AR-DLSN routing algorithm is implemented using PyCharm Professional 2025.2.3, and statistical analysis and result evaluation of the relevant simulation data are completed. All experiments are conducted on the same computing platform to ensure the fairness of the results: a 12th Gen Intel® Core™ i5-12400F (2.50 GHz) processor, 32 GB of memory, and a 64-bit Windows / x64 architecture. The methods compared in the experiments are as follows: Low latency priority routing algorithm, without considering load balancing strategy; The randomized hierarchical algorithm considers the case where the clustered relay satellites are randomly assigned to each layer. In extreme cases, relay satellites located in the same cluster are placed in the same layer, resulting in all available routing links being high-latency links; The CLAPS (Client-Location-Aware Path Selection) algorithm was originally designed to optimize path selection in the Tor network. Its core idea is to generate optimal route weights under given constraints by solving a linear programming problem to achieve a specific optimization objective.

[0080] II. Simulation Content and Result Analysis According to a specific embodiment of the present invention, the performance of the routing algorithm designed in the present invention was evaluated, and the low-latency priority routing algorithm of the present invention was compared with the performance indicators of the random hierarchical algorithm and extreme cases.

[0081] Experiment 1: Theoretical experiments on entropy and time delay were conducted using the diversified layering algorithm of this invention, the existing random layering algorithm, and in extreme cases. The results are as follows: Figure 6-8 As shown. Among them, Figure 6 The theoretical and experimental results of entropy and time delay using the diversified hierarchical algorithms of this invention are presented. Figure 7 To utilize experimental results based on the entropy and time delay theory of random hierarchical structures, Figure 8 These are experimental results of entropy and time delay theory under extreme conditions. Figure 6-8 Figure (a) in the diagram represents entropy and randomness. τ The relationship between time delay and randomness is shown in Figures (b). τ The relationship.

[0082] It can be seen that when diverse layering algorithms are used to generate layered network structures ( Figure 6 When ), its performance is comparable to that of a scenario where relay satellites are placed completely randomly ( Figure 7 They are almost identical in terms of anonymity and latency. This is due to the settings at each layer. W With 128 relay satellites, random placement is sufficient to provide adequate low-latency links, allowing routing strategies to perform well in terms of latency optimization. However, diversified hierarchical algorithms still have practical value, helping to avoid certain extreme cases ( Figure 8 Although the probability of such situations is low, they can lead to significant performance degradation when they occur. In extreme cases, the entropy of the three routing strategies only decreases slightly compared to the diversified and random structures, but their latency increases significantly, as expected. When comparing the diversified layering algorithm with the random algorithm and the extreme cases, we conclude that the latency is significantly higher in the extreme cases, while the random layering algorithm exhibits slightly higher latency than the diversified layering algorithm, further validating the advantages of the diversified layering algorithm in maintaining low latency and high anonymity. It can also be observed from the figure that both entropy and latency increase with the randomness parameter. τIt rises with the increase of [something]. When [something] increases. τ When =1, all routing strategies tend to be perfectly uniform, forming the maximum entropy value (for W With a relay satellite count of 128, the maximum entropy is log2128 = 7 bits, but no further latency optimization is performed at this point. In the other extreme case, i.e. τ When = 0, the non-load balancing strategy (red curve) achieves the most significant latency optimization, with latency compared to τ When the entropy is 1, the fully uniform route reduces latency by approximately 68%. However, this optimization comes at the cost of complete anonymity; the entropy becomes 0, meaning that once the first relay satellite's entry satellite is known, its final relay satellite's exit satellite can be uniquely determined, making the routing path completely predictable. In contrast, load balancing strategies, while sacrificing some latency optimization capabilities, significantly increase uncertainty in the routing process and enhance anonymity. Specifically, the naive load balancing strategy (purple curve) in... τ The delay when = 0 is compared to τ When the value is 1, the latency is reduced by approximately 41%. Although the optimization is relatively small, its entropy value still reaches 4.2 bits, about 2.8 bits less than that of completely random routing, providing good anonymity protection. The greedy load balancing strategy (blue curve) falls between the two in terms of performance. τ The time delay when =1, in τ When the value is 0, approximately 55% latency optimization is achieved while retaining about 2.6 bits of entropy, compared to a loss of 4.4 bits of anonymity compared to completely random routing. This demonstrates that the load balancing strategy can achieve a good trade-off between anonymity and latency by significantly reducing latency while sacrificing a small amount of anonymity.

[0083] Experiment 2: Entropy and time delay simulation experiments were conducted using the diversified hierarchical algorithm of this invention. The results are as follows: Figure 9 As shown in the figure, this diagram visually illustrates different randomness parameters. τ The entropy and latency performance under routing strategies. Figure (a) shows the relationship between entropy and randomness. τ The relationship between time delay and randomness is shown in Figures (b). τ The simulation samples at the granularity of a single message, thus obtaining not only the expected values ​​of entropy and delay, but also their complete value distribution. In each set of random parameters... τ Combined with routing strategies (naive load balancing, greedy load balancing, and no load balancing), the results are presented as box plots: the boxes represent the 25th to 75th percentile intervals on either side of the median (i.e., the middle 50% of the samples), the lines give the full range of the samples, and the discrete points indicate outliers. This presentation method allows for a more comprehensive evaluation of different... τThe impact of configuration and routing strategies on entropy and latency volatility. From a latency perspective, significant fluctuations exist between experimental samples: in the optimal case, message transmission can be achieved with a latency of almost 100 ms, while in the worst case, the latency can reach 450 ms. The main reason for this difference is that the mixed latency introduced by each relay satellite follows an exponential distribution, which has a long tail, causing some messages to experience significant delays. From the perspective of entropy distribution, its volatility is significantly less than that of latency, especially under higher randomness parameters. τ Under these values, only a very small number of outlier samples have entropy values ​​below 9 bits. Under the current message traffic and mixed processing conditions for each relay satellite, even in the most extreme cases, such as low... τ Even with the unbalanced strategy, no message becomes completely traceable (entropy value is 0), and the system always maintains a certain degree of anonymity. When At that time, compared to uniform routing ( τ The routing strategy with =1) can significantly reduce latency. Taking the greedy equilibrium strategy (blue box plot) as an example, the latency of the first quartile decreases from 430 ms to 320 ms, a reduction of approximately 110 ms; the median latency decreases from 400 ms to 270 ms, a reduction of approximately 130 ms. At the same time, the anonymity cost is also relatively controllable: in τ When =0, the median loss of the entropy value is approximately 2 bits, while τ At a resolution of 0.6, only 0.2 bits are lost, demonstrating a superior trade-off between latency and anonymity. From Figure 9 It can also be observed that the median change trend of entropy and time delay is similar to... Figure 6 The average values ​​of the theoretical results are basically consistent: Within the interval, the latency remains basically stable; while when Subsequently, the latency increased significantly. Conversely, the entropy value decreased. It continued to grow within the range, until τ= The saturation point is reached at 0.8, and further increases are no longer significant. In summary, τ= 0.6 remains the "sweet spot" for balancing latency and entropy, maintaining low latency while causing only a slight loss of anonymity.

[0084] Experiment 3: Performance evaluation experiment for relay satellite network scaling. The experiment aims to analyze the proposed routing strategy under different relay satellite network scales (…). N The time delay and entropy performance under three different randomness parameters (e.g., 102, 216, 327, 435, 540, each group being a three-layer structure) are considered. τ Low randomness τ =0.1: Corresponds to approximately deterministic routing; high randomness. τ=0.9: Corresponds to near-uniform random routing; moderate randomness. τ =0.6: Provides an optimal trade-off between latency and anonymity. In all experiments, relay satellites were assigned to different layers using clustering and diversified hierarchical algorithms, and routing weights were generated using a greedy load balancing strategy. See also Figure 10 , Figure 10 The theoretical experimental results of entropy and time delay under different relay satellite network scales are shown. From the perspective of time delay variation (solid line): in approximately uniform routing (… τ= Under the condition of 0.9 (blue line), the latency remains relatively stable at approximately 440ms as the scale of the relay satellite network changes; under moderate randomness (… τ =0.6 (purple line), as the scale of the relay satellite network increases, the ms decreases from 360ms to 240ms, a reduction of 33%; under approximately deterministic routing ( τ =0.1 (red line) Under this condition, the expansion of the relay satellite network scale brings more optional paths and increases the optimization space. For example, when the relay satellite network scale is... N The average latency at =540 is 179 ms, while N In small relay satellite networks with a latency of 102, the latency is 308 ms, a reduction of 42%, which is similar to the latency optimization range of moderately random configurations. This is especially true in large-scale relay satellite networks. τ =0.6 and τ A value of 0.1 can achieve low path latency. Looking at the entropy change (dashed line), the entropy value continuously increases with the expansion of the relay satellite network, showing a stable growth trend. By definition, the theoretical upper limit of entropy is... ,in W This indicates the number of relay satellites at each layer of the relay satellite network. When W As the number of satellites increases, the number of candidate relay satellites that could serve as the last hop in the message path also increases, further enhancing the uncertainty of the path, especially with larger random parameters. τ Under these conditions, routing decisions exhibit greater randomness, resulting in a more significant increase in entropy. Different randomness parameters... τ Theoretical and experimental results show that in highly random routing ( τ At a value of 0.9 (blue line), the entropy increase is most significant; while at low randomness ( τ At a value of 0.1 (red line), the entropy increase is relatively limited due to the near-deterministic nature of the routing; for moderately random configurations (…), τ =0.6 (purple line), the entropy increase is between the two, but still shows a significant improvement, especially in its latency performance which is much lower than that of high. τ Configuration ( τ =0.9), slightly higher than low τ Configuration ( τEven with a resolution of 0.1, it still provides high anonymity, demonstrating excellent overall performance advantages.

[0085] Experiment 4: In different Under the given values, the results of CLAPS are compared with the VLEO layer routing strategies (including naive equilibrium and greedy equilibrium) in the AR-DLSN algorithm.

[0086] To evaluate the overall performance in terms of entropy and latency, a metric is constructed here to measure the trade-off between the two. This indicator reaches larger values ​​when entropy is higher and the average total latency of the relay satellite network is lower; therefore, a larger value indicates a better trade-off. Please see [link to relevant documentation]. Figure 11 , Figure 11 Entropy / delay with respect to parameters in CLAPS and AR-DLSN algorithms τ The diagram illustrating the changing relationships shows that: in all Under the given values, both load balancing strategies of the VLEO layer routing policy in the AR-DLSN algorithm are significantly better than CLAPS. When When the initial value is small, the naive equilibrium strategy outperforms the greedy equilibrium strategy, and the performance of the greedy equilibrium strategy gradually approaches that of the naive equilibrium strategy; as the initial value decreases... The increase, in At these values, the performance of the two strategies is almost identical. Overall, the naive equilibrium strategy achieves optimal performance in most cases. Similarly, consistent with previous experiments, It can achieve a good trade-off between entropy and latency.

[0087] In addition, this experiment also recorded the generation of route weights using three strategies: CLAPS, greedy equilibrium, and naive equilibrium. The running time during the process, and the experimental results are as follows: Figure 12 As shown, it can be seen that as the scale of the relay satellite network continues to expand, the running time of the greedy equilibrium and naive equilibrium strategies remains at a low level, and increases slowly with the scale in an approximately linear manner. This is because the computational complexity of the two equilibrium strategies is low, and they can still maintain high computational efficiency even when the scale of the relay satellite network increases significantly. In contrast, the running time of CLAPS increases significantly non-linearly with the increase of the relay satellite network scale. When the network size is small (e.g., N<60), the running time of CLAPS is similar to that of the two equilibrium strategies. However, when N>100, the running time of CLAPS rises rapidly and is much higher than that of the greedy equilibrium and naive equilibrium strategies. This trend reflects that CLAPS's dependence on solving linear programming leads to a sharp increase in computational complexity with the problem size.

[0088] Furthermore, the running time scaling of CLAPS relative to naive equilibrium and greedy equilibrium was calculated, and the experimental results are as follows: Figure 13 As shown. From Figure 13 It can be observed that, under relay satellite networks of any size, the runtime ratio of CLAPS / naive equilibrium is slightly higher than that of CLAPS / greedy equilibrium, meaning that the solution time of naive equilibrium is consistently shorter than that of greedy equilibrium. Taking a scenario with three layers and 70 relay satellites per layer as an example, the runtime of CLAPS in solving a linear programming problem is approximately 90 times that of the two equilibrium strategies. As the scale of the relay satellite network continues to expand, this ratio shows a stable upward trend, indicating that the additional computational overhead introduced by CLAPS in terms of global optimization and solution accuracy is more significant in large-scale network environments.

[0089] In summary, although CLAPS can theoretically achieve joint optimization of routing entropy and latency through linear programming, its compromise performance is not superior to the naive equilibrium strategy or the greedy equilibrium strategy. Furthermore, because the computational complexity of CLAPS increases dramatically with the scale of the relay satellite network, its runtime grows too rapidly in large-scale ultra-low orbit satellite constellation scenarios, making it difficult to meet the timeliness requirements of frequent route weight updates. In contrast, the naive equilibrium and greedy equilibrium strategies have significant advantages in high joint optimization performance of entropy and latency, low computational time, and strong scalability, making them more suitable for handling high-frequency, heavy-load route weight calculation tasks. In actual system deployments, satellite link congestion and topology changes occur frequently, and route weights typically require high-frequency dynamic updates. If CLAPS is directly extended to the current Starlink network scale (over 6000 satellites), its solution time is expected to exceed one hour, making it difficult to meet real-time or near-real-time route adjustment requirements. In comparison, the naive equilibrium and greedy equilibrium strategies have significant advantages in computational efficiency and convergence speed. Based on the experimental results above, it can be seen that under low randomness conditions, the naive equilibrium strategy outperforms the greedy equilibrium strategy in terms of the trade-off between entropy and time delay, while its computational cost is slightly lower. Therefore, the naive equilibrium strategy is more suitable as the core mechanism for high-frequency routing updates in large-scale satellite networks.

[0090] The above description, in conjunction with specific preferred embodiments, provides a further detailed explanation of the present invention. It should not be construed that the specific implementation of the present invention is limited to these descriptions. For those skilled in the art, various simple deductions or substitutions can be made without departing from the concept of the present invention, and all such modifications and substitutions should be considered within the scope of protection of the present invention.

Claims

1. An anonymous routing method for a two-layer satellite network oriented towards latency-sensitive services, applied to a VLEO / LEO two-layer satellite network, characterized in that, The VLEO / LEO dual-layer satellite network is implemented based on a software-defined networking (SDN) architecture, including a data plane, a control plane, and an application plane. The data plane includes a VLEO layer, a VLEO / LEO cross-layer, and an LEO layer. The application plane deploys the anonymous routing method of the dual-layer satellite network. Routing decisions are generated by running the anonymous routing method of the dual-layer satellite network and forwarded to the data plane for execution via the control plane. The anonymous routing method for the two-layer satellite network includes: Based on the hierarchical structure of the VLEO / LEO dual-layer satellite network, a hierarchical decoupled routing calculation framework is constructed. The hierarchical decoupled routing calculation framework divides routing decisions into multi-layer path selection, including LEO layer intra-layer routing, VLEO / LEO cross-layer routing, and VLEO layer intra-layer routing. By leveraging the collaborative computing capabilities of the SDN controller and satellite nodes, the multi-layer path selection is reconfigured and weighted in real time to generate the optimal routing path; Specifically, the LEO layer intra-layer routing is implemented using a K-shortest path algorithm and a weighted random selection routing decision mechanism, and communicates via laser inter-satellite links; the VLEO / LEO cross-layer routing is implemented using a two-stage routing decision mechanism of weighted random selection and load-aware optimization, and communicates via VLISL cross-layer links that satisfy the constraints of no Earth obstruction and maximum communication distance; the VLEO layer intra-layer routing is implemented using a hierarchical optimization, latency-aware, and load-balancing routing decision mechanism, and communicates via microwave inter-satellite links.

2. The anonymous routing method for a two-layer satellite network according to claim 1, characterized in that, Both the VLEO layer and the LEO layer satellite constellations adopt a polar orbit Walker constellation configuration, and the inter-satellite links follow the Iridium system's four-link principle, including two intra-orbit links between adjacent satellites in the same orbit and two inter-orbit links between adjacent orbital planes. When a satellite enters the polar region, the inter-orbit links are closed; when a satellite leaves the polar region, the inter-orbit links are re-established. The polar region is an area with latitudes exceeding a preset value.

3. The anonymous routing method for a two-layer satellite network according to claim 1, characterized in that, The routing decision mechanism using the K shortest path algorithm and weighted random selection within the LEO layer includes: The KSP algorithm is used to calculate multiple optimal candidate paths for the source satellite node and the destination satellite node; Each optimal candidate path is assigned a weight, and the selection probability of each optimal candidate path is calculated based on a weighted random mechanism; wherein, the weight of each optimal candidate path is negatively correlated with the path length.

4. The anonymous routing method for a two-layer satellite network according to claim 1, characterized in that, The two-stage routing decision mechanism adopted for VLEO / LEO cross-layer routing, which combines weighted random selection and load-aware optimization, includes: The initial forwarding probability is calculated based on the Euclidean distance between VLEO satellites and LEO satellites, and the forwarding probability is used to pre-select the next-hop LEO satellite for VLEO satellites, thereby realizing the weighted random selection of satellites across layers. Based on the weighted random selection results of cross-layer satellites, the node degree of LEO satellites is statistically analyzed to sense the load of LEO satellites. The load of LEO satellites and the Euclidean distance between VLEO satellites and LEO satellites are optimized to obtain the final probability of VLEO satellites selecting LEO satellites as cross-layer relay nodes.

5. The anonymous routing method for a two-layer satellite network according to claim 4, characterized in that, The payload of the LEO satellite and the Euclidean distance between the VLEO satellite and the LEO satellite are optimized to obtain the final probability of the VLEO satellite selecting the LEO satellite as the cross-layer relay node, including: The node degree of the LEO satellite and the Euclidean distance between the VLEO satellite and the LEO satellite are normalized to obtain the normalized node degree and Euclidean distance. Calculate the distance weight and load weight based on the normalized node degree and Euclidean distance, respectively. Based on the distance weight and the load weight, an adjustment factor is introduced. To balance distance and load, by increasing The value is used to increase the distance weight, thereby prioritizing LEO satellites with closer physical distances; by decreasing... The value is used to reduce the distance weight, thereby prioritizing the selection of LEO satellites with lower loads, and finally obtaining the final probability of VLEO satellites selecting LEO satellites as cross-layer relay nodes.

6. The anonymous routing method for a two-layer satellite network according to claim 1, characterized in that, In the VLEO / LEO cross-layer, the maximum communication distance of the VLISL cross-layer link is calculated by combining the Earth's radius, VLEO orbital altitude, LEO orbital altitude, and atmospheric altitude; the satellites of both communicating parties must be on the same side of the Earth and have no line of sight obstructed by the Earth.

7. The anonymous routing method for a two-layer satellite network according to claim 1, characterized in that, The routing decision-making mechanism employed within the VLEO layer, which includes hierarchical optimization, latency awareness, and load balancing, includes: The K-medoids clustering algorithm is used to cluster satellites based on their geographical location, and the relay satellites are divided into multi-layer structures according to the distribution of cluster centers to construct a hierarchical network topology with geographical diversity. Based on the aforementioned hierarchical network topology, a balance parameter τ, τ∈[0,1], is introduced to balance end-to-end latency and anonymity during the relay satellite selection process. When τ=0, it indicates that the goal is to minimize the average total latency of the relay satellite network, prioritizing the selection of the relay satellite with the lowest latency to optimize the average total latency of the network to the greatest extent. When τ=1, it indicates that the goal is to enhance anonymity, randomly selecting relay satellites to maximize the anonymity of the network. At the VLEO layer, for scenarios where τ < 1, a load balancing mechanism is adopted, which achieves traffic balance by adjusting the relay satellite routing weight.

8. The anonymous routing method for a two-layer satellite network according to claim 7, characterized in that, The method of dividing relay satellites into a multi-layered structure based on cluster center distribution includes: Based on the number of clusters K Number of nodes per layer W Relationships can be categorized into the following three patterns: when W < K At that time, clusters are randomly selected for each layer according to the principle of the farthest cluster center, and relay satellites are randomly selected from that cluster; the above process is repeated until all relay satellites have been assigned to the corresponding layer; when W > K At that time, according to the cluster size ratio, one relay satellite is randomly selected from each cluster as part of the members of that layer, and the remaining relay satellites are randomly selected from each cluster according to the cluster size ratio; repeat the above process until all relay satellites are assigned to the corresponding layer; when W = K At that time, a relay satellite is randomly selected from each cluster as a member of that layer; the above process is repeated until all relay satellites are assigned to the corresponding layer.

9. The anonymous routing method for a two-layer satellite network according to claim 7, characterized in that, The load balancing mechanism includes a naive load balancing algorithm; the naive load balancing algorithm achieves load balancing through a single weight adjustment and load redistribution operation, specifically including: Based on the scattering matrix column and the calculation of the load of each relay satellite, the relay satellites are divided into underloaded relay satellites and overloaded relay satellites according to a preset threshold. Reduce the routing weight of the overloaded relay satellites and calculate the remaining load released; Based on the shortfall ratio of the underloaded relay satellites, the remaining load is redistributed to the underloaded relay satellites.

10. The anonymous routing method for a two-layer satellite network according to claim 7, characterized in that, The load balancing mechanism includes a greedy load balancing algorithm, which identifies overloaded relay satellites and reduces their routing weights through multiple rounds of iteration, while increasing the weights of underloaded relay satellites. In each round, the weights are recalculated based on the latest load distribution, and the remaining load is distributed proportionally to gradually achieve load migration.

Images