Extensible video caching method and system for low earth orbit satellite network

Abstract

The invention relates to the technical field of satellite communication, and particularly discloses an extensible video caching method and system for a low earth orbit satellite network, and the method comprises the steps: firstly determining a caching model of the low earth orbit satellite network; then, based on the cache model, an extensible video cache optimization problem is constructed, and the problem takes minimization of total video access delay experienced by all ground users as a target and takes storage constraint of each satellite, binary constraint of hierarchical video content placement variables on the satellite and bandwidth allocation constraint of the satellite as constraint conditions; and finally, solving an extensible video cache optimization problem by adopting a multi-agent online learning algorithm to obtain a layered video content placement strategy and a downlink bandwidth allocation strategy. Different from the previous research that the system state is assumed to be known, the method is operated in an online environment, and the user demand is unknown and changes along with time. A simulation result shows that the access delay is remarkably reduced compared with an existing baseline.

Description

Technical Field

[0001] The present invention relates to the field of satellite communication technology, and in particular to an expandable video caching method and system for a low earth orbit satellite network. Background Art

[0002] The proliferation of 5G networks has led to a sharp increase in wireless data traffic, which poses significant challenges to traditional cloud-based video transmission, especially in remote and bandwidth-constrained areas, where high latency and limited capacity hinder service quality. Edge caching effectively reduces latency and alleviates backhaul network congestion by storing content closer to users. However, ground-based edge nodes have limited coverage and are expensive to expand in underserved areas. In contrast, low-Earth orbit (LEO) satellite networks provide extensive coverage and are able to deliver communication services to areas where deploying terrestrial infrastructure is impractical. Combining LEO networks with edge caching enables low-latency, high-throughput video transmission, improving multimedia delivery in challenging environments.

[0003] Recent research has explored satellite networks equipped with caching units to enhance content delivery. For example, one paper proposed a cooperative caching strategy combining LSTM (Long Short-Term Memory) and the Soft Actor-Critic (SAC) algorithm. Others applied game theory and ridge regression to improve distributed caching and coverage partitioning. Another paper proposed a method based on MAPPO (Multi-Agent Proximal Policy Optimization) for optimizing caching under dynamic satellite topologies, while another paper utilized the Alternating Direction Method of Multipliers (ADMM) to reduce user access latency through distributed caching. However, most of these studies focused on traditional file caching, lacking the flexibility to accommodate video services with diverse quality requirements. To address this issue, Scalable Video Coding (SVC) has been used for edge caching. SVC encodes video into a base layer and multiple enhancement layers, with decoding the higher layers requiring all lower layers. This layered structure enables fine-grained caching and delivery, adapting to user preferences and channel variations while maximizing the utilization of limited storage and bandwidth.

[0004] Most existing SVC-based caching studies rely on offline optimization, assuming prior knowledge of system conditions. In practice, although the distribution of user request probabilities is known, the actual request volume varies over time and is difficult to predict. Furthermore, in LEO satellite networks, satellite mobility leads to dynamically changing channel conditions and fluctuating link quality. These uncertainties in user demands and communication links complicate caching decisions. In a cooperative network, each satellite observes only local user behavior and channel status, but its caching choices affect not only its own users but also the service capabilities of neighboring satellites. This interdependence highlights the need for adaptive, decentralized caching strategies that can effectively cope with dynamic and partially observable environments. Summary of the Invention

[0005] The present invention provides a scalable video caching method and system for low-Earth orbit satellite networks, which solves the technical problem that existing SVC-based caching research relies on offline optimization and lacks an adaptive, decentralized caching strategy that can effectively cope with dynamic and partially observable environments.

[0006] To solve the above technical problems, the present invention provides a scalable video caching method for a low earth orbit satellite network, comprising the steps of:

[0007] S1. Determine a caching model for a low-Earth orbit satellite network; the caching model includes a satellite network consisting of M low-Earth orbit satellites, each of which establishes a direct communication link with a ground station and is responsible for providing video content to K ground users within a coverage area;

[0008] S2. Based on the cache model, construct a scalable video cache optimization problem, where the problem aims to minimize the total video access delay experienced by all terrestrial users, with a layered video content placement variable on the satellite and a downlink bandwidth allocation variable from the satellite to terrestrial users as solution variables, and with a storage constraint on each satellite, a binary constraint on the layered video content placement variable on the satellite, and a bandwidth allocation constraint on the satellite as constraints;

[0009] S3. Use a multi-agent online learning algorithm to solve the scalable video cache optimization problem, and obtain a layered video content placement strategy on each satellite in each time period and a downlink bandwidth allocation strategy for each satellite to ground users.

[0010] Furthermore, in step S1, in the cache model, when a ground user requests a video layer, the relevant satellite first checks whether its local cache contains the content. If the content is available, the satellite will directly transmit it to the ground user. Otherwise, the satellite forwards the query to the neighboring satellite via the intersatellite link and attempts to obtain the missing layer from the nearest satellite that stores the layer. If the requested content is not cached in the entire satellite network, the satellite obtains the content from the ground station via the backhaul link and then transmits it to the ground user.

[0011] Furthermore, in step S1, the cache model maintains a video content library consisting of N videos, the popularity of each video follows Zipf's law, and each video is encoded into L hierarchical quality layers using SVC; all ground users have the same preference for different quality layers; the cache model operates in T discrete time periods, and in each time period t, the ground user numbered k will generate a predetermined number of video layer requests The cache state of layer l of video n on satellite m in time period t is represented by the binary variable Indicates that if the layer is cached locally, then Otherwise 0.

[0012] Furthermore, in step S2, the objective function of the scalable video cache optimization problem is:

[0013]

[0014] in, is the downlink bandwidth allocated by satellite m to ground user k in time period t, To solve for the variable, φ nl is the request probability of the lth layer of video n, represents the set of ground users associated with satellite m, is the transmission delay between satellite m and ground user k if the video content has been cached on satellite m.

[0015] Furthermore, the storage constraint for each satellite is:

[0016]

[0017] Among them, C m represents the storage budget of satellite m, It means arbitrary;

[0018] The binary constraints on the layered video content placement variables on each satellite are:

[0019]

[0020] The bandwidth allocation constraint for each satellite is:

[0021]

[0022] B is the total downlink bandwidth available for each satellite.

[0023] Furthermore, step S3 specifically includes:

[0024] S31, by removing the constant, the scalable video cache optimization problem is transformed into a given time period t and satellite m, the goal is to The total bandwidth B is distributed among them to minimize the overall expected access delay A new optimization problem, represents the effective video data load of ground user k associated with satellite m during time period t;

[0025] S32. For the new optimization problem, a closed-form solution is obtained through Lagrange duality. form;

[0026] S33, based on closed-form solution In the form of, a multi-armed bandit algorithm combined with an upper confidence bound algorithm is used to solve the scalable video cache optimization problem, and the optimal cache action and the optimal bandwidth allocation are obtained.

[0027] Furthermore, step S33 specifically includes the following steps:

[0028] S331, initialization t = 1, the upper confidence bound of each action a of each satellite m is infinite, the average reward of each action a of each satellite m is 0, the cache budget of each satellite m is ω m C m , the action set of each satellite m

[0029] S332: In each time period t, perform the following steps:

[0030] For each satellite m, when ω m >0, select the optimal action and judge Is it true? If so, then a * From the action space Remove, if otherwise cache a * And update ω m for And a * From the action space Remove; According to the closed-form solution The optimal bandwidth allocation is obtained in the form of

[0031] For each satellite m, for each of its selected actions a, calculate the reward that satellite m evaluates for each selected action based on the observed system performance at the end of time period t according to Update the reward for the next moment according to Update the upper confidence bound for the next moment

[0032] Reset m ,

[0033] Enter the next time period.

[0034] Furthermore, in each time period t, the upper confidence bound of each action a of each satellite m is for:

[0035]

[0036] in, represents the number of times satellite m chooses action a until time period t, and represents the total number of actions performed by satellite m up to time period t, represents the empirical average reward of satellite m choosing action a up to time period t, Indicates the previous time period β represents the learning rate; Denotes the reward of each selected action a evaluated by satellite m based on the observed system performance at the end of time period t:

[0037]

[0038] in, represents the normalized reduction in the total video access delay for all GUs associated with satellite m at the end of time period t compared to the worst-case baseline, represents the maximum delay reduction experienced by a user terminal associated with satellite m at the end of time period t when the requested video layer is served from the local cache, represents the synergistic benefit brought by the cache content of satellite m being used to serve user terminals associated with other satellites through inter-satellite links at the end of time period t, and η1, η2, η3∈[0,1] are weight parameters that control the contribution of each component.

[0039] Furthermore, Specifically:

[0040]

[0041] in, represents the total video access delay experienced by all GUs associated with satellite m at the end of time period t, is a normalizing constant;

[0042] Specifically:

[0043]

[0044] in, is the number of times ground user k visits the quality layer corresponding to action a in time period t, o a represents the size of the video layer corresponding to action a, is the transmission delay per bit between satellite m and the ground station in time period t, is the transmission delay per bit between satellite m and ground user k in time period t, is the propagation delay per bit between satellite m and the ground station in time period t, is the propagation delay per bit between satellite m and ground user k in time period t;

[0045] Specifically:

[0046]

[0047] Among them, j refers to other satellites, represents the set of ground users associated with satellite j, and k′ refers to the set of ground users associated with satellite j. Ground users in is the number of times ground user k′ visits the quality layer corresponding to action a within time period t.

[0048] The present invention also provides a scalable video caching system for a low-Earth orbit satellite network, the key of which is that it includes a cache model construction module, an optimization problem construction module and an optimization problem solving module, which are respectively used to execute steps S1, S2 and S3 of the scalable video caching method for the low-Earth orbit satellite network.

[0049] The present invention provides a scalable video caching method and system for a low-Earth orbit satellite network. First, a caching model for the low-Earth orbit satellite network is determined. Then, based on the caching model, a scalable video caching optimization problem is constructed. This problem aims to minimize the total video access delay experienced by all terrestrial users. The problem uses the layered video content placement variables on the satellites and the downlink bandwidth allocation variables from the satellites to terrestrial users as solution variables. Constraints include each satellite's storage constraint, a binary constraint on the layered video content placement variables on the satellites, and the satellite's bandwidth allocation constraint. Finally, a multi-agent online learning algorithm is used to solve the scalable video caching optimization problem, yielding a layered video content placement strategy for each satellite and a downlink bandwidth allocation strategy for each satellite to terrestrial users for each time period. The present invention minimizes video access delay by jointly optimizing layered video caching and bandwidth allocation between satellites, while taking into account collaboration and resource constraints. Unlike previous studies that assumed known system states, the present invention operates in an online environment where user demands are unknown and change over time. A multi-agent online learning algorithm based on a multi-armed bandit (MAB) framework is proposed, in which each satellite independently learns its caching strategy. Lagrangian duality is further applied to efficiently allocate downlink bandwidth. Simulation results show that the proposed method significantly reduces access latency compared to existing baselines, highlighting the importance of satellite collaboration and adaptive learning in dynamic environments. BRIEF DESCRIPTION OF THE DRAWINGS

[0050] Figure 1 is a flow chart of a scalable video caching method for a low earth orbit satellite network provided by an embodiment of the present invention;

[0051] Figure 2 This is a comparison chart of online optimization performance of video access delay under different solutions provided by the embodiments of the present invention;

[0052] Figure 3 is a diagram showing the impact of satellite cache capacity on average video access delay under different solutions provided by an embodiment of the present invention;

[0053] Figure 4 is a graph showing the effect of an increase in the average number of video layer requests from GUs on the total video access delay, provided by an embodiment of the present invention;

[0054] Figure 5 This is a diagram showing the influence of the skewness parameter on the average total video access delay under different solutions provided by the embodiments of the present invention. DETAILED DESCRIPTION

[0055] The following describes the embodiments of the present invention in detail with reference to the accompanying drawings. The embodiments are provided for illustrative purposes only and are not to be construed as limiting the present invention. The accompanying drawings are provided for reference and illustration only and do not constitute a limitation on the scope of protection of the present invention. Many changes may be made to the present invention without departing from the spirit and scope of the present invention.

[0056] The embodiment of the present invention provides a scalable video caching method for a low earth orbit satellite network, such as Figure 1 As shown in the flowchart, the steps include:

[0057] S1. Determine a caching model for a low-Earth orbit satellite network. The caching model includes a satellite network consisting of M low-Earth orbit (LEO) satellites. Each satellite establishes a direct communication link with a ground station (GS) and is responsible for providing video content to K ground users (GUs) within the coverage area.

[0058] S2. Based on the cache model, construct a scalable video cache optimization problem, where the problem aims to minimize the total video access delay experienced by all terrestrial users, with a layered video content placement variable on the satellite and a downlink bandwidth allocation variable from the satellite to terrestrial users as solution variables, and with a storage constraint on each satellite, a binary constraint on the layered video content placement variable on the satellite, and a bandwidth allocation constraint on the satellite as constraints;

[0059] S3. Use a multi-agent online learning algorithm to solve the scalable video cache optimization problem, and obtain a layered video content placement strategy on each satellite in each time period and a downlink bandwidth allocation strategy for each satellite to ground users.

[0060] Leveraging the extensive coverage of low-Earth orbit satellites, this paper proposes a collaborative edge caching framework for Scalable Video Coding (SVC) video. A multi-agent online learning algorithm is developed to minimize video access latency by jointly optimizing hierarchical video caching and bandwidth allocation across satellites, while accounting for collaboration and resource constraints. Unlike previous studies that assumed known system states, this approach operates in an online environment where user demands are unknown and time-varying. This approach enables each satellite to adaptively balance exploration and exploitation under local observation, effectively addressing dynamic user demands and time-varying channel conditions, and significantly reducing long-term video access latency.

[0061] The following describes each step in detail.

[0062] (1) Step S1: Determine the cache model of the low earth orbit satellite network

[0063] To ensure efficient video multimedia transmission in remote or disaster-stricken areas, consider a satellite network consisting of M low earth orbit (LEO) satellites. LEO ={LEO1,LEO2,…,LEO M Each satellite establishes a direct communication link with the ground station (GS) and is responsible for providing video content to K ground users (GUs) within the coverage area. These users use the set K GU ={GU1,GU2,…,GU K} represents. For m∈M LEO Each satellite of m provides services to GUs within its coverage area. The set of GUs associated with satellite m is expressed as In order to reduce content retrieval latency and reduce dependence on GS, edge caching is introduced on the satellite side. Each satellite (satellite m) is equipped with a limited storage capacity C m > 0 (in bits). In addition, satellites are connected to each other via intersatellite links, forming a cooperative network that allows content to be shared between satellites.

[0064] The network maintains a video content library consisting of N videos, with videos indexed by n = 1, ..., N. The popularity of each video follows Zipf's law, and the request probability of video n is defined as Where θ is a skewness parameter that controls the concentration of requests, and q is a temporary index variable used to iterate over all items (in this case, the request probabilities of the video layer) and sum them to ensure the normalization of the probability distribution. To adapt to different device capabilities and channel conditions, each video is encoded using SVC into L hierarchical quality layers. The first layer corresponds to the base quality, and each subsequent layer provides incremental quality improvement. The size of the lth layer of video n is o nl , satisfying the condition o n1 ≥o n2 ≥…≥o nL The preference weight of quality layer l is expressed as z l Assuming that all GUs have the same preference for different quality layers, the request probability of the lth layer of video n is φ nl =r n z l .

[0065] The network operates in discrete time periods t∈{1,2,…,T}. In each time period, each GU generates a certain number of video layer requests, expressed as This example defines a binary variable To indicate the cache status of layer l of video n on satellite m in time period t. Specifically, if the layer is cached locally, then Otherwise, it is 0. The cache decision must satisfy the storage constraints of the satellite, i.e. When a GU requests a video layer, the relevant satellite first checks its local cache to see if the content exists. If the content is available, the satellite transmits it directly to the GU. Otherwise, the satellite forwards the query to neighboring satellites via intersatellite links and attempts to retrieve the missing layer from the nearest satellite storing the layer. If the requested content is not cached anywhere in the satellite network, the satellite retrieves the content from the GS via the backhaul link. This tiered content retrieval strategy helps balance latency, bandwidth usage, and storage efficiency across the network.

[0066] To support layered video transmission and cooperative caching in a dynamic LEO satellite network, this example considers three communication links: the satellite-to-GU link, the inter-satellite link (ISL), and the satellite-to-GS backhaul link. To maintain symbol consistency, all ground nodes are indexed as Where k=0 represents GS.

[0067] 1) Satellite to GU link

[0068] Each satellite provides services to its associated GU through the downlink channel. In time period t, satellite m and the kth GU (GU k ) is modeled as in and are the antenna gains of the satellite as the transmitter and the GU as the receiver, λ1 is the signal wavelength, It is satellite m and GU k The distance in time period t, and F rain represents rain attenuation. Therefore, satellite m and GU k The achievable rate between in The satellite m is assigned to GU in time period t k Bandwidth, P s is the transmit power of each satellite, and is the noise power of each GU. The bandwidth allocation of the satellite must satisfy Where B is the total downlink bandwidth available for each satellite. Therefore, in time period t, satellite m and GU k The transmission delay per bit is The propagation delay is c represents the speed of light.

[0069] 2) Intersatellite links

[0070] In order to achieve content sharing between satellites, this example considers a free space optical (FSO) intersatellite link. In time period t, the achievable rate between satellite j and satellite m is

[0071] B fis the bandwidth of the intersatellite link (ISL), is the intersatellite distance between satellite j and satellite m at time period t, is the noise power of the intersatellite link. λ2 is the wavelength of the FSO signal. The parameters η, κ, and e represent the optical efficiency, link gain, and pointing error loss, respectively. Therefore, the transmission delay per bit between satellite j and satellite m in time period t is The propagation delay is

[0072] 3) Return link from ground station to satellite

[0073] When the GU associated with satellite m requests content that is not cached in the satellite network, satellite m retrieves the content from GS. The channel gain between satellite m and GS is modeled as in and are the antenna gains of the ground station as the transmitter and the satellite as the receiver respectively. The achievable rate between satellite m and GS is Among them B g represents the fixed backhaul bandwidth, P g is the transmission power of the ground station. Therefore, the transmission delay per bit between satellite m and GS in time period t is The propagation delay is

[0074] Consider GU The case of requesting the quality level l of video n in time period t. If the video content has been cached on satellite m, the transmission delay between satellite m and GU k is If the content is not cached, the access latency is: Here represents the inter-layer satellite transmission delay, and represents the delay in transmitting this layer from GS. Due to the dependency of Scalable Video Coding (SVC), decoding the quality of layer l requires transmitting all preceding layers first. Therefore, the total delay for requesting the quality of layer l for video n is min means taking the minimum value.

[0075] (2) Step S2: Constructing a scalable video cache optimization problem

[0076] This example considers the problem of layered video caching in a LEO satellite network with caching capabilities, with the goal of minimizing the total video access latency experienced by all GUs. This can be achieved by jointly optimizing the layered video content placement variables on the satellite. and the downlink bandwidth allocated to GU The scalable video cache optimization problem can be formulated as follows:

[0077]

[0078] Solving problem (P1) is challenging because the demand for GUs varies over time, and each satellite can only partially observe these demands, which leads to uncertainty in caching decisions. Moreover, in the case of inter-satellite cooperation, a satellite's caching decision not only affects the GUs it serves but also the availability of content for other satellites. Furthermore, the problem involves binary variables, its objective function is highly coupled across the GU, video layer, and time frame dimensions, and its overall non-convex structure makes it difficult to directly solve using conventional optimization techniques.

[0079] (3) Step S3: Solving the scalable video cache optimization problem

[0080] This example proposes a distributed online learning framework in which each satellite independently optimizes its caching strategy using a multi-armed bandit (MAB) approach. During each time period, the satellite selects the video layer to cache based on observed GU requests. Because access latency is influenced by both the caching strategy and bandwidth allocation, the satellite also solves a local bandwidth optimization subproblem to evaluate the current caching decision. The resulting access latency serves as feedback to guide future caching behavior.

[0081] 1) MAB online learning

[0082] To achieve online decision making under uncertain conditions, this example models each satellite as an agent operating in a MAB environment. In the classic MAB setting, the agent sequentially selects from a finite set of actions. The goal is to maximize the cumulative reward over time while balancing the trade-off between exploration (selecting underexplored arms to improve reward estimation) and exploitation (selecting the arm that is currently believed to bring the highest reward).

[0083] set up represents an action (e.g., caching a specific video layer), and y(a) represents its expected reward. Since y(a) is unknown in advance, the agent must learn from past interactions. This example uses the upper confidence bound (UCB) algorithm, a widely used MAB strategy with provably sublinear regret. At each time step t, the satellite m chooses the action that maximizes the following expression

[0084]

[0085] in, represents the empirical average reward of satellite m choosing action a up to time period t, represents the number of times satellite m chooses action a until time period t, and It represents the total number of actions performed by satellite m up to time period t. Used to encourage exploration, tending to choose actions that have been tried less times.

[0086] In this framework, each caching action (i.e., selecting a video layer to store) corresponds to an arm. In each time period, the satellite observes the request and delayed feedback of the local GU and updates the reward estimate This process enables each agent to learn an effective caching strategy online without relying on prior knowledge of user needs or channel conditions.

[0087] At each time period t, satellite m needs to determine the video layer to cache under the constraint of its remaining cache capacity. Therefore, in this example, the cache action space of satellite m is defined as:

[0088]

[0089] Satellite m maintains a remaining cache capacity ω m , the budget is initialized to the total capacity C at the beginning of each time period m The satellite then performs a sequential decision process: at each step, it selects from the set of actions Select the action a with the largest UCB value * :

[0090]

[0091] set up Representation and action * The size of the corresponding video layer. The layer is cached and the remaining capacity is updated to And a * From the action space Remove, that is, the action space Updated to Otherwise, skip the action and set the action space Updated to This selection process continues until there are no actions in the action space that can be accommodated by the current remaining capacity.

[0092] At the end of time period t, satellite m evaluates the reward of each selected action a based on the observed system performance The reward is expressed as a weighted sum of the following three components:

[0093]

[0094] in, represents the normalized reduction in the total video access delay for all GUs associated with satellite m at the end of time period t compared to the worst-case baseline, represents the maximum delay reduction experienced by a user terminal associated with satellite m at the end of time period t when the requested video layer is served from the local cache, represents the synergistic benefit brought by the cache content of satellite m being used to serve user terminals associated with other satellites through inter-satellite links at the end of time period t, and η1, η2, η3∈[0,1] are weight parameters that control the contribution of each component.

[0095] Satellite-level delay gain Specifically:

[0096]

[0097] in, represents the total video access delay experienced by all GUs associated with satellite m at the end of time period t, while is a normalizing constant. Although this metric is calculated based on local delay observations, it forms part of the global system objective. Because overall system efficiency is determined by the aggregated results of each satellite's local performance, improving local performance under a coordinated mechanism can effectively drive overall performance improvements.

[0098] Local service efficiency Specifically:

[0099]

[0100] in, is the number of times GU k visits the quality layer corresponding to action a in time period t, o a Indicates the size of the video layer corresponding to action a.

[0101] Synergistic gain (serving other satellites) Specifically:

[0102]

[0103] Among them, j refers to other satellites, represents the set of ground users associated with satellite j, and k′ refers to the set of ground users associated with satellite j. Ground users in is the number of times ground user k′ visits the quality layer corresponding to action a within time period t.

[0104] Then update the estimated reward value:

[0105]

[0106] Among them, β∈(0,1) represents the learning rate.

[0107] At the same time, the counter and The caching strategy is updated to record the latest action selection statistics. Based on the updated reward estimates and selection statistics, the UCB value of the selected action is recalculated accordingly. Through this iterative process, each satellite continuously optimizes its caching strategy to effectively balance long-term learning and short-term user terminal needs under dynamic network conditions.

[0108] 2) Downlink bandwidth allocation for delay estimation

[0109] To accurately evaluate the reward for each caching action, the satellite must estimate the access latency perceived by the GU, which is determined by both the cache allocation and the downlink bandwidth allocation. Therefore, this example formulates a local bandwidth allocation problem for each satellite per time period. For a given time period t and satellite m, the goal is to allocate its total bandwidth B among the GUs k∈u(m) to minimize the overall expected access latency. By removing the constant, the new optimization problem is formulated as follows:

[0110]

[0111] in, represents the effective video data load of GU k in time period t, and its calculation formula is:

[0112]

[0113] Problem (P2) is a convex and separable optimization problem that can be solved in closed form by Lagrangian duality. Specifically, we introduce a bandwidth-constrained Lagrangian multiplier α ≥ 0 and construct the Lagrangian function:

[0114]

[0115] right about Taking the derivative, setting it to zero, and solving the stationary condition, we get: Substitute it into the bandwidth constraint In the solution, Finally, the closed-form solution for the optimal bandwidth allocation is:

[0116]

[0117] This allocation method balances data volume and demand across user terminals (GUs) and is used to calculate the actual delay value under the current caching decision. This is then used as feedback to update the satellite's caching strategy in a multi-armed bandit (MAB) framework. The entire algorithm is summarized in Algorithm 1 shown in Table 1.

[0118] Table 1 Pseudocode of Algorithm 1

[0119]

[0120]

[0121] As shown in Table 1, the video layer cache optimization algorithm based on UCB-MAB specifically includes the following steps:

[0122] 1. Initialize each satellite m For infinity, is 0, cache budget ω m C m , action set

[0123] 2. At each time step t, the traversal iteratively performs the following steps (this phase is terminated until the cache budget cannot meet the cache requirements of any action):

[0124] For each satellite m, when ω m When >0, select the optimal action according to formula (6) and judge Is it true? If so, then a * From the action space Remove (action space Updated to ), if otherwise cache a * And update ω m for And a * From the action space Remove; according to formula (15) to obtain the optimal bandwidth allocation

[0125] For each satellite m, for each selected action a, the reward is calculated according to equations (7)-(10): Update according to formula (11) Update according to formula (4)

[0126] Reset

[0127] Enter the next time period.

[0128] It should be noted that the various forms of processes shown above can be used to reorder, add, or delete steps. For example, the steps described in the present invention can be executed in parallel, sequentially, or in a different order, as long as the desired results of the technical solution of the present invention can be achieved. This embodiment is not limited here.

[0129] Based on the above-mentioned scalable video caching method, an embodiment of the present invention also provides a scalable video caching system for a low-Earth orbit satellite network, which includes a cache model construction module, an optimization problem construction module and an optimization problem solving module, which are respectively used to execute steps S1, S2 and S3 in the above-mentioned method.

[0130] The embodiments described herein can be implemented in a computing system that includes backend components (e.g., as a data server), or a computing system that includes middleware components (e.g., an application server), or a computing system that includes frontend components (e.g., a user computer with a graphical user interface or web browser through which a user can interact with embodiments of the systems and techniques described herein), or a computing system that includes any combination of such backend components, middleware components, or frontend components. The components of the system can be interconnected by any form or medium of digital data communication (e.g., a communication network). Examples of communication networks include: a local area network (LAN), a wide area network (WAN), a blockchain network, and the Internet.

[0131] Computer programs for implementing the methods and systems of the present invention can be written in any combination of one or more programming languages ​​and stored in a computer-readable storage medium. These computer programs can be provided to a processor of a general-purpose computer, a special-purpose computer, or other programmable data processing device, so that when the computer program is executed by the processor, the functions / operations specified in the flowcharts and / or block diagrams are implemented. The computer program can be executed entirely on the machine, partially on the machine, as a stand-alone software package, partially on the machine and partially on a remote machine, or entirely on a remote machine or server.

[0132] Computer-readable storage media can be tangible media that can contain or store a computer program for use with an instruction execution system, device or equipment or used in combination with an instruction execution system, device or equipment. Computer-readable storage media can include, but are not limited to, electronic, magnetic, optical, electromagnetic, infrared or semiconductor systems, devices or equipment, or any suitable combination of the foregoing. Alternatively, computer-readable storage media can be a machine-readable signal medium. More specific examples of machine-readable storage media can include electrical connections based on one or more lines, portable computer disks, hard disks, random access memories (RAM), read-only memories (ROM), erasable programmable read-only memories (EPROM or flash memory), optical fibers, compact disc read-only memories (CD ROM), optical storage devices, magnetic storage devices, or any suitable combination of the foregoing.

[0133] In summary, the scalable video caching method and system for a low-Earth orbit satellite network provided by the embodiments of the present invention have the following beneficial effects:

[0134] We propose an online framework for SVC video caching in cooperative satellite networks that jointly optimizes tiered content allocation and downlink bandwidth allocation to minimize overall video access latency. The framework accounts for time-varying user requests, dynamic channel conditions, and limited satellite resources. By integrating bandwidth allocation into the caching process, it enables latency-aware decision-making under partial observation in a fully decentralized setting.

[0135] A multi-agent online learning algorithm based on a multi-armed bandit framework is developed, where each satellite independently selects cache actions based on local observations. Bandwidth is allocated using a Lagrangian duality approach to estimate user-perceived latency, which serves as a cache update reward based on the Upper Confidence Bound (UCB) for optimization under dynamic demand and decentralized coordination.

[0136] A comprehensive simulation evaluation is conducted to evaluate the performance of the proposed method and system. The experiments are conducted in a satellite network consisting of M = 2 LEO satellites, each of which is responsible for providing video content to a subset of K = 10 GUs. Each satellite is connected to a ground station (GS) via a direct communication link. The simulation is performed using the Satellite Tool Kit (STK). Each satellite orbits approximately 500 km above the Earth's surface and has a storage capacity of C m = 10Gbit. The video content library consists of N = 15 SVC videos, each with L = 4 quality layers. Video popularity follows a Zipf distribution with a skewness parameter θ = 0.5, indicating that some videos are significantly more likely to be requested than others. The sizes of each layer for each video are o n1 =400Mbit, o n2 =400Mbit, o n3 =400Mbit and o n4 =400Mbit.

[0137] The request pattern of GU is modeled as dynamically changing. The number of requests issued by each GU in time period t is The average number of requests per time period is 5. The dynamic change of requests is represented by Where D=5 represents the average number of requests per time period. Time variability is introduced, following a first-order Markov process, namely: Indicates the number of user requests for the kth GU in the t+1 time period Equal to the number of user requests in the tth time period (that is, the previous time period) Add a random variable e k,t+1 (used to indicate that the number of user requests in each time period is randomly generated and changes), here Indicates that this random variable obeys a normal distribution with a mean of 0 and a variance of 2. Here, Control the time dependency of GU requests for environmental factors, where represents independent requests across time periods, while This represents a time-invariant scenario where requests are highly correlated. Unless otherwise specified, the system parameters are set as follows: B=20MHz,B f =100MHz,B g =10MHz,P s =P g =30dBm,

[0138] To evaluate the performance of the proposed approach, this example compares it with several baseline strategies. In the random caching benchmark, each satellite randomly selects a video layer for local storage caching. The maximum probability request caching benchmark caches the video layer with the highest request probability. The genetic algorithm-based caching scheme treats caching decisions as individuals in a population, optimizing the total video access latency experienced by all relevant user terminals. Finally, the average bandwidth allocation scheme applies the multi-armed bandit (MAB) framework but uses an average bandwidth allocation strategy instead of dynamically optimizing bandwidth for each user.

[0139] Figure 2 The comparison of online optimization performance of video access delay under different schemes is shown. Figure 2 It shows that the average bandwidth allocation scheme performs the worst, highlighting the clear advantage of the Lagrangian-based bandwidth allocation scheme in this example. The maximum probability request caching scheme follows closely behind because it involves all satellites caching the video layer with the highest probability of request, which limits the effective satellite collaboration capabilities. In contrast, the algorithm proposed in this example significantly outperforms the baseline method. Although the initial latency is high due to the exploration phase in the online learning process, this exploration phase is crucial for the algorithm to collect enough information to transition to the utilization phase. As the understanding of the algorithm deepens, the satellite selects the video layer that can more effectively satisfy the GU cache request, thereby significantly reducing the long-term total video access latency. This example also The performance of the algorithm is evaluated in the case of , which represents a more dynamic environment. As expected, In contrast, latency decreases more slowly and ultimately results in higher latency. This suggests that in a dynamic environment, the algorithm faces more dramatic fluctuations in GU demand, making adaptation more challenging. To cope with these rapid changes, the algorithm requires more exploration and, therefore, takes longer to reach the optimal configuration. This further demonstrates the balance between exploration and exploitation in online learning systems.

[0140] Figure 3The study demonstrates the impact of satellite cache capacity on average video access latency under different scenarios. As expected, total latency decreases with increasing satellite cache capacity. Larger cache capacity means more video requests can be processed locally, reducing reliance on backhaul links and ultimately lowering content retrieval latency. This result highlights the importance of cache capacity in improving system performance by minimizing latency.

[0141] Figure 4 The study shows how increasing the average number of video layer requests from GUs affects total video access latency under different scenarios. As the number of video requests increases, total video access latency also increases. This is because more requests place greater demands on the network and satellite storage capacity, leading to increased congestion and latency. This finding highlights the trade-off between resource demands and the system's ability to efficiently handle such demands.

[0142] Figure 5 The impact of the skewness parameter on average total video access latency for different schemes is demonstrated. With the exception of the random caching method, all methods show a trend of decreasing latency as the skewness parameter increases. This trend can be attributed to the concentration of GU requests. Higher skewness indicates a greater concentration of requests, allowing the satellite to optimize its cache by storing only the most frequently requested video tiers. This optimization maximizes storage efficiency, enabling the satellite to effectively serve the needs of the majority of GUs, thereby reducing overall latency.

[0143] Simulation results demonstrate that the proposed method and system enable each satellite to adaptively balance exploration and exploitation under local observation, effectively responding to dynamic user demands and time-varying channel conditions. Compared to the baseline algorithm, the proposed method significantly reduces long-term video access latency.

[0144] The above embodiments are preferred implementation modes of the present invention, but the implementation modes of the present invention are not limited to the above embodiments. Any other changes, modifications, substitutions, combinations, and simplifications that do not deviate from the spirit and principles of the present invention should be considered as equivalent replacement methods and are included in the scope of protection of the present invention.

Claims

1. A scalable video caching method for a low earth orbit satellite network, characterized in that: Including steps: S1. Determine a caching model for a low-Earth orbit satellite network; the caching model includes a satellite network consisting of M low-Earth orbit satellites, each of which establishes a direct communication link with a ground station and is responsible for providing video content to K ground users within a coverage area; S2. Based on the cache model, construct a scalable video cache optimization problem, where the problem aims to minimize the total video access delay experienced by all terrestrial users, with a layered video content placement variable on the satellite and a downlink bandwidth allocation variable from the satellite to terrestrial users as solution variables, and with a storage constraint on each satellite, a binary constraint on the layered video content placement variable on the satellite, and a bandwidth allocation constraint on the satellite as constraints; S3. Use a multi-agent online learning algorithm to solve the scalable video cache optimization problem, and obtain a layered video content placement strategy on each satellite in each time period and a downlink bandwidth allocation strategy for each satellite to ground users.

2. The scalable video caching method for a low earth orbit satellite network according to claim 1, wherein: In step S1, in the caching model, when a ground user requests a video layer, the relevant satellite first checks whether its local cache contains the content. If the content is available, the satellite will directly transmit it to the ground user. Otherwise, the satellite forwards the query to the neighboring satellite via the intersatellite link and tries to obtain the missing layer from the nearest satellite that stores the layer. If the requested content is not cached in the entire satellite network, the satellite obtains the content from the ground station via the backhaul link and then transmits it to the ground user.

3. The scalable video caching method for a low earth orbit satellite network according to claim 2, wherein: In step S1, the cache model maintains a video content library consisting of N videos. The popularity of each video follows Zipf's law. Each video is encoded into L hierarchical quality layers using SVC. All ground users have the same preference for different quality layers. The cache model operates in T discrete time periods. In each time period t, the ground user numbered k generates a predetermined number of video layer requests. The cache state of layer l of video n on satellite m in time period t is represented by the binary variable Indicates that if the layer is cached locally, then Otherwise 0.

4. The scalable video caching method for a low earth orbit satellite network according to claim 3, wherein: In step S2, the objective function of the scalable video cache optimization problem is: in, is the downlink bandwidth allocated by satellite m to ground user k in time period t, To solve for the variable, φ nl is the request probability of the lth layer of video n, represents the set of ground users associated with satellite m, is the transmission delay between satellite m and ground user k if the video content has been cached on satellite m.

5. The scalable video caching method for a low earth orbit satellite network according to claim 4, wherein: The storage constraint for each satellite is: Among them, C m represents the storage budget of satellite m, It means arbitrary; The binary constraints on the layered video content placement variables on each satellite are: The bandwidth allocation constraint for each satellite is: B is the total downlink bandwidth available for each satellite.

6. The scalable video caching method for a low earth orbit satellite network according to claim 5, wherein: Step S3 specifically includes: S31, by removing the constant, the scalable video cache optimization problem is transformed into a given time period t and satellite m, the goal is to The total bandwidth B is distributed among them to minimize the overall expected access delay A new optimization problem, represents the effective video data load of ground user k associated with satellite m during time period t; S32. For the new optimization problem, a closed-form solution is obtained through Lagrange duality. form; S33, based on closed-form solution In the form of, a multi-armed bandit algorithm combined with an upper confidence bound algorithm is used to solve the scalable video cache optimization problem, and the optimal cache action and the optimal bandwidth allocation are obtained.

7. The scalable video caching method for a low earth orbit satellite network according to claim 6, wherein: Step S33 specifically includes the following steps: S331, initialization t = 1, the upper confidence bound of each action a of each satellite m is infinite, the average reward of each action a of each satellite m is 0, the cache budget of each satellite m is ω m C m , the action set of each satellite m S332: In each time period t, perform the following steps: For each satellite m, when ω m >0, select the optimal action and judge Is it true? If so, then a * From the action space Remove, if otherwise cache a * And update ω m for And a * From the action space Remove; According to the closed-form solution The optimal bandwidth allocation is obtained in the form of For each satellite m, for each of its selected actions a, calculate the reward that satellite m evaluates for each selected action based on the observed system performance at the end of time period t according to Update the reward for the next moment according to Update the upper confidence bound for the next moment Reset Enter the next time period.

8. The scalable video caching method for a low earth orbit satellite network according to claim 7, wherein: The upper confidence bound of each action a of each satellite m at each time period t is for: in, represents the number of times satellite m chooses action a until time period t, and represents the total number of actions performed by satellite m up to time period t, represents the empirical average reward of satellite m choosing action a up to time period t, Indicates the previous time period β represents the learning rate; Denotes the reward of each selected action a evaluated by satellite m based on the observed system performance at the end of time period t: in, represents the normalized reduction in the total video access delay for all GUs associated with satellite m at the end of time period t compared to the worst-case baseline, represents the maximum delay reduction experienced by a user terminal associated with satellite m at the end of time period t when the requested video layer is served from the local cache, represents the synergistic benefit brought by the cache content of satellite m being used to serve user terminals associated with other satellites through inter-satellite links at the end of time period t, and η1, η2, η3∈[0,1] are weight parameters that control the contribution of each component.

9. The scalable video caching method for a low earth orbit satellite network according to claim 8, wherein: Specifically: in, represents the total video access delay experienced by all GUs associated with satellite m at the end of time period t, is a normalizing constant; Specifically: in, is the number of times ground user k visits the quality layer corresponding to action a in time period t, o a represents the size of the video layer corresponding to action a, is the transmission delay per bit between satellite m and the ground station in time period t, is the transmission delay per bit between satellite m and ground user k in time period t, is the propagation delay per bit between satellite m and the ground station in time period t, is the propagation delay per bit between satellite m and ground user k in time period t; Specifically: Among them, j refers to other satellites, represents the set of ground users associated with satellite j, and k′ refers to the set of ground users associated with satellite j. Ground users in is the number of times ground user k′ visits the quality layer corresponding to action a within time period t.

10. A scalable video caching system for a low earth orbit satellite network, characterized in that: It includes a cache model construction module, an optimization problem construction module and an optimization problem solving module, which are respectively used to execute steps S1, S2 and S3 of the scalable video caching method for a low earth orbit satellite network as described in any one of claims 1 to 9.

Images