A data stream feature recognition method and system for high-orbit satellite network

Abstract

This invention relates to a data flow feature identification method and system for high-orbit satellite networks, belonging to the field of satellite internet and onboard network traffic engineering technology. The method includes the following steps: Step 1, extracting detailed coefficients of the traffic sequence and updating the macroscopic congestion status flag through threshold comparison; Step 2, triggering dynamic sampling based on the congestion status and extracting the initial segment of the payload as the sample to be evaluated; Step 3, statistically analyzing the frequency of sample bytes and accumulating mapping values ​​based on a pre-set lookup table to obtain an approximate Shannon entropy; Step 4, adaptively updating the dynamic sampling span based on the difference between the current and historical entropy values ​​and the congestion status. This scheme significantly reduces onboard computing power overhead. By introducing a denormalized integer wavelet transform probe and an entropy dimensionality reduction calculation mechanism based on a pre-set lookup table, it avoids intensive floating-point and logarithmic operations at the algorithm's underlying level, greatly reducing the consumption of limited processor resources on high-orbit satellites.

Description

Technical Field

[0001] This invention relates to the fields of satellite internet, space communication, and onboard network traffic engineering technology. Specifically, it relates to a data flow feature identification method and system for high-orbit satellite networks. This invention uses the dynamically calculated approximate Shannon entropy value of the data flow as the core feature value characterizing the redundancy of the data flow. It aims to solve the problem of wasted onboard computing resources and energy consumption bottlenecks caused by indiscriminate traffic feature evaluation in high-orbit satellite environments with long latency and highly limited onboard computing power through an extremely lightweight algorithm architecture. When dealing with micro-burst congestion in satellite-to-ground or inter-satellite links, this invention provides extremely low-latency decision support for the subsequent dynamic switching of onboard data adaptive compression algorithms through micro-burst perception and rapid feature value evaluation mechanisms. This achieves an optimal dynamic balance between onboard computing resources, long-latency network queuing delay, and end-to-end throughput between space and ground. Background Technology

[0002] With the rapid development of satellite internet, high-orbit satellite network traffic exhibits extreme heterogeneity and high burstiness. Limited by the stringent computing power bottlenecks of onboard processors and expensive link bandwidth, network data transmission has evolved into a multi-dimensional resource scheduling problem that balances low latency and high throughput. Against this backdrop, traditional network load compression mechanisms have revealed numerous technical shortcomings that urgently need to be addressed.

[0003] Traditional static online data compression mechanisms struggle to adapt to the highly dynamic changes in continuous, high-load, heterogeneous payload data streams. Using a single static compression algorithm is too rigid and inefficient: a fixed high compression ratio leads to significant computational waste and processing delays when encountering incompressible data; a fixed no-compression or fast compression method wastes link bandwidth when dealing with highly redundant data. Therefore, the system must dynamically extract eigenvalues ​​reflecting the redundancy of the data itself (i.e., Shannon entropy) to allocate or switch the most suitable compression algorithm in real time. However, the standard Shannon entropy formula involves numerous floating-point divisions and floating-point logarithmic operations. In the general-purpose CPUs of onboard computing platforms, frequent calls to this formula will result in extremely severe performance bottlenecks.

[0004] Furthermore, applying an indiscriminate feature evaluation strategy to all time periods and all traffic will introduce processing delays that significantly degrade transaction response times, generating substantial unnecessary overhead during periods of low network load. Simultaneously, micro-bursts in high-speed networks can easily trigger transient congestion. Traditional congestion control mechanisms rely on passive responses based on packet loss rates over long periods, resulting in significant lag; while existing micro-burst identification algorithms suffer from excessively high computational complexity, making them difficult to port to the data plane. Currently, there is a lack of an extremely lightweight macroscopic monitoring probe capable of proactively sensing traffic throughput gradients, preventing data stream compression engines from achieving "on-demand activation."

[0005] For continuous, high-load, heterogeneous data streams, fixed-interval data sampling can lead to significant CPU cycle waste or slow response to data mutations. Existing adaptive sampling methods often employ complex nonlinear prediction models, whose inference latency hinders real-time applications.

[0006] In summary, existing network data compression schemes have significant technological gaps in addressing the dynamic balance between computational overhead, network queuing latency, and data compression rate. How to achieve high-throughput, low-latency network traffic micro-burst detection and rapid entropy calculation through innovative system logic architecture, thereby providing efficient decision-making services for subsequent adaptive stream compression processing, is a major technical challenge that urgently needs to be solved in this field. Summary of the Invention

[0007] To address the severe challenges of limited onboard computing power, long latency of satellite-to-ground links, and the waste of onboard computing resources and increased queuing delays caused by highly heterogeneous and bursty traffic in high-orbit satellite networks, this invention provides a data flow feature identification method and system for high-orbit satellite networks. This invention addresses the dynamic variability of modern data flow characteristics, pointing out that traditional static compression algorithms cannot adapt to the complex onboard network environment and require real-time data flow feature values ​​as the basis for algorithm switching. This method aims to reduce the ineffective evaluation overhead during periods of network congestion, perceive network micro-bursts in real time, and provide extremely low-latency decision support for subsequent adaptive flow processing (such as dynamic matching and real-time switching of intelligent compression algorithms) through dynamic sampling and a fast feature value evaluation mechanism, thereby achieving a dynamic balance between computing resources and network throughput.

[0008] To achieve the above objectives, the technical solution of the present invention is as follows: a data stream feature identification method for high-orbit satellite networks, the method comprising the following steps:

[0009] Step 1: Extract detailed coefficients of the traffic sequence and update the macroscopic congestion status flags by threshold comparison;

[0010] Step 2: Trigger dynamic sampling based on the congestion status and extract the initial segment of the load as the sample to be evaluated;

[0011] Step 3: Count the frequency of sample bytes, accumulate the mapping values ​​based on the preset lookup table, and obtain the approximate Shannon entropy;

[0012] Step 4: Based on the difference between the current and historical entropy values ​​and the congestion status, adaptively update the dynamic sampling span.

[0013] The specific implementation process of step 1 is as follows:

[0014] Step 1-1: Define the discrete sequence of network interface traffic counts extracted within the time window as follows: The count values ​​of adjacent time windows are respectively denoted as and Extracting detail coefficients through single-level decomposition The calculation formula is: ,in This represents an arithmetic right shift operation;

[0015] Step 1-2: Calculate the detail coefficients. With preset micro-burst alarm threshold Compare the values ​​and use the absolute flow count. Each with a preset high load threshold and idle fallback threshold A comparison was conducted, among which When satisfied or When the macroscopic congestion status flag is set to the wake-up state; when the condition is met... and When the macroscopic congestion status flag is set to a dormant state; when the condition is met... and At that time, the macroscopic congestion status flag is kept in the wake-up state or sleep state of the previous time window;

[0016] This step employs a non-normalized integer wavelet transform and utilizes arithmetic shift operations to extract the changing gradient, avoiding the computational overhead of floating-point scaling factors in conventional algorithms and effectively reducing the computational burden on the onboard processor. Simultaneously, a hysteresis state machine based on amplitude and gradient, along with a dual-threshold determination mechanism, is introduced. By setting a fault-tolerant interval, the high-frequency switching of macroscopic congestion state flags near critical values ​​is reduced, improving the stability of state determination.

[0017] The specific implementation process of step 2 is as follows:

[0018] Step 2-1: Read the status data of the macroscopic congestion status flag; when the macroscopic congestion status flag is in a dormant state, skip the evaluation sampling operation;

[0019] Step 2-2: When the macroscopic congestion status flag is in the awake state, the dynamic sampler is triggered to capture the initial payload of the current continuously high-load heterogeneous payload data packets. One byte is used as the sample to be evaluated, of which The set sampling byte length constant;

[0020] This step avoids indiscriminate sampling of all data streams during periods of network congestion by using a macroscopic congestion status flag as a precondition for triggering dynamic sampling. This mechanism ensures that the system only intervenes for evaluation when it detects micro-bursts or sustained high loads, effectively reducing invalid feature value collection operations and lowering the additional queuing delays caused by data processing.

[0021] The specific implementation process of step 3 is as follows:

[0022] Step 3-1: Count the frequency of each byte value in the sample to be evaluated; define a pre-calculated one-dimensional array loaded into memory as a pre-set lookup table, where the index value in the lookup table is... The element value is ;

[0023] Step 3-2: Obtain the frequency of each byte value Read the index value in the lookup table that equals the frequency. The element values ​​are summed to obtain the total. ;

[0024] Step 3-3, according to the formula The current approximate Shannon entropy is calculated. ;in, The current approximate Shannon entropy, The sum obtained by the accumulation, The set sampling byte length constant, For the first The frequency of each byte value appearing in the sample to be evaluated. This is the index variable for the lookup table.

[0025] This step transforms the standard Shannon entropy calculation into memory read and scalar accumulation operations based on a pre-defined lookup table through mathematical elimination. This effectively avoids intensive floating-point division and logarithmic operations in the data plane, significantly reducing the time complexity of feature value calculation and meeting the real-time requirements for line-speed feature evaluation of data streams in high-speed network environments.

[0026] The specific implementation process of step 4 is as follows:

[0027] Step 4-1: Extract the historical sampling entropy value obtained from the previous evaluation calculation. and the previous dynamic sampling span Calculate the current approximate Shannon entropy. With the historical sampling entropy value absolute value of the difference ;

[0028] Step 4-2: Based on the absolute value of the difference and the macroscopic congestion status flag is updated to the dynamic sampling span. When satisfied And when the macroscopic congestion status flag is in a dormant state, according to the formula Execute an additive strategy to update the dynamic sampling span; where, For the set feature stability threshold, The maximum sampling distance threshold, The minimum sampling distance threshold, This is the set accumulation step size constant.

[0029] Step 4-3, when the conditions are met Or, when the macroscopic congestion status flag is in the wake-up state, according to the formula... Execute a multiplicative-subtractive strategy to update the dynamic sampling span; This indicates an arithmetic right shift operation by two bits.

[0030] This step employs a lightweight additive increment and multiplicative subtraction strategy to dynamically adjust the sampling span. When data feature values ​​are stable, the sampling distance is gradually widened through additive increment to save detection overhead; when feature values ​​change abruptly or network alarms occur, the sampling span is quickly reduced using low-level arithmetic shift instructions. This strategy effectively balances the consumption of onboard computing resources with the sensitivity to detecting abrupt changes in the feature values ​​of continuously high-load heterogeneous payload data streams.

[0031] A data stream feature recognition system for high-orbit satellite networks, the system comprising the following steps:

[0032] The micro-burst sensing module, corresponding to step 1, performs detail coefficient extraction and threshold discrimination on the network interface traffic sequence, and outputs macro-congestion status flag bits;

[0033] The dynamic sampling module, corresponding to step 2, is communicatively connected to the micro-burst sensing module and is used to perform interception actions on data packets of continuous high load heterogeneous load based on the macro-congestion status flag bit to generate samples to be evaluated.

[0034] The eigenvalue evaluation module, corresponding to step 3, is communicatively connected to the dynamic sampling module. It is used to acquire the sample to be evaluated and perform dimensionality reduction calculation based on the preset lookup table, and output an approximate Shannon entropy.

[0035] The span update module, corresponding to step 4, is communicatively connected to the micro-burst sensing module, the dynamic sampling module, and the feature value evaluation module, respectively. It is used to update the dynamic sampling span based on the degree of change of the approximate Shannon entropy and the macroscopic congestion status flag, and to feed back the dynamic sampling span to the dynamic sampling module.

[0036] The various functional modules within the system logically form a closed-loop control relationship of perception, interception, evaluation, and feedback. This structured design reduces the coupling between modules, facilitates efficient software logic mapping and pipelined deployment on the data plane of the onboard network equipment, and improves the overall processing efficiency of data flow feature recognition.

[0037] Beneficial Effects: Compared with existing technologies, the data stream feature recognition method and system for high-orbit satellite networks provided by this invention have the following beneficial effects:

[0038] (1) Significantly reduce onboard computing power overhead: By introducing a non-normalized integer wavelet transform probe and an entropy value dimensionality reduction calculation mechanism based on a pre-set lookup table, this invention avoids dense floating-point operations and logarithmic operations at the algorithm level, which greatly reduces the occupation of limited processor resources of high-orbit satellites.

[0039] (2) Effective control of network queuing delay: The present invention constructs a feature evaluation mechanism based on network congestion state triggering on demand, which remains silent during the low load smooth period of the network and only activates the sampling action when transient congestion or high load is detected; combined with fast approximate Shannon entropy calculation, it avoids packet queuing backlog and forwarding delay caused by indiscriminate and high complexity feature evaluation.

[0040] (3) Enhance the dynamic adaptability to continuous high-load heterogeneous load data streams: Introduce the AIMD adaptive distance decision strategy, which is similar to TCP congestion control. It can dynamically and with low overhead update the sampling span based on the changes in the current data stream characteristics and the macro-congestion status. While reducing the long-term probing load, it ensures the system's timely perception of sudden changes in heterogeneous data stream characteristics and network micro-bursts. Attached Figure Description

[0041] Figure 1 This is a flowchart of the data stream feature recognition method for high-orbit satellite networks according to the present invention;

[0042] Figure 2 This is a comparison chart of simulation experiments on the relationship between queuing delay and data packet arrival rate in this invention;

[0043] Figure 3 This is a comparison chart of TCP throughput simulation experiments under micro-burst congestion of high-orbit satellites according to the present invention;

[0044] Figure 4 This is a schematic diagram of the module architecture of the data stream feature recognition system for high-orbit satellite networks according to the present invention. Detailed Implementation

[0045] Example 1: A Data Stream Feature Recognition Method for High-Orbit Satellite Networks

[0046] This embodiment provides a data flow feature identification method for high-orbit satellite networks. The method operates in the data plane of the onboard switching routing payload, the space-ground integrated edge computing gateway, or the satellite-ground access node of the high-orbit communication satellite.

[0047] Before executing the specific method flow, during the system mounting and initialization phase, a pre-built lookup table (LUT) for fast dimensionality reduction operations and initial state variables are generated and loaded in memory. The pseudocode description of its core logic is as follows:

[0048] Algorithm 1.1: System Initialization State

[0049] function Initialize_System()

[0050] LUT_ARRAY = Precompute_Log_LUT(); / / Precompute logarithmic elimination lookup table

[0051] S_distance = S_min; / / Initialize the dynamic sampling distance to a safe baseline

[0052] Flag_congested = false; / / Initialize the macroscopic congestion flag.

[0053] After system initialization is complete, when network packets arrive at the network interface, the system does not immediately perform indiscriminate feature extraction on all packets, but instead executes... Figure 1 The complete method steps are shown below:

[0054] Step 1: Extract detailed coefficients of the traffic sequence and update the macroscopic congestion status flags through threshold comparison; the specific implementation process is as follows:

[0055] Step 1-1: Define the discrete sequence of network interface traffic counts extracted within the time window as follows: The count values ​​of adjacent time windows are respectively denoted as and Extracting detail coefficients through single-level decomposition The calculation formula is: ,in This represents an arithmetic right shift operation;

[0056] This embodiment uses a denormalized integer wavelet transform as the underlying probe for flow sensing. Specifically, the system removes the orthogonal normalization coefficients from the standard Haar wavelet basis (i.e., The feature extraction logic is simplified to arithmetic shift operations based on fixed-point integers. In this embodiment, the system extracts the raw traffic value from the hardware counter with a preset discretization period of 100 microseconds, which is set based on the line-rate throughput characteristics of the onboard relay environment. This processing method retains the ability to capture traffic gradients while avoiding the occupation of onboard computing resources by floating-point operations.

[0057] Step 1-2: Calculate the detail coefficients. With preset micro-burst alarm threshold Compare the values ​​and use the absolute flow count. Each with a preset high load threshold and idle fallback threshold A comparison was conducted, among which When satisfied or When the macroscopic congestion status flag is set to the wake-up state; when the condition is met... and When the macroscopic congestion status flag is set to a dormant state; when the condition is met... and At that time, the macroscopic congestion status flag is kept in the wake-up state or sleep state of the previous time window;

[0058] This embodiment employs a hysteresis determination logic based on dual thresholds to combat fluctuations in satellite link traffic near critical loads. By setting... and The tolerance range between these ranges allows the system to withstand small fluctuations in traffic without significant gradient changes (i.e., The system maintains its original state during periods of high congestion. This mechanism effectively mitigates the oscillation effect caused by frequent state switching and ensures the stability of macroscopic congestion determination results.

[0059] The pseudocode description of the specific processing logic is as follows:

[0060] Algorithm 1.2: Packet Processing Logic Based on Joint Sensing of Micro-Burst and High Load

[0061] function Process_Packet_Motivation (packet, x_2i, d_1i)

[0062] if (d_1i >= lambda_burst OR x_2i >= lambda_heavy) {

[0063] Flag_congested = true; / / Detects a surge or sustained high traffic volume and triggers the system.

[0064] } else if (x_2i < lambda_idle) {

[0065] Flag_congested = false; / / System shuts down when flow rate falls below the safety threshold.

[0066] }

[0067] / / Given the fluctuations between the two, maintain the Flag_congested state from the previous cycle.

[0068] if (Flag_congested == true) {

[0069] Trigger_Dynamic_Sampling(packet); / / With the system enabled, perform entropy calculation.

[0070] } else {

[0071] Bypass_Evaluation(packet); / / The system is in a shutdown state, so the entropy calculation is skipped.

[0072] }

[0073] Step 2: Trigger dynamic sampling based on the congestion status, and extract the initial segment of the load as the sample to be evaluated; the specific implementation process is as follows:

[0074] Step 2-1: Read the status data of the macroscopic congestion status flag; when the macroscopic congestion status flag is in a dormant state, skip the evaluation sampling operation;

[0075] Step 2-2: When the macroscopic congestion status flag is in the awake state, the dynamic sampler is triggered to capture the initial payload of the current continuously high-load heterogeneous payload data packets. One byte is used as the sample to be evaluated, of which The set sampling byte length constant;

[0076] In this embodiment, a constant sampling byte length is set. The value is 1024. Only when the flag is in the wake-up state, the system extracts the first 1KB field of the current active data stream load as the sample to be evaluated. By selecting the initial segment of the load, the system can capture the significant differences in statistical characteristics of continuously high-load heterogeneous load data streams, while effectively reducing the onboard computing load caused by indiscriminate sampling in a full-flow environment.

[0077] Step 3: Count the frequency of sample bytes, accumulate the mapping values ​​based on a pre-set lookup table, and obtain the approximate Shannon entropy; the specific implementation process is as follows:

[0078] Step 3-1: Count the frequency of each byte value in the sample to be evaluated; define a pre-calculated one-dimensional array loaded into memory as a pre-set lookup table, where the index value in the lookup table is... The element value is ;

[0079] The core of this step lies in transforming complex floating-point logarithmic operations into memory addressing operations through mathematical elimination. The formula for calculating standard Shannon entropy is: ,in For Shannon entropy, For the first The probability of the occurrence of a certain symbol. Since this embodiment uses a fixed load length constant... If the array is 1024 bytes, then the probability of a certain byte value appearing is... ,in This represents the frequency of the byte value in the sample. Substituting the probability formula into the standard formula and simplifying by extracting the common denominator, we can obtain the formula for calculating the equivalent entropy value: To eliminate the summation term The system pre-calculates the logarithmic operations that consume clock cycles. arrive The corresponding integers within the range The mapped values ​​are then fixed into a one-dimensional pre-defined lookup table of length 256 (corresponding to a single byte space of 0-255) and resides in the onboard memory. This is the index variable for the preset lookup table.

[0080] Step 3-2: Obtain the frequency of each byte value Read the index value in the lookup table that equals the frequency. The element values ​​are summed to obtain the total. ;

[0081] When processing real-time data streams, the system will count the frequency of each byte. The preset lookup table is accessed directly as a memory address index. The sum is obtained through 256 loop reads and scalar accumulation operations. (Right now This step reduces the complexity of nonlinear operations to a dimensionality-based addressing and fixed-point addition, making it suitable for operating environments with limited onboard computing power.

[0082] Step 3-3, according to the formula The current approximate Shannon entropy is calculated. ;in, The current approximate Shannon entropy, The sum obtained by the accumulation, The set sampling byte length constant, For the first The frequency of each byte value appearing in the sample to be evaluated. This is the index variable for the lookup table.

[0083] The sum obtained from the above accumulation Substitute this into the aforementioned formula for calculating the equivalent entropy value. In this embodiment, a sampling constant is set. Therefore, the constant term Therefore, the system ultimately relies on the formula. The current approximate Shannon entropy is calculated and output as an eigenvalue. In the actual execution by the onboard hardware, it is divided by... The division operation (i.e., 1024) is further mapped to an arithmetic right shift by ten bits instruction. This enables low-latency feature extraction.

[0084] The pseudocode description of the specific processing logic is as follows:

[0085] Algorithm 1.3: Approximate Shannon Entropy LUT Dimensionality Reduction Acceleration Algorithm

[0086] function Calculate_Entropy (sample_1KB)

[0087] Map_freq = Count_Bytes(sample_1KB); / / Count the frequency of each byte Ci

[0088] Sum_LUT = 0;

[0089] for (each x in Map_freq) {

[0090] Sum_LUT = Sum_LUT + LUT_ARRAY[Map_freq[x]]; / / Look up the table and accumulate the mapped values

[0091] }

[0092] H_new = 10 - (Sum_LUT >> 10); / / Right shift by 10 bits instead of dividing by 1024 for calculation

[0093] return H_new;

[0094] Step 4: Adaptively update the dynamic sampling span based on the difference between the current and historical entropy values ​​and the congestion status. The specific implementation process is as follows:

[0095] Step 4-1: Extract the historical sampling entropy value obtained from the previous evaluation calculation. and the previous dynamic sampling span Calculate the current approximate Shannon entropy. With the historical sampling entropy value absolute value of the difference ;

[0096] Step 4-2: Based on the absolute value of the difference and the macroscopic congestion status flag is updated to the dynamic sampling span. When satisfied And when the macroscopic congestion status flag is in a dormant state, according to the formula Execute an additive strategy to update the dynamic sampling span; where, For the set feature stability threshold, The maximum sampling distance threshold, The minimum sampling distance threshold, This is the set accumulation step size constant.

[0097] When the data stream characteristics are stable and the link is in a smooth period, the system linearly widens the sampling distance through scalar addition, thereby effectively reducing the long-term detection load of the onboard processor.

[0098] Step 4-3, when the conditions are met Or, when the macroscopic congestion status flag is in the wake-up state, according to the formula... Execute a multiplicative-subtractive strategy to update the dynamic sampling span; This indicates an arithmetic right shift operation by two bits.

[0099] Once a sudden change in data stream characteristic values ​​is detected or a macroscopic congestion alarm is received, the system utilizes arithmetic shift instructions from the underlying hardware. (Equivalent to dividing the span value by 4) This rapidly reduces the sampling span. The above additive increase and multiplicative decrease (AIMD) mechanism avoids complex nonlinear predictions, reducing computational overhead while ensuring the system's sensitivity to capturing burst traffic characteristics.

[0100] The specific adaptive strategies are shown in Table 1:

[0101] Table 1:

[0102]

[0103] The core pseudocode for the execution logic is as follows:

[0104] Algorithm 1.4: Dynamic Sampling Span Algorithm

[0105] function Update_Sampling_Distance (H_new, H_old, Flag_congested)

[0106] delta_H = abs(H_new - H_old);

[0107] if (delta_H < T_stable && Flag_congested == false) {

[0108] S_distance = min(S_max, S_distance + delta_S); / / Additive increment, widens the sampling width

[0109] }

[0110] else {

[0111] S_distance = max(S_min, S_distance >> 2); / / Multiplicative subtraction, displacement command quickly wakes up sampling

[0112] }

[0113] H_old = H_new; / / Update the historical entropy value.

[0114] based on Figure 2 Simulation experiments verifying the relationship between queuing delay and packet arrival rate: Simulation results show that the data flow feature value calculation based on the pre-set lookup table in this embodiment can still maintain line-rate packet processing even when facing a sustained high-load heterogeneous payload packet arrival rate (20 Kpps), with extremely low additional queuing delay introduced by the system. This mechanism effectively eliminates the excessive occupation of onboard CPU computing power by standard floating-point arithmetic instructions, avoids buffer queue backlog and passive congestion packet loss caused by the time-consuming feature evaluation calculation, and provides highly sensitive and timely underlying decision support for micro-burst congestion intervention in the data plane.

[0115] based on Figure 3Simulation experiments verifying TCP throughput performance under micro-burst congestion on high-orbit satellites: Macroscopic transport layer simulation results further demonstrate that, when facing transient micro-burst congestion, this embodiment can achieve early detection before the physical buffer queue overflows and promptly output data stream characteristic values ​​to the subsequent adaptive intervention module. This low-latency feedback mechanism effectively blocks the performance degradation path caused by physical packet loss due to assessment delays, which in turn leads to a sudden drop in the transport layer TCP congestion window. Simulation experiments confirm that this embodiment can maintain a stable high throughput (7.1 Mbps) during micro-burst congestion and recover rapidly after the congestion is resolved, achieving an optimal dynamic balance between limited onboard computing resources, long-latency network queuing delays, and end-to-end transmission bandwidth between space and ground.

[0116] Example 2: A data stream feature recognition system for high-orbit satellite networks

[0117] This embodiment provides a data stream feature recognition system for high-orbit satellite networks, such as Figure 4 As shown, the system specifically includes the following functional modules:

[0118] The micro-burst sensing module, corresponding to step 1, performs detailed coefficient extraction and threshold discrimination on the network interface traffic sequence and outputs a macro-level congestion status flag. This module utilizes denormalized integer wavelet transform to sense the changing gradient of link throughput in real time. Through its output macro-level congestion status flag, the system can determine the start and stop of subsequent dynamic sampling and feature value evaluation modules based on the real-time network status, thereby avoiding unnecessary computational overhead at the macro level.

[0119] The dynamic sampling module, corresponding to step 2, is communicatively connected to the micro-burst sensing module. It is used to intercept data packets from continuously high-load heterogeneous payloads based on the macro-congestion status flag, generating samples to be evaluated. The core function of this module is to perform on-demand interception under the control of the macro-congestion status flag. When the macro-congestion status flag is in the active state, the module intercepts data from the initial segment of the payload to form samples, ensuring that feature acquisition only occurs during peak traffic periods that truly require evaluation.

[0120] The eigenvalue evaluation module, corresponding to step 3, is communicatively connected to the dynamic sampling module. It acquires the samples to be evaluated and performs dimensionality reduction calculations based on the preset lookup table, outputting an approximate Shannon entropy. This module transforms the originally high-overhead logarithmic operation into a low-latency lookup and accumulation process by accessing the preset lookup table resident in memory. The final output approximate Shannon entropy can reflect the redundancy characteristics of the current data stream in real time, providing a decision indicator for subsequent traffic processing.

[0121] The span update module, corresponding to step 4, is communicatively connected to the micro-burst sensing module, the dynamic sampling module, and the feature value evaluation module. It updates the dynamic sampling span based on the change in the approximate Shannon entropy and the macroscopic congestion status flag, and feeds the dynamic sampling span back to the dynamic sampling module. This module dynamically extends or shortens the sampling span using additive increment and multiplicative subtraction strategies, dynamically adjusting the sampling frequency to achieve a balance between detection accuracy and system resource consumption.

[0122] It should be noted that the above embodiments are not intended to limit the scope of protection of the present invention. Equivalent transformations or substitutions made based on the above technical solutions all fall within the scope of protection of the claims of the present invention.

Claims

1. A data stream feature identification method for high-orbit satellite networks, characterized in that, The method includes the following steps: Step 1: Extract detailed coefficients of the traffic sequence and update the macroscopic congestion status flags by threshold comparison; Step 2: Trigger dynamic sampling based on the congestion status and extract the initial segment of the load as the sample to be evaluated; Step 3: Count the frequency of sample bytes, accumulate the mapping values ​​based on the preset lookup table, and obtain the approximate Shannon entropy; Step 4: Based on the difference between the current and historical entropy values ​​and the congestion status, adaptively update the dynamic sampling span.

2. The data stream feature identification method for high-orbit satellite networks according to claim 1, characterized in that, The specific implementation process of step 1 is as follows: Step 1-1: Define the discrete sequence of network interface traffic counts extracted within the time window as follows: The flow count values ​​for adjacent time windows are respectively denoted as and Extracting detail coefficients through single-level decomposition The calculation formula is: ,in This represents an arithmetic right shift operation; Step 1-2: Adjust the detail coefficients With preset micro-burst alarm threshold Compare and record the flow count values. Each with a preset high load threshold and idle fallback threshold A comparison was conducted, among which When satisfied or When the macroscopic congestion status flag is set to the wake-up state; when the condition is met... and When the macroscopic congestion status flag is set to dormant, the condition is met. and At that time, the macro congestion status flag is kept in the wake-up or dormant state of the previous time window.

3. The data stream feature identification method for high-orbit satellite networks according to claim 2, characterized in that, The specific implementation process of step 2 is as follows: Step 2-1: Read the status data of the macroscopic congestion status flag. When the macroscopic congestion status flag is in a dormant state, skip the evaluation sampling operation. Step 2-2: When the macroscopic congestion status flag is in the awakened state, the dynamic sampler is triggered to capture the beginning of the effective payload of the current continuously high-load heterogeneous payload data packets. One byte is used as the sample to be evaluated, of which This is the set sampling byte length constant.

4. The data stream feature identification method for high-orbit satellite networks according to claim 3, characterized in that, The specific implementation process of step 3 is as follows: Step 3-1: Count the frequency of each byte value in the sample to be evaluated. Define a pre-calculated one-dimensional array loaded into memory as a pre-set lookup table, where the index value is... The element value is ; Step 3-2: Obtain the frequency of each byte value in the sample to be evaluated. Read the index value in the lookup table equal to the frequency. The element values ​​are summed to obtain the total. ; Step 3-3, according to the formula The current approximate Shannon entropy is calculated. ;in, For the current approximate Shannon entropy, The sum obtained by accumulation The set sampling byte length constant, This is the index variable for the lookup table.

5. The data stream feature identification method for high-orbit satellite networks according to claim 4, characterized in that, The specific implementation process of step 4 is as follows: Step 4-1: Extract the historical sampling entropy value obtained from the previous evaluation calculation. and the previous dynamic sampling span Calculate the current approximate Shannon entropy. Compared with historical sampling entropy values absolute value of the difference ; Step 4-2: Based on the absolute value of the difference and the macroscopic congestion status flag is updated dynamically across the sampling span. When satisfied And when the macroscopic congestion status flag is in a dormant state, according to the formula Execute an additive strategy to update the dynamic sampling span; where, For the set feature stability threshold, The maximum sampling distance threshold, The minimum sampling distance threshold, The set accumulation step size constant; Step 4-3, when the conditions are met Or, when the macroscopic congestion status flag is in the awake state, according to the formula... Execute a multiplicative-subtractive strategy to update the dynamic sampling span; This indicates an arithmetic right shift operation by two bits.

6. The data stream feature identification method for high-orbit satellite networks according to claim 5, characterized in that, Before performing step 1, system initialization is also included: a preset lookup table is generated and loaded in memory in advance, the dynamic sampling span is initialized to the minimum sampling distance threshold, and the macroscopic congestion status flag is initialized to a dormant state.

7. A data stream feature recognition system for high-orbit satellite networks, characterized in that, This method is used to implement the data stream feature recognition method for high-orbit satellite networks as described in any one of claims 1-6. The system includes: The micro-burst sensing module performs detail coefficient extraction and threshold discrimination on the network interface traffic sequence, and outputs macro-congestion status flags; The dynamic sampling module, which communicates with the micro-burst sensing module, is used to intercept data packets of continuous high-load heterogeneous load based on the macro-congestion status flag and generate samples to be evaluated. The eigenvalue evaluation module communicates with the dynamic sampling module to obtain the sample to be evaluated and perform dimensionality reduction calculation based on a preset lookup table, outputting an approximate Shannon entropy; The span update module is communicatively connected to the micro-burst sensing module, the dynamic sampling module, and the feature value evaluation module, respectively. It is used to update the dynamic sampling span based on the degree of change of the approximate Shannon entropy and the macroscopic congestion status flag, and to feed back the dynamic sampling span to the dynamic sampling module.

8. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the program, it implements a data stream feature recognition method for high-orbit satellite networks as described in any one of claims 1 to 6.

9. A computer-readable storage medium storing computer instructions thereon, characterized in that, When the computer instruction is executed by the processor, it implements a data stream feature recognition method for high-orbit satellite networks as described in any one of claims 1-6.

Images