A 5G multipath redundancy transmission method and system based on spatiotemporal co-operation and digital twins

By constructing a nine-dimensional feature perception network and a digital twin, and dynamically adjusting multi-path transmission resources, the problems of insufficient prediction accuracy and rigid redundancy strategies in existing technologies are solved, and stable transmission of highly reliable and low-latency services is achieved.

CN123124050APending Publication Date: 2026-11-17AEROSPACE XINTONG TECH CO LTD
View PDF 3 Cites 0 Cited by

Patent Information

Application Number
CN202611525151.8
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2026-09-29
Publication Date
2026-11-17

AI Technical Summary

Technical Problem

Existing multipath transmission technologies suffer from insufficient prediction accuracy, rigid redundancy strategies, lagging mobility response, and lack of security adaptation, making it difficult to adapt to the dynamic changes in high-reliability, low-latency service scenarios.

Method used

A nine-dimensional feature perception network with three-level collaboration between the terminal, edge, and core network is constructed. The spatiotemporal joint quality index is updated using the Double DQN reinforcement learning algorithm. Terminal light twins, edge medium twins, and core heavy twins are established. Prediction is performed through spatiotemporal Transformer and conditional diffusion models to dynamically adjust the number of multipaths and resource allocation, and to implement differentiated security strategies.

Benefits of technology

It improves long-term forecast accuracy and resource utilization, reduces handover interruption time, and adapts to dynamic changes in high-reliability, low-latency service scenarios.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN123124050A_ABST
    Figure CN123124050A_ABST
Patent Text Reader

Abstract

This invention relates to the field of communication technology and discloses a 5G multi-path redundant transmission method and system based on spatiotemporal joint and digital twin. The method includes the following steps: S1, constructing a nine-dimensional feature sensing network with three-level collaboration between the terminal, edge, and core network to collect multi-dimensional data in real time; S2, calculating the spatiotemporal joint quality index ST-QI based on the data collected by the nine-dimensional feature sensing network, and updating the weight of each component in ST-QI according to a set period; S3, establishing a hierarchical digital twin; S4, calculating the dynamic redundancy K; S5, dynamically adjusting the number of multipaths and weighted allocating service data; S6, triggering Xn interface pre-construction, P4 flow table pre-configuration, and TSN microsecond-level scheduling through trajectory prediction adaptive time window; S7, updating quantum keys and performing differentiated blockchain auditing after priority allocation. This invention can achieve a synergistic improvement in prediction accuracy, redundancy adaptability, mobility handover continuity, and security level adaptability in 5G multi-path transmission.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] This invention relates to the field of communication technology, specifically to a 5G multipath redundancy transmission method and system based on spatiotemporal co-operation and digital twins. Background Technology

[0002] As 5G-A penetrates vertical industries such as industrial control, autonomous driving, and telemedicine, time-sensitive services place higher demands on the reliability, low latency, and security of communication links. Multi-path redundancy transmission improves reliability by carrying services on multiple links in parallel; however, existing mechanisms often rely on fixed ratios or localized states, making it difficult to adapt to dynamic changes in network topology, service priorities, and terminal mobility. When air interface quality, edge computing power, and backbone network load fluctuate rapidly, fixed strategies can easily lead to resource waste or insufficient protection for critical services. Therefore, a transmission method that integrates multi-dimensional sensing, spatiotemporal joint prediction, and dynamic redundancy decision-making is needed to provide a necessary foundation for subsequent solution design.

[0003] Currently, existing technologies have been explored in this area. For example, patent document CN120512723A discloses a data transmission optimization scheme that supports heterogeneous network access for 5G, NB-IoT, LoRa, and Wi-Fi 6, uses deep reinforcement learning for routing decisions, utilizes LSTM to predict network states, and improves reliability through multi-path parallelism and forward error correction. Patent document CN121396886A proposes a four-layer collaborative architecture consisting of a terminal layer, an edge node layer, a central cloud service layer, and a trusted chain layer. It employs sharded reinforcement learning to generate regional routing strategies, combines privacy-preserving federated learning for global optimization, and utilizes dynamic expert models and dual-gated network routing. Patent document CN121486863A, targeting low-latency vehicle-to-everything (V2X) networks, discloses link redundancy scheduling based on latency fluctuation entropy and redundancy requirement assessment matrices, as well as a dynamic adjustment scheme based on latency uncertainty assessment. These schemes provide a foundation for multi-path transmission and dynamic routing, but their prediction, decision-making, execution, and security still largely unfold along independent paths, resulting in limited collaborative depth.

[0004] Furthermore, existing multipath transmission still faces several challenges: (1) The prediction dimension is relatively limited, and a single model is difficult to balance short-term accurate prediction and long-term error correction, and the network topology dynamic modeling capability is insufficient; (2) Redundancy strategies are prone to rigidity, and the service value density, ST-QI change rate and historical reliability are not fully considered, and the reliability guarantee is insufficient in URLLC burst scenarios; (3) Mobility response is lagging, and the pre-construction preparation and resource reservation in high-speed scenarios are difficult to meet the seamless switching requirements; (4) Security mechanisms mostly adopt fixed-period key updates and are not adapted according to the service security level; (5) There is insufficient cross-level twin data interaction, and it is difficult for the terminal, edge and core network to form an end-to-end closed-loop optimization.

[0005] Therefore, how to organically integrate spatiotemporal joint prediction, hierarchical digital twins, dynamic redundant decision-making, microsecond-level mobility collaboration, and priority-based intrinsic security has become a technical problem that urgently needs to be solved in this field. Summary of the Invention

[0006] This invention aims to provide a 5G multipath redundancy transmission method and system based on spatiotemporal co-operation and digital twins, which effectively solves the problems of insufficient prediction accuracy, rigid redundancy strategy, lag in mobility response and lack of security adaptation in existing multipath transmission technologies. It reduces handover interruption time, improves long-term prediction accuracy and resource utilization, and is suitable for high-reliability and low-latency service scenarios.

[0007] To achieve the above objectives, the basic solution provided by this invention is as follows: Option 1 The 5G multipath redundancy transmission method based on spatiotemporal co-operation and digital twins includes the following steps: S1. Construct a nine-dimensional feature perception network with three-level collaboration between terminal, edge, and core network. Collect in real time the air interface parameters, motion status, service characteristics, and energy consumption of the terminal side, the wireless resources, computing resources, and MEC computing power of the edge side, and the network topology and SRv6 tunnel status of the core network side. Then, obtain the backbone network load index based on the entropy weight method. S2. Calculate the spatiotemporal joint quality index ST-QI based on the data collected by the nine-dimensional feature perception network, and update the weights of each component in ST-QI according to a set period using the Double DQN reinforcement learning algorithm. S3. Establish a hierarchical digital twin including a terminal light twin, an edge twin, and a core heavy twin; the edge twin uses a spatiotemporal Transformer to predict short-term ST-QI of 0-100ms and a conditional diffusion model to correct errors in long-term prediction results of 100ms-1s; the core heavy twin is based on network simulation to perform strategy evaluation and cross-domain collaboration. S4. Calculate the dynamic redundancy K based on the ST-QI forecast value, ST-QI change rate, business priority, business value density, mobility factor, and historical reliability factor. S5. Dynamically adjust the number of multi-paths according to the dynamic redundancy K, and allocate service data according to the ST-QI weight of each path, wherein the number of paths is dynamically adjusted between 1 and 8. S6. Adaptively adjust the prediction window based on the terminal trajectory prediction result, and trigger the pre-construction of Xn interface, pre-configuration of P4 flow table and reservation of target base station resources according to the prediction window. Perform multi-path resource scheduling at the microsecond level through the time-sensitive network scheduler. S7. Perform quantum key hierarchical updates according to business priorities, and perform differentiated blockchain auditing and intrusion-tolerant transmission according to data security levels.

[0008] Option 2 A 5G multipath redundancy transmission system based on spatiotemporal co-operation and digital twins is used to execute the 5G multipath redundancy transmission method based on spatiotemporal co-operation and digital twins as described in Scheme 1; including: A nine-dimensional feature-based cross-domain collaborative sensing network is used to execute S1; The spatiotemporal joint mass index optimization module is used to execute S2; A hierarchical digital twin prediction engine, comprising a terminal light twin unit, an edge twin unit, and a core heavy twin unit, is used to execute S3; The business adaptive redundancy decision module is used to execute S4 and S5; The mobility collaborative execution unit includes a trajectory prediction subunit, a time window adjustment subunit, a path pre-construction subunit, and a resource scheduling subunit, which are used to execute S6; The dynamic intrinsic security protection module includes a key management subunit, a blockchain auditing subunit, and an intrusion tolerance subunit, which are used to execute S7.

[0009] The working principle and advantages of this invention are as follows: This invention presents a 5G multi-path redundant transmission method and system based on spatiotemporal co-location and digital twins. It effectively solves problems in existing multi-path transmission technologies such as insufficient prediction accuracy, rigid redundancy strategies, lag in mobility response, and lack of security adaptation. It reduces handover interruption time, improves long-term prediction accuracy and resource utilization, and is suitable for high-reliability, low-latency service scenarios. The key points are: First, this invention utilizes a nine-dimensional feature perception network with three levels of collaboration—terminal, edge, and core network—to incorporate multi-dimensional data into unified collection and fusion, and calculates the ST-QI based on this data. Compared to schemes that rely solely on sequence embedding or single network metrics, this approach more comprehensively reflects link quality, service requirements, and computing power status. Double DQN is employed to dynamically update the weights of each component of the ST-QI periodically, adjusting the importance of factors such as latency, jitter, bandwidth, reliability, and computing power according to changes in service type and network status. Consequently, subsequent redundancy decisions no longer rely on fixed weights but are based on a unified and dynamic quality evaluation, which improves the policy's adaptability to highly dynamic scenarios and avoids decision-making biases caused by insufficient perception dimensions or rigid weights.

[0010] Second, this invention establishes a three-layer digital twin architecture: a terminal light twin, an edge twin, and a core twin, enabling a progressive prediction and simulation system from the terminal to the core network. The terminal side is responsible for low-overhead, short-term prediction; the edge side is responsible for accurate short-term prediction and uses generative models to correct long-term errors; and the core side is responsible for policy simulation and cross-domain evaluation. Compared to a single prediction model, this architecture can simultaneously meet real-time and forward-looking requirements, and forms a closed loop through inter-layer data interaction, improving prediction availability, supporting proactive optimization, and helping to reduce redundant configuration errors caused by prediction bias.

[0011] Third, this invention calculates the dynamic redundancy K by integrating multiple indicators and adjusts the number of multipaths accordingly. It no longer employs a fixed segmentation ratio or a weighted strategy based solely on transmission rate, but instead incorporates service value density and historical transmission success rate into redundancy decisions, enabling network resources to be allocated to high-value, high-reliability services. Furthermore, it allocates service data according to the weighted result of the ST-QI of each path. Compared to multipath schemes with a fixed number of paths or static weights, this approach ensures that data allocation matches the real-time quality of the paths and allows for smooth adjustments when service demands change. Compared to static redundancy schemes, this method is more adaptable to URLLC burst scenarios and prioritizes truly important services when resources are limited, thus achieving a balance between reliability and resource utilization.

[0012] Fourth, this invention adaptively adjusts the time window through trajectory prediction, triggering path pre-construction, flow table pre-configuration, and resource reservation in advance, and executing them with fine-grained scheduling. Compared to schemes that rely on network triggering or fixed pre-construction durations, this approach is better suited to high-speed mobile scenarios, reducing handover waiting and interruption risks, and ensuring the continuity of time-sensitive services. Simultaneously, the linkage between the prediction time window and pre-construction and reservation actions enables mobility management to shift from passive response to proactive preparation, improving the stability of end-to-end transmission.

[0013] Fifth, this invention performs hierarchical updates of quantum keys according to business priorities and implements differentiated blockchain auditing and intrusion-tolerant transmission according to data security levels. Compared to fixed-period key updates and unified auditing strategies, this approach matches security overhead with business security levels, better meets the differentiated security needs of different businesses, and is also conducive to balancing security strength and real-time transmission performance. Attached Figure Description

[0014] Figure 1 This is a schematic diagram of the structure of an embodiment of the 5G multipath redundancy transmission method and system based on spatiotemporal co-operation and digital twin of the present invention. Detailed Implementation

[0015] The following detailed explanation illustrates the specific implementation methods: The basic implementation examples are as follows: Figure 1As shown: A 5G multipath redundancy transmission method based on spatiotemporal co-operation and digital twins includes the following steps: S1. Construct a nine-dimensional feature perception network with three-level collaboration between the terminal, edge, and core network. Collect in real time the air interface parameters, motion status, service characteristics, and energy consumption of the terminal side, the wireless resources, computing resources, and MEC computing power of the edge side, and the overall network topology and SRv6 tunnel status of the core network side. Then, obtain the backbone network load index based on the entropy weight method.

[0016] In the nine-dimensional feature perception network, the terminal side deploys an air interface parameter perception module, a motion state perception module, a service feature extraction module, and an energy consumption monitoring module. The air interface parameter perception module uses multi-antenna joint detection technology to collect RSRP, SINR, and latency jitter. Specifically, the RSRP acquisition range is -140 to -44 dBm with an accuracy of ±0.5 dB; the SINR acquisition range is -23 to 40 dB with an accuracy of ±0.2 dB; and the latency jitter sampling frequency is 2 ms / time with an accuracy of ±0.1 μs.

[0017] The motion state sensing module collects motion data through a 9-axis IMU and a dual-frequency GPS. The 9-axis IMU includes an accelerometer and a gyroscope, with the accelerometer range being ±16g and the gyroscope range being ±2000° / s; the dual-frequency GPS uses the L1+L5 frequency bands and has a positioning accuracy of 0.1m.

[0018] The business feature extraction module parses the data packet header information in real time to obtain the QCI level and a 3-bit data importance identifier. The energy consumption monitoring module uses a TI INA226 chip to measure power consumption with an accuracy of ±0.1% and a sampling frequency of 1kHz.

[0019] The edge layer deploys a wireless resource monitoring module, a computing resource status monitoring module, and a MEC computing power monitoring module. The wireless resource monitoring module monitors PRB utilization at a 1% granularity and a 1ms period, and calculates interference levels based on RSRQ. The computing resource status monitoring module collects CPU / GPU utilization, memory usage, and storage IOPS at a 10ms period. The MEC computing power monitoring module uses the NVIDIA DCGM toolkit to monitor computing power redundancy in real time, in TOPS.

[0020] The core network side collects the entire network topology information every 100ms through the BGP-LS protocol of the SDN controller; monitors the SRv6 tunnel status in real time, including SID reachability, forwarding latency and packet loss rate; and calculates the backbone network load index based on the entropy weight method by fusing 12-dimensional indicators such as bandwidth utilization, latency, and jitter.

[0021] Furthermore, an improved OPTICS clustering algorithm is used to analyze business characteristics in real time. The parameters of the improved OPTICS clustering algorithm are ε=0.15 and MinPts=20. Under this clustering condition, if the current business flow is clustered into the industrial control business cluster, and its QCI level belongs to the preset industrial control business level set, and the value of the 3-bit data importance identifier of the current business flow is not lower than the preset importance threshold, then the current business flow is determined to be an industrial control business, and the ST-QI weight is adjusted accordingly: the delay factor in ST-QI is adjusted... Increased to 0.6, and bandwidth factor The weight is reduced to 0.05. If any of the above conditions are not met, the current service flow is determined to be another service type, the current ST-QI weight is maintained, or the ST-QI weight is updated by Double DQN at a set period.

[0022] The preset industrial control business level set and preset importance threshold can be pre-configured.

[0023] S2. Calculate the spatiotemporal joint quality index ST-QI based on the data collected by the nine-dimensional feature perception network, and use the Double DQN reinforcement learning algorithm to update the weights of each component in ST-QI at a set period.

[0024] The ST-QI is calculated according to the following formula: ; ; in, These are the weighting coefficients. For mass components; This is an environmental noise correction term based on historical error feedback learning.

[0025] The specific mass components are as follows: , which is the time delay stability component; This is the time delay variation coefficient calculated using the Welford online algorithm based on the most recent 10 sampling points. The smaller, The larger the value, the more stable the latency.

[0026] , where is the jitter suppression component; J(t) is the time delay jitter calculated using a 20-sample sliding window and exponentially decaying weights. This serves as the jitter normalization baseline or maximum permissible jitter value, used to map jitter to a comparable normalization range. The smaller J(t), the better. The larger the value, the better the jitter suppression.

[0027] , which is the bandwidth availability component; The available bandwidth is predicted based on the ARIMA(2,1,2) model, with a predicted horizontal value of 5. ARIMA(2,1,2) indicates that the autoregressive order is 2, the differencing order is 1, and the moving average order is 2. This is the nominal bandwidth or reference bandwidth, used to normalize the predicted available bandwidth. The closer to or beyond , The larger the value, the better the bandwidth availability.

[0028] , which is the reliability component; To adopt The packet loss rate is calculated using the exponentially weighted moving average. As a smoothing factor, This indicates that recent packet loss rates are given a certain weight in the smoothing calculation; The smaller, The larger the value, the higher the reliability.

[0029] , which is the computing power availability component; This represents the remaining computational resources for edge nodes after normalization using the sigmoid function.

[0030] The Double DQN reinforcement learning algorithm is used to update the weight coefficients every 50ms. The state space includes service type, network load level, mobility status, and time characteristics; the action space includes... The adjustment step size is 0.05, and the reward function is determined based on the latency score, reliability score, and resource utilization score, i.e.: reward function .

[0031] Latency score Determined according to the following formula:

[0032] in, For the current end-to-end delay, The maximum allowable latency for the business; Reliability score Determined according to the following formula: (t) in (t) represents the exponentially weighted moving average packet loss rate; Resource utilization score Determined according to the following formula:

[0033] in, This refers to the utilization rate of PRB resources.

[0034] S3. Establish a hierarchical digital twin including a terminal light twin, an edge twin, and a core heavy twin; the edge twin uses a spatiotemporal Transformer to predict short-term ST-QI (0-100ms) and a conditional diffusion model to correct errors in long-term prediction results (100ms-1s); the core heavy twin is based on network simulation to perform strategy evaluation and cross-domain collaboration. Specifically: The terminal's lightweight twin employs a lightweight ST-GCN, with parameters not exceeding 500kB and computational complexity not exceeding 0.5 TOPS. Its input features include GPS coordinates (longitude, latitude, and elevation), IMU data (three-axis acceleration and angular velocity), and environmental characteristics. Its output is a trajectory prediction for the next 50ms, with an error not exceeding 0.8m and a confidence interval of 95%. Power consumption control utilizes dynamic voltage and frequency adjustment technology, with a maximum power consumption not exceeding 100mW.

[0035] Preferably, the graph structure of the lightweight ST-GCN is constructed based on a spatiotemporal graph. The node set V includes historical trajectory point nodes and sensor feature nodes; the historical trajectory point nodes correspond to the terminal positions at several past moments, and the sensor feature nodes correspond to GPS, IMU, and environmental features. The edge set E includes temporal edges and spatial edges: temporal edges connect nodes of the same type at adjacent moments to capture trajectory temporal dependencies; spatial edges connect different sensor nodes or trajectory point nodes at the same moment to fuse multi-source motion features. The graph structure can be represented as follows: Where A is the adjacency matrix. The adjacency matrix can be determined based on the Euclidean distance between nodes, sensor physical association, or learnable attention weights. Based on this spatiotemporal graph structure, ST-GCN aggregates node features in both the temporal and spatial dimensions and outputs the terminal trajectory prediction results for the next 50ms.

[0036] The training data for this network includes: historical trajectory datasets from terminals, measured data from IMU and GPS, environmental features from high-precision maps, and simulation data generated through NS-3 or trajectory simulation platforms. The training method employs end-to-end supervised learning, using the real trajectory from the next 50ms as the label. The loss function can use mean squared error or negative log-likelihood, and lightweight training is achieved by incorporating dynamic voltage and frequency adjustment constraints, such as adding parameter constraints, computational constraints, or power constraints to the loss function. After training, the model has no more than 500kB of parameters and no more than 0.5 TOPS of computation, and is deployed on a lightweight twin of the terminal to perform trajectory prediction.

[0037] The edge twin includes a short-term prediction module and a long-term prediction module.

[0038] The short-term prediction module covers a prediction range of 0-100ms, employing a 12-layer spatiotemporal Transformer with an embedding dimension of 256 and 8 attention heads. Inputs include historical ST-QI sequences, a network topology adjacency matrix, and a service load vector; the preferred length of the historical ST-QI sequence is 50. The output of the short-term prediction module is the ST-QI prediction sequence for the next 100ms, which can be used as conditional input for subsequent long-term prediction modules and for calculating the dynamic redundancy K.

[0039] In this embodiment, the network topology adjacency matrix is ​​constructed as follows: based on the network topology information collected by the SDN controller every 100ms via the BGP-LS protocol, gNBs, MEC nodes, UPF / core network gateways, and SRv6 SID endpoints participating in multipath transmission are taken as graph nodes, and physical links, Xn interfaces, SRv6 tunnels, or available multipath links between nodes are taken as edges; the (i,j)th element of the adjacency matrix A(t) The SID is determined by weighting based on link reachability, latency, jitter, packet loss rate, available bandwidth, and SRv6 SID reachability, and normalized to [0,1]. If unreachable, it is set to 0.

[0040] The adjacency matrix is ​​dynamically refreshed according to the BGP-LS topology update cycle or sliding window, and then input into the spatiotemporal Transformer after being aligned with the historical ST-QI sequence in the time dimension.

[0041] The service load vector is constructed as follows: edge-side PRB utilization, CPU utilization, GPU utilization, memory utilization, storage IOPS and MEC computing power margin are collected, as well as the backbone network load index obtained by fusion based on the entropy weight method on the core network side. After normalizing each component to [0,1], they are concatenated in a fixed order to form the service load vector L(t).

[0042] For data with different sampling periods, time alignment and interpolation are performed based on the ST-QI sampling time. If necessary, service type identifiers can be added according to service type or QCI level so that L(t) together with the historical ST-QI sequence and the network topology adjacency matrix constitute the input of the spatiotemporal Transformer.

[0043] The long-term prediction module covers a prediction range of 100ms-1s and employs a conditional diffusion model. The conditional diffusion model uses a 1000-step denoising process, with noise scheduling using the cosine rule. Inputs include short-term prediction results, environmental context features, and historical error patterns; the output is the corrected long-term prediction trend. By using this conditional diffusion model to correct errors in short-term prediction results, the long-term prediction accuracy is improved by 15%.

[0044] In this embodiment, the conditional injection method of the conditional diffusion model is as follows: short-term prediction results, environmental context features, and historical error patterns are encoded into conditional vectors, and then injected into the denoising network through a cross-attention mechanism and adaptive feature modulation. Specifically, short-term prediction results are encoded into short-term conditional vectors through one-dimensional convolution; environmental context features are mapped into environmental conditional vectors through fully connected layers; historical error patterns are generated into error conditional vectors through a temporal encoder; in each step of the 1000-step denoising process, cross-attention is performed with noise latent variables as queries and conditional vectors as keys and values, and the intermediate features of the denoising network are modulated through adaptive layer normalization or FiLM, so that the back-diffusion process gradually generates the corrected long-term ST-QI trend sequence under cosine noise scheduling; Historical error patterns are represented as time-series sequences of long-term prediction errors and their statistical characteristics, including the error vector, mean error, variance error, autocorrelation coefficient error, and rate of change error between the actual ST-QI and the predicted ST-QI for each prediction. These historical error patterns can be stored in buckets according to service type, mobility status, network load level, and time window to form a historical error library. During input, the most recent N error sequences are encoded into fixed-dimensional error embedding vectors using a one-dimensional convolutional network or LSTM, and then concatenated or fused with short-term prediction results and environmental context features. This serves as one of the conditional inputs to the conditional diffusion model, guiding the error correction of long-term prediction results.

[0045] The core heavy twin is based on the NS-3 extended simulation platform, supporting parallel simulation with more than 10,000 nodes, and its functions include: Network state reconstruction: Recovering the entire network state from limited observation data based on compressed sensing technology; Strategy evaluation: Multiple redundancy strategies are tested in parallel, and Monte Carlo tree search is used for optimization. The evaluation time is no more than 8ms. Cross-domain collaboration: Supports joint optimization of backbone networks from multiple operators and achieves privacy protection based on federated learning.

[0046] S4. Calculate the dynamic redundancy K based on the ST-QI forecast value, ST-QI change rate, business priority, business value density, mobility factor, and historical reliability factor.

[0047] First, construct a seven-dimensional business feature profile, including: Latency requirements: End-to-end latency limit, in milliseconds, categorized as Extreme <10ms, High <20ms, and Normal <50ms; Bandwidth requirement: Peak / average bandwidth ratio, categorized as burst >5, fluctuation 2-5, and stable <2; Jitter tolerance: Maximum permissible jitter value, in milliseconds, categorized as strict (<1ms), moderate (<5ms), and lenient (<10ms); Priority: 3GPP QCI levels 1-9, mapped to 0-1 normalized values; Data Importance: Security Level 1-3, categorized based on data sensitivity; Movement modes: Static, Low speed <30km / h, Medium speed 30-120km / h, High speed >120km / h; Business value density: the ratio of data importance to transmission cost, normalized to [0,1].

[0048] The importance of data can be determined based on the 3-bit data importance identifier obtained by the business feature extraction module. Specifically, the value of the 3-bit data importance identifier is mapped to a value between 0 and 1. For example, a linear mapping method is used so that the larger the identifier value, the higher the data importance. The mapped value is denoted as the data importance component I, and I∈[0,1].

[0049] Transmission cost is determined based on the network resource costs of service transmission, including at least one of the following: bandwidth requirement, end-to-end latency requirement, jitter tolerance, packet loss sensitivity, path resource utilization, MEC computing power consumption, and SRv6 tunnel overhead. After normalizing each of these components to [0,1], they are weighted and summed according to preset weights to obtain the transmission cost C, where C∈[0,1]. The higher the bandwidth requirement, the stricter the end-to-end latency requirement, the lower the jitter tolerance, the higher the packet loss sensitivity, the higher the path resource utilization, the greater the MEC computing power consumption, or the greater the SRv6 tunnel overhead, the higher the transmission cost C. The weights of each component can be pre-configured or dynamically adjusted according to the service type or network status.

[0050] The dynamic redundancy K is calculated according to the following formula: ; in, This is the predicted value for ST-QI. The rate of change of ST-QI For business priority, M is the service value density, M is the mobility factor (with a value of 0-1, based on speed mapping), and H is the historical reliability factor (determined based on the success rate of the past 100 transmissions). .

[0051] The dynamic redundancy K is a continuous value with upper and lower bound constraints. All input components of K are normalized to [0,1], and the coefficients... Since each value is limited to a preset interval, the theoretical range of K is [0,1]. For ease of engineering implementation, K can be truncated, i.e., K=min(1,max(0,K)), so that it always lies within the [0,1] interval. The larger the value of K, the higher the service's demand for redundant transmission, and more paths should be configured; the smaller the value of K, the lower the redundancy requirement, and the fewer paths can be used to save resources.

[0052] In this invention, a path refers to an end-to-end logical transmission channel from the source terminal to the target node (which may be an edge node, MEC node, UPF, core network gateway, or application server depending on the business scenario), capable of independently carrying business data streams and participating in multi-path redundant transmission. The path is composed of one or more links, interfaces, or tunnels connected in series, and has a unique path identifier and a corresponding ST-QI attribute.

[0053] In this embodiment, the mobility factor M is calculated and mapped as follows: the terminal speed v (unit km / h) is acquired in real time and normalized to [0,1] using a piecewise linear mapping or a continuous function; preferably, a speed classification consistent with the mobility mode is used for piecewise mapping, i.e., when v=0 at static speed, M=0; when at low speed... Let M be linearly mapped between 0.1 and 0.3; at medium speed Let M be linearly mapped between 0.3 and 0.6; at high speed Let M be linearly mapped between 0.6 and 0.9; at extremely high speeds... Let M=1. In the above piecewise mapping, the specific values ​​in each interval can be determined by linear interpolation based on the relative position of the velocity within that interval, thus making M change smoothly with the terminal velocity.

[0054] The historical reliability factor H is calculated as follows: A sliding window of length 100 is maintained, recording the success or failure of the most recent 100 service transmissions. Each successful transmission is recorded as 1, and each failure as 0. Whenever a new service transmission is completed, the result is written into the sliding window, and the oldest record is removed, keeping the window length constant at 100. The historical reliability factor H is equal to the ratio of the number of successful transmissions within the sliding window to the total number of transmissions (100), i.e., H = (number of successful transmissions) / 100, therefore H ∈ [0, 1]. The more successful transmissions, the closer H is to 1, indicating higher historical reliability; conversely, the closer H is to 0, the lower the historical reliability.

[0055] The business value density is determined based on the ratio of data importance to transmission cost and normalized to [0,1].

[0056] S5. Dynamically adjust the number of multi-paths according to the dynamic redundancy K, and allocate service data according to the ST-QI weight of each path, wherein the number of paths is dynamically adjusted between 1 and 8.

[0057] The data weights for each path are calculated using the following formula: ; in, This is the sharpening factor related to the business type; the default value is 2.0. The number of paths is adjusted between 1 and 8 based on the dynamic redundancy K, and a smooth transition algorithm and transition buffer are used to perform path switching, with a switching response time of no more than 10ms. Through the above multi-path management, resource utilization can be improved while ensuring the reliability of time-sensitive services.

[0058] Specifically, the mapping relationship between the number of paths and K is as follows: Let the number of paths be N, The K can be mapped to N using either a piecewise or continuous mapping. A piecewise mapping is preferred. when At that time, N=1; when At that time, N=2; when At that time, N=3; when At that time, N=4; when At that time, N=5; when At that time, N=6; when N=7; when At that time, N=8.

[0059] The above mapping allows the number of paths to change continuously and remain stable as redundancy requirements increase.

[0060] The smooth transition algorithm includes the following steps: The first step is to calculate the number of target paths based on the current K value. and the current number of paths Compare; The second step, if Then, select the highest ST-QI paths from the candidate paths as new paths, and establish the session state, flow table entries and key context of the new paths in the transition buffer. The third step, Ruoruo If the path to be deleted is marked as "in transition", it will not be released immediately, but its data allocation weight will be gradually reduced. The fourth step is to adjust the data allocation ratio of each path according to the linear or exponential gradual change of the weights of the old and new paths during the transition period, so that the business data can be smoothly migrated from the old path to the new path. Fifth, after the transition period ends, release the resources of the path to be deleted to complete the path switch. The preferred transition period is 5ms to 10ms, which can be adjusted according to business latency requirements.

[0061] The transition buffer is set in the multipath management unit or the sender's buffer and is used to temporarily store data packets to be reallocated during path switching. Its size is determined based on the service rate and transition period, and is preferably capable of accommodating... Bits of data, where R is the service rate. For a transition period, for example, when the service rate is 100Mbps and the transition period is 10ms, the buffer size should be no less than 1Mbit. The triggering conditions for the transition buffer include: the number of target paths is inconsistent with the number of current paths, the ST-QI of a certain path is lower than a preset threshold, a path fails, the service priority or service value density changes, and the rate of change of the K value exceeds a preset threshold. When the triggering conditions are met, the smooth transition algorithm is activated, and the data in the transition buffer is redistributed according to the new weights.

[0062] The measurement method for a switching response time of no more than 10ms is as follows: The time interval is measured from the moment the path switching is triggered to the moment the business data is stably transmitted on the new path and the resources of the old path are fully released. Measurement scenarios include path addition, path reduction, path failure, business priority switching, and sudden changes in the K-value. Measurement points can be set at the sending end's multi-path management unit, recording the trigger and completion times with timestamps, and statistically analyzing the response times of multiple switches, taking the maximum value or the 95th percentile as the indicator. If a path fails, its weight is immediately reset to zero, and the data it carries is redistributed according to the ST-QI weights of the remaining paths. Simultaneously, the path with the highest ST-QI is selected from the candidate paths to supplement it. If the number of remaining paths is insufficient, a path pre-construction and resource reservation process is triggered to ensure that the switching response time does not exceed 10ms.

[0063] S6. Adaptively adjust the prediction window based on the terminal trajectory prediction result, and trigger Xn interface pre-construction, P4 flow table pre-configuration, and target base station resource reservation according to the prediction window. Multi-path resource scheduling is then performed at a microsecond granularity using a time-sensitive network scheduler. Specifically: First, intelligent trajectory prediction is performed using a fusion algorithm of adaptive Kalman filtering and Attention-LSTM. Input features include GPS location, IMU data, high-precision map information, and traffic flow data; the GPS location uses WGS84 coordinates, and the IMU data uses the vehicle coordinate system. The output is a trajectory prediction for the next 1 second, with an error of no more than 0.5m, measured using RMS. The prediction window is adaptively adjusted according to the terminal speed: 50ms for low speed (<30km / h); 65ms for medium speed (30-120km / h); 80ms for high speed (>120km / h); and 100ms for extreme high speed (>200km / h).

[0064] Based on the prediction window, the mobility coordination mechanism is triggered as follows: through pre-construction via the Xn interface, direct communication between gNBs is enabled, reducing core network signaling overhead, and the pre-construction delay is less than 5ms; through P4 flow table pre-configuration, flow table entries are pre-configured in the programmable data plane, supporting nanosecond-level activation; through resource reservation, the target base station reserves PRB resources in advance, with a reservation duration of 1s, and a soft reservation mechanism is adopted.

[0065] For resource scheduling, an enhanced scheduler based on time-sensitive networking is adopted, with a scheduling granularity of 20μs and support for 32 priority queues. Adaptations are made for different services: industrial robots have a cycle time of 1ms and a jitter requirement of less than 10μs; autonomous driving has a cycle time of 10ms and a reliability requirement of greater than 99.999%; and telemedicine has a cycle time of 5ms and a latency requirement of less than 20ms. Through the above-mentioned mobility-coordinated execution, the risk of handover interruptions in high-speed scenarios can be reduced, meeting the requirements of low-latency and high-reliability services.

[0066] S7. Perform hierarchical updates of quantum keys according to business priorities, and implement differentiated blockchain auditing and intrusion-tolerant transmission according to data security levels. Specifically: The post-quantum key hierarchical update adopts an improved CRYSTALS-Kyber-1024 post-quantum encryption algorithm; among them, the highest priority services, such as medical control, adopt a 10ms key update cycle and perform forward security protection; medium priority services, such as video transmission, adopt a 50ms key update cycle and perform two-way authentication; ordinary services, such as data acquisition, adopt a 100ms key update cycle and perform lightweight authentication; key distribution is based on blockchain and adopts threshold signature technology.

[0067] The differentiated blockchain auditing is based on Hyperledger Fabric light nodes and state channels, using Reed-Solomon erasure coding for storage. The erasure coding form is 10+6, which reduces storage overhead by 60%. The on-chain strategy is as follows: high-density data is on-chain every 5ms and uses full-replica storage and digital signature verification; ordinary data is on-chain every 100ms and uses erasure coding storage and batch authentication. Performance indicators are: memory usage no more than 512MB, throughput no less than 2000TPS, and latency less than 10ms.

[0068] The intrusion-tolerant transmission employs anomaly detection combining deep packet inspection and deep learning, with anomaly detection time not exceeding 3ms. The backup encrypted channel uses hot backup, with an activation time not exceeding 5ms. The self-healing mechanism supports automatic key updates and path reconstruction, with a recovery time of less than 10ms. Through the above security protections, security overhead can be adapted to the business security level as needed, avoiding the risk of key leakage in high-security scenarios under fixed-period key update mechanisms.

[0069] This embodiment also provides a 5G multipath redundancy transmission system based on spatiotemporal coherence and digital twins, used to execute the 5G multipath redundancy transmission method based on spatiotemporal coherence and digital twins as described above; including: A nine-dimensional feature-based cross-domain collaborative sensing network is used to execute S1; The spatiotemporal joint mass index optimization module is used to execute S2; A hierarchical digital twin prediction engine, comprising a terminal light twin unit, an edge twin unit, and a core heavy twin unit, is used to execute S3; The business adaptive redundancy decision module is used to execute S4 and S5; The mobility collaborative execution unit includes a trajectory prediction subunit, a time window adjustment subunit, a path pre-construction subunit, and a resource scheduling subunit, which are used to execute S6; The dynamic intrinsic security protection module includes a key management subunit, a blockchain auditing subunit, and an intrusion tolerance subunit, which are used to execute S7.

[0070] Furthermore, in the hierarchical digital twin prediction engine, the terminal light twin unit deploys a lightweight ST-GCN, the edge twin unit deploys a spatiotemporal Transformer and a conditional diffusion model, and the core heavy twin unit deploys a simulation platform based on NS-3 extensions.

[0071] The business adaptive redundancy decision module includes a business profile construction unit, a redundancy calculation unit, and a multi-path management unit. The business profile construction unit constructs a seven-dimensional business feature profile including latency requirements, bandwidth requirements, jitter tolerance, priority, data importance, mobility mode, and business value density.

[0072] In the dynamic intrinsic security protection module, the key management subunit performs priority-based quantum key updates, the blockchain audit subunit performs differentiated on-chain and erasure coding storage, and the intrusion tolerance subunit performs anomaly detection, emergency switching, and self-healing recovery.

[0073] This embodiment provides a 5G multipath redundancy transmission method and system based on spatiotemporal co-operation and digital twins, which effectively solves the problems of insufficient prediction accuracy, rigid redundancy strategy, lag in mobility response and lack of security adaptation in existing multipath transmission technologies. It reduces handover interruption time, improves long-term prediction accuracy and resource utilization, and is suitable for high-reliability and low-latency service scenarios.

[0074] The above descriptions are merely embodiments of the present invention. Commonly known structures and characteristics of the solutions are not described in detail here. Those skilled in the art are aware of all common technical knowledge in the field prior to the application date or priority date, are aware of all existing technologies in that field, and have the ability to apply conventional experimental methods prior to that date. Those skilled in the art can, under the guidance of this application, improve and implement this solution in combination with their own capabilities. Some typical known structures or methods should not be obstacles for those skilled in the art to implement this application. It should be noted that those skilled in the art can make several modifications and improvements without departing from the structure of the present invention. These should also be considered within the scope of protection of the present invention, and will not affect the effectiveness of the implementation of the present invention or the practicality of the patent.

Claims

1. A 5G multi-path redundant transmission method based on space-time joint and digital twinning, characterized in that, Includes the following steps: S1. Construct a nine-dimensional feature perception network with three-level collaboration between terminal, edge, and core network. Collect in real time the air interface parameters, motion status, service characteristics, and energy consumption of the terminal side, the wireless resources, computing resources, and MEC computing power of the edge side, and the network topology and SRv6 tunnel status of the core network side. Then, obtain the backbone network load index based on the entropy weight method. S2. Calculate the spatiotemporal joint quality index ST-QI based on the data collected by the nine-dimensional feature perception network, and update the weights of each component in ST-QI according to a set period using the Double DQN reinforcement learning algorithm. S3. Establish a hierarchical digital twin including a terminal light twin, an edge twin, and a core heavy twin; the edge twin uses a spatiotemporal Transformer to predict short-term ST-QI of 0-100ms and a conditional diffusion model to correct errors in long-term prediction results of 100ms-1s; the core heavy twin is based on network simulation to perform strategy evaluation and cross-domain collaboration. S4. Calculate the dynamic redundancy K based on the ST-QI forecast value, ST-QI change rate, business priority, business value density, mobility factor, and historical reliability factor. S5. Dynamically adjust the number of multi-paths according to the dynamic redundancy K, and allocate service data according to the ST-QI weight of each path, wherein the number of paths is dynamically adjusted between 1 and 8. S6. Adaptively adjust the prediction window based on the terminal trajectory prediction result, and trigger the pre-construction of Xn interface, pre-configuration of P4 flow table and reservation of target base station resources according to the prediction window. Perform multi-path resource scheduling at the microsecond level through the time-sensitive network scheduler. S7. Perform quantum key hierarchical updates according to business priorities, and perform differentiated blockchain auditing and intrusion-tolerant transmission according to data security levels.

2. The 5G multi-path redundancy transmission method based on spatiotemporal co-location and digital twin as described in claim 1, characterized in that, In S1, the service characteristics are analyzed in real time by using an improved OPTICS clustering algorithm, parameters of the improved OPTICS clustering algorithm are ε=0.15 and MinPts=20; when the industrial control type service is detected, the time delay factor in the ST-QI is increased to 0.6, and the bandwidth factor is decreased to 0.

05. ​​ In the nine-dimensional feature perception network, the terminal side collects RSRP, SINR and time delay jitter, integrates a 9-axis IMU and dual-frequency GPS to collect motion data, parses the data packet header to obtain QCI level and data importance identifier, and collects terminal power consumption. At the edge, PRB utilization is monitored with a granularity of 1% and a cycle of 1ms. CPU / GPU utilization, memory usage and storage IOPS are collected with a cycle of 10ms, and MEC computing power margin is monitored. The core network side collects network topology information through the SDN controller, monitors the SRv6 tunnel status, and calculates the backbone network load index based on at least 12 indicators from bandwidth utilization, latency and jitter using the entropy weight method.

3. The 5G multi-path redundancy transmission method based on spatiotemporal co-operation and digital twin as described in claim 1, characterized in that, In S2, the ST-QI is calculated according to the following formula: ; ; in, These are the weighting coefficients. For mass components; This is an environmental noise correction term learned based on historical error feedback. The weight coefficients are updated every 50ms using the Double DQN reinforcement learning algorithm. The state space includes service type, network load level, mobility status, and time characteristics; the action space includes... The adjustment step size is 0.05, and the reward function is determined based on the latency score, reliability score, and resource utilization score.

4. The 5G multi-path redundancy transmission method based on spatiotemporal co-operation and digital twin as described in claim 1, characterized in that, In S3, the terminal lightweight twin adopts a lightweight ST-GCN with no more than 500kB of parameters and no more than 0.5TOPS of computation. It outputs a trajectory prediction for the next 50ms and uses dynamic voltage and frequency adjustment technology to control power consumption. The spatiotemporal Transformer of the twin in the edge adopts a 12-layer spatiotemporal Transformer with an embedding dimension of 256 and 8 attention heads. The input includes historical ST-QI sequence, network topology adjacency matrix and service load vector, and the output is the predicted ST-QI value within the next 100ms. The conditional diffusion model employs a 1000-step denoising process, with noise scheduling using the cosine rule. The inputs include short-term prediction results, environmental context features, and historical error patterns. The core heavy twin is based on the NS-3 extended simulation platform, which supports parallel simulation of more than 10,000 nodes. It uses Monte Carlo tree search for redundancy strategy evaluation and performs cross-domain collaboration based on federated learning.

5. The 5G multi-path redundancy transmission method based on spatiotemporal co-operation and digital twin as described in claim 1, characterized in that, In S4, the dynamic redundancy K is calculated according to the following formula: ; in, This is the predicted value for ST-QI. The rate of change of ST-QI For business priority, Where M is the business value density, H is the mobility factor, and H is the historical reliability factor. ; The business value density is determined based on the ratio of data importance to transmission cost and normalized to [0,1].

6. The 5G multipath redundancy transmission method based on spatiotemporal co-location and digital twin as described in claim 1, characterized in that, In S5, the data allocation weights for each path are calculated according to the following formula: ; in, This is the sharpening factor related to the business type; the default value is 2.

0. The number of paths is adjusted between 1 and 8 based on the dynamic redundancy K, and a smooth transition algorithm and transition buffer are used to perform path switching, with a switching response time of no more than 10ms.

7. The 5G multi-path redundancy transmission method based on spatiotemporal co-operation and digital twin as described in claim 1, characterized in that, In S6, an adaptive Kalman filter and Attention-LSTM fusion algorithm are used to predict the terminal trajectory. The input includes GPS location, IMU data, high-precision map information and traffic flow data, and the output is the trajectory prediction for the next 1 second. When adaptively adjusting the prediction window, it is adjusted according to the terminal speed: the prediction window is 50ms for low speed, 65ms for medium speed, 80ms for high speed, and 100ms for extreme high speed. The pre-construction delay of the Xn interface is less than 5ms; the P4 flow table is pre-configured in the programmable data plane and supports nanosecond-level activation; the target base station resource reservation adopts a soft reservation mechanism with a reservation duration of 1s. The time-sensitive network scheduler has a scheduling granularity of 20μs and supports 32 priority queues.

8. The 5G multipath redundancy transmission method based on spatiotemporal co-operation and digital twin as described in claim 1, characterized in that, In S7, the post-quantum key hierarchical update adopts an improved CRYSTALS-Kyber-1024 post-quantum encryption algorithm; among them, the highest priority service uses a 10ms key update cycle and performs forward security protection, the medium priority service uses a 50ms key update cycle and performs two-way authentication, and the ordinary service uses a 100ms key update cycle and performs lightweight authentication; the key distribution is based on blockchain and uses threshold signature technology; The differentiated blockchain audit is based on Hyperledger Fabric light nodes and state channels, and uses Reed-Solomon erasure coding for storage. High-density data is uploaded to the chain every 5ms and uses full-replica storage and digital signature verification, while ordinary data is uploaded to the chain every 100ms and uses erasure coding for storage and batch authentication. The intrusion-tolerant transmission employs anomaly detection combining deep packet inspection and deep learning, with anomaly detection time not exceeding 3ms, backup encrypted channel activation time not exceeding 5ms, and self-healing recovery time less than 10ms.

9. A 5G multi-path redundant transmission system based on spatiotemporal co-operation and digital twin, characterized in that, For performing the 5G multipath redundancy transmission method based on spatiotemporal co-location and digital twin as described in any one of claims 1-8; comprising: A nine-dimensional feature-based cross-domain collaborative sensing network is used to execute S1; The spatiotemporal joint mass index optimization module is used to execute S2; A hierarchical digital twin prediction engine, comprising a terminal light twin unit, an edge twin unit, and a core heavy twin unit, is used to execute S3; The business adaptive redundancy decision module is used to execute S4 and S5; The mobility collaborative execution unit includes a trajectory prediction subunit, a time window adjustment subunit, a path pre-construction subunit, and a resource scheduling subunit, which are used to execute S6; The dynamic intrinsic security protection module includes a key management subunit, a blockchain auditing subunit, and an intrusion tolerance subunit, which are used to execute S7.

10. The 5G multi-path redundant transmission system based on spatiotemporal co-location and digital twin as described in claim 9, characterized in that, In the hierarchical digital twin prediction engine, the terminal light twin unit deploys a lightweight ST-GCN, the edge twin unit deploys a spatiotemporal Transformer and conditional diffusion model, and the core heavy twin unit deploys a simulation platform based on NS-3 extension. The business adaptive redundancy decision module includes a business profile construction unit, a redundancy calculation unit, and a multi-path management unit. The business profile construction unit constructs a seven-dimensional business feature profile including latency requirements, bandwidth requirements, jitter tolerance, priority, data importance, mobility mode, and business value density. In the dynamic intrinsic security protection module, the key management subunit performs priority-based quantum key updates, the blockchain audit subunit performs differentiated on-chain and erasure coding storage, and the intrusion tolerance subunit performs anomaly detection, emergency switching, and self-healing recovery.

Citation Information

Patent Citations

  • Internet of Things data transmission optimization system based on intelligent routing

    CN120512723A

  • Dynamic route selection method and system, electronic equipment and medium

    CN121396886A

  • Communication link redundancy scheduling method for low-delay Internet of Vehicles

    CN121486863A