Packet link fault locating method and system based on error code time sequence fingerprint

Abstract

The application discloses a packet link fault positioning method and system based on error code time sequence fingerprint, and relates to the network operation and maintenance field.The method comprises the following steps: acquiring error code count and message digest of an observation port, determining an error code event window, a directed forwarding path and a message replication relationship; constructing message progress phase and interval range by using common anchor messages before and after the window and message sequence positions, converting error code increment into measured direct error code quantity distribution, obtaining measured message loss quantity distribution from upstream and downstream digest differences, and forming measured error code time sequence fingerprint; constructing an observation template for a candidate fault section based on the forwarding path and the replication relationship; adopting a monotonic elastic mapping that maintains message order, inverting damaged message quantity of the candidate section under the constraint of unique distribution of measured quantity and generating a predicted time sequence fingerprint, calculating positioning residuals according to phase position, propagation direction, propagation range consistency and quantity conservation relationship, and determining a target fault section set.

Description

Technical Field

[0001] This application relates to the field of intelligent network operation and maintenance management technology, and in particular to a method and system for locating packet link faults based on bit error timing fingerprints. Background Technology

[0002] As packet-switched networks continue to expand, the number of devices, links, and service forwarding relationships within the network increases, making the network operating environment more complex. During network operation, some links or network devices may experience intermittent anomalies that are short-lived and infrequent. These anomalies typically do not cause complete link outages, and the link and routing states may remain normal. However, they can lead to occasional packet loss, transmission latency fluctuations, decreased throughput, or service request timeouts, making them difficult to detect and locate promptly using conventional link failure alarms.

[0003] Current network operation and maintenance systems typically collect link status, port quality counts, interface drop counts, and service performance indicators. These are then combined with alarm information, network topology, statistical analysis, or active probing results to analyze and assess network operation. The aforementioned monitoring data are mostly cumulative or statistical values ​​generated according to preset periods, primarily reflecting the overall operation of ports, links, or services within a certain timeframe. When the duration of an anomaly is short or the number of affected service packets is small, the corresponding anomaly characteristics are easily influenced by normal data within the statistical period. Therefore, relying solely on a single monitoring indicator or a simple combination of different indicators is usually insufficient to accurately determine the scope of the anomaly.

[0004] Furthermore, different network devices may have different data collection cycles, collection times, and reporting latency. Service packets are also affected by factors such as network load, forwarding paths, and traffic changes during multi-hop forwarding. Therefore, abnormal data collected from different devices or monitoring locations may deviate in terms of occurrence time and manifestation. In cases of multi-path forwarding, dynamic changes in network status, or multiple anomalies occurring simultaneously, the correlation between anomaly information becomes more complex, potentially leading to a larger candidate range or less stable localization results. Summary of the Invention

[0005] This application provides a method, system, storage medium, computer program product, and electronic device for locating packet link faults based on bit error timing fingerprints, which at least solves the problem of difficulty in accurately locating intermittent link faults in complex network environments in the prior art.

[0006] In a first aspect, embodiments of this application provide a method for packet link fault location based on bit error timing fingerprints. The method includes: acquiring the cumulative direct bit error count and normal message digest of multiple observation ports; determining the bit error increment based on the cumulative direct bit error count; determining the bit error event window based on the bit error increment and acquiring the directed forwarding path and message replication relationship; determining the preceding anchor message and following anchor message jointly observed by multiple observation ports on the directed forwarding path from the normal message digests before and after the bit error event window; constructing a message progress phase and the phase range of each sampling interval based on the message sequence position of each observation port between the preceding anchor message and the following anchor message; converting the bit error increment into a measured direct bit error distribution based on each phase range; and comparing the normal message digests of observation ports with upstream and downstream relationships in the directed forwarding path to obtain the measured direct bit error rate that exists upstream but is missing downstream. The message missing quantity distribution is used to generate a measured bit error time sequence fingerprint. Based on the directed forwarding path and the message replication relationship, observation templates are constructed for multiple candidate fault sections. Each observation template maps the unit damaged message quantity of the corresponding candidate fault section to the expected direct bit error quantity distribution and expected message missing quantity distribution of each observation port. A monotonically elastic mapping that preserves message order is performed on the message progress phase. Under the constraint that each measured direct bit error quantity or measured message missing quantity is assigned to at most one candidate fault section, the damaged message quantity of each candidate fault section is jointly inverted based on the measured bit error time sequence fingerprint and each observation template, and predicted time sequence fingerprints are generated respectively. The location residual of each candidate fault section is determined based on the consistency and quantity conservation relationship between the phase position, propagation direction, and propagation range of each predicted time sequence fingerprint and the measured bit error time sequence fingerprint, and the target fault section set is determined based on each location residual.

[0007] Secondly, embodiments of this application provide a packet link fault location system based on bit error timing fingerprints. The system includes: a bit error event acquisition unit, used to acquire the cumulative direct bit error count and normal message digest of multiple observation ports, determine the bit error increment based on the cumulative direct bit error count, determine the bit error event window based on the bit error increment, and acquire the directed forwarding path and message replication relationship; a phase reference construction unit, used to determine the preceding anchor message and following anchor message jointly observed by multiple observation ports on the directed forwarding path from the normal message digests before and after the bit error event window, respectively, and construct the message progress phase and the phase range of each sampling interval based on the message sequence position of each observation port between the preceding anchor message and the following anchor message; and a measured fingerprint generation unit, used to convert the bit error increment into a measured direct bit error distribution based on each phase range, and compare the normal message digests of observation ports with upstream and downstream relationships in the directed forwarding path to obtain the measured direct bit error distribution that exists upstream but is missing downstream. The system includes a message missing quantity distribution unit to generate a measured bit error rate timing fingerprint; an observation template construction unit to construct observation templates for multiple candidate fault sections based on the directed forwarding path and the message replication relationship, whereby each observation template maps the unit damaged message quantity of the corresponding candidate fault section to the expected direct bit error rate distribution and expected message missing quantity distribution of each observation port; a damaged message inversion unit to perform a monotonically elastic mapping on the message progress phase to maintain message order, under the constraint that each measured direct bit error rate or measured message missing quantity is assigned to at most one candidate fault section, jointly inverting the damaged message quantity of each candidate fault section based on the measured bit error rate timing fingerprint and each observation template, and generating predicted timing fingerprints respectively; and a fault section localization unit to determine the localization residual of each candidate fault section based on the consistency and quantity conservation relationship between the predicted timing fingerprint and the measured bit error rate timing fingerprint in terms of phase position, propagation direction, and propagation range, and to determine the target fault section set based on the localization residual.

[0008] Thirdly, an electronic device is provided, comprising: at least one processor, and a memory communicatively connected to the at least one processor, wherein the memory stores instructions executable by the at least one processor, the instructions being executed by the at least one processor to enable the at least one processor to perform the steps of the packet link fault location method based on bit error timing fingerprint of any embodiment of the present application.

[0009] Fourthly, embodiments of this application provide a storage medium storing a computer program thereon, which, when executed by a processor, implements the steps of the packet link fault location method based on bit error timing fingerprints according to any embodiment of this application.

[0010] Fifthly, embodiments of this application provide a computer program product, including a computer program / instructions, which, when executed by a processor, implement the steps of the packet link fault location method based on bit error timing fingerprints according to any embodiment of this application.

[0011] The packet link fault location method and system based on bit error timing fingerprinting provided in this application can achieve at least the following technical effects:

[0012] By converting the cumulative direct bit error count into bit error increments and limiting these increments to the corresponding bit error event windows, observational data directly related to the anomaly process can be extracted. Simultaneously, by establishing a message progress phase using jointly observed preceding and following anchor messages, the bit error increments and message missing information at different observation ports can be uniformly expressed according to message sequence. The resulting measured bit error timing fingerprint takes into account both direct bit error rates and upstream / downstream message missing rates, characterizing the link damage process from both the anomaly occurrence and propagation along the path perspectives.

[0013] By establishing observation templates for each candidate fault section based on directed forwarding paths and message replication relationships, the direct error distribution and message missing distribution that a unit of damaged message volume should form at different observation ports can be predefined, thereby establishing a correspondence between candidate fault locations and observable anomalies. By adapting to local progress shifts during message propagation through monotonic elastic mapping, and by jointly inverting the damaged message volume of each candidate fault section under the constraint that each measured anomaly volume belongs to at most one candidate fault section, competitive duplication of the same observation evidence by different candidate locations can be suppressed, and different damage contributions can be distinguished when multiple paths, message replication, or multiple fault sections coexist.

[0014] This technical solution transforms packet link fault location into a problem of temporal fingerprint matching and damaged packet quantity inversion under directed topology constraints. The inversion results are jointly verified by phase position, propagation direction, propagation range and quantity conservation relationship, so that the determination of the target fault section has a complete evidence chain from topology propagation mechanism, abnormal temporal manifestation to damaged quantity relationship, which helps to narrow down the candidate fault range and enhance the consistency, interpretability and reliability of the location results in complex forwarding environment. Attached Figure Description

[0015] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0016] Figure 1 A flowchart illustrating an example of a packet link fault location method based on bit error timing fingerprinting according to an embodiment of this application is shown;

[0017] Figure 2 This document shows an example of an operation flowchart for generating a measured bit error timing fingerprint in a method according to an embodiment of this application;

[0018] Figure 3 A flowchart illustrating an example of determining the target fault profile set in a method according to an embodiment of this application is shown.

[0019] Figure 4 A schematic diagram illustrating the system operation mechanism of an example of a packet link fault location method based on bit error timing fingerprinting according to an embodiment of this application is shown.

[0020] Figure 5 A comparative simulation diagram illustrating the Top-1 fault location accuracy of the method of this application and the timing correlation method under different device clock offsets and normal message digest sampling rates is shown.

[0021] Figure 6 The diagram shows a comparative experimental simulation of the Pareto distribution between the diagnostic computation delay and Top-1 fault location accuracy of different methods, as well as an example of the candidate fault section location residual spectrum of the method presented in this paper.

[0022] Figure 7 A structural block diagram of an example of a packet link fault location system based on bit error timing fingerprinting according to an embodiment of this application is shown. Detailed Implementation

[0023] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, the technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0024] It should be noted that, regarding specific fault location methods, some current research in related technologies employs network layering or probabilistic inference. This involves estimating the operational status of different links based on the end-to-end success rate, packet loss rate, transmission latency, or throughput of multiple service flows, combined with the set of paths traversed by the corresponding service flows. For example, the Flock system uses PGM (Probabilistic Graphical Model) to analyze end-to-end flow observations to infer potentially faulty network components. This type of approach primarily relies on statistical inference based on a certain number of service flow samples, and its judgment results may be affected by the sample size, path coverage, and model parameters. When anomalies only affect a small number of packets, or when the service path sets corresponding to multiple candidate links are similar, different candidate locations may lead to similar statistical interpretations.

[0025] Other studies reconstruct the network devices, forwarding paths, and processing states of packets by recording their hop-by-hop forwarding history. For example, NetSight provides packet-level forwarding information for network diagnostics by collecting packet history at different network nodes. This approach provides relatively direct and detailed diagnostic data, but typically requires the configuration of corresponding packet collection, compression, storage, and retrieval facilities. When the monitoring scope is large or the forwarding history of production service packets needs to be stored for a long time, it also consumes significant network and computing resources. Furthermore, in environments with traditional network equipment, cross-vendor networks, or varying equipment capabilities, it may not be possible to continuously obtain complete packet-by-packet forwarding records.

[0026] To address packet loss in virtualized networks, some studies employ packet tracing systems such as VTrace. These systems configure packet marking, matching, and logging rules within virtual forwarding devices and reconstruct the processing of target packets based on logs generated at different forwarding locations. This approach is suitable for analyzing persistent packet loss in virtual forwarding environments, but its applicability is limited by the rule configuration and log collection capabilities of the virtual forwarding devices, and is also constrained by network architecture and device functionality. For anomalies occurring outside the scope of virtual forwarding processing, the information provided by the corresponding logs may be limited.

[0027] Furthermore, proactive probing systems like NetBouncer employ an IP-in-IP (Internet Protocol in Internet Protocol) approach, routing probe packets through a specified network path and then inferring potentially abnormal devices or links based on the probe results. While proactive probing can generate relatively well-defined measurement results, probe packets may differ from actual service packets in terms of packet length, traffic type, transmission rhythm, and path selection. Intermittent anomalies may not necessarily occur during the probing period. Continuously increasing the probe frequency can increase the likelihood of obtaining anomaly observations, but it consumes additional network transmission and device processing resources.

[0028] For link and port operating status, some related technologies can also collect indicators such as BER (Bit Error Rate), CRC (Cyclic Redundancy Check) errors, FCS (Frame Check Sequence) errors, physical coding anomalies, and interface drops, and determine candidate anomaly locations through correlation analysis, probabilistic inference, or classification models. This approach can comprehensively utilize various device operating information; however, port counts typically reflect cumulative changes between adjacent collection times, making it difficult to present the specific details of each abnormal packet within the corresponding sampling interval. When using the time-series characteristics of indicators as the primary basis for judgment, the results may also be affected by factors such as device type, statistical caliber, monitoring granularity, network load, and sample scenario.

[0029] Under the condition that the receiving device performs full frame verification and does not continue to forward failed verification packets normally, Ethernet packets with cyclic redundancy check errors, framing errors, or abnormal lengths are identified as damaged packets at the receiving side, and their subsequent paths show that the corresponding normal packets are missing. Therefore, device operation indicators, packet forwarding records, and end-to-end service performance obtained at different locations in the network may have different data meanings and observation ranges. Currently, related technologies utilize the above information for fault analysis, but due to limitations such as observation granularity, device capabilities, and applicable environments, there may still be certain limitations when conducting detailed analysis of short-term, low-frequency anomalies.

[0030] It should be understood that the above description of the relevant technologies is intended only to help the public better understand the inventive spirit and motivation of this application, and is not intended to limit this application. Furthermore, the technical solutions described in the above-mentioned relevant technologies are not prior art, and may also be undisclosed technical solutions, such as those under research or in the laboratory stage.

[0031] Figure 1A flowchart illustrating an example of a packet link fault location method based on bit error timing fingerprints according to an embodiment of this application is shown.

[0032] Regarding the execution entity of the method in the embodiments of this application, it can be executed by a network management system, a network controller, a network telemetry analysis platform, or a fault location system deployed in a network operation and maintenance server. The execution entity can implement the method in software, hardware, or a combination of software and hardware by running programs or instructions stored in the memory through a processor. For ease of explanation, the entity executing the method will be referred to as the system below.

[0033] The observation port can be a port in a switching device, routing device, gateway device, or other packet forwarding device that can provide bit error counts or packet observation results, or it can be a network observation point deployed in the forwarding path. Multiple observation ports can be distributed at different forwarding locations on the same packet link to respectively form direct bit error observation and packet forwarding observation. It should be understood that the above-described execution entity and network scenario are only used to illustrate the processing procedure of the embodiments of this application and do not constitute a limitation on the scope of protection of this application.

[0034] like Figure 1 As shown, in step S110, the cumulative direct bit error count and normal message digest of multiple observation ports are obtained, the bit error increment is determined according to the cumulative direct bit error count, the bit error event window is determined according to the bit error increment, and the directed forwarding path and message replication relationship are obtained.

[0035] The cumulative direct bit error count is used to cumulatively record the direct bit error events detected by the corresponding observation port. It can originate from the bit error counter maintained by the physical layer, data link layer, or port receiving and processing module. The system can read the cumulative direct bit error count of each observation port according to a preset acquisition period, and determine the number of newly added bit errors in the corresponding sampling interval based on the changes in the cumulative direct bit error count at adjacent acquisition times, thus obtaining the bit error increment. For changes in counter status or sampling anomalies, the system can verify the corresponding observation results to avoid using count changes caused by non-link bit error reasons as the bit error increment.

[0036] The normal message digest is a message observation record obtained by extracting digest information from messages that were not determined to be direct errors by the corresponding observation port. The normal message digest may include a digest identifier generated based on message header fields, transmission sequence number, payload content, or a combination thereof, and may also include sequence information reflecting the message order, observation port identifier, and acquisition time. Multiple observation ports can use corresponding digest generation rules to enable matching based on corresponding normal message digests when the same message passes through different observation ports.

[0037] The system identifies the time range of concentrated error occurrences based on the change in error increment relative to normal operating conditions, and determines the error event window by combining the normal observation status before and after the error occurrence. For the error event window, the system obtains the actual effective forwarding information within the corresponding time period, determines the directed forwarding path of the packet propagation from the upstream observation port to the downstream observation port, and determines the packet replication relationship where the packet forms one or more downstream replicas during the forwarding process. Therefore, error counts, normal packet digests, and actual forwarding status can be constrained within the same error event range, reducing the mismatch in observation relationships caused by the mismatch between the forwarding status and the error occurrence time.

[0038] In step S120, the preceding anchor message and the following anchor message jointly observed by multiple observation ports on the directed forwarding path are determined from the normal message digests before and after the bit error event window, respectively. Based on the message sequence position of each observation port between the preceding anchor message and the following anchor message, the message progress phase and the phase range of each sampling interval are constructed.

[0039] The preceding anchor message and the following anchor message are used to indicate the message forwarding progress before and after the bit error event, respectively. The system can identify normal messages observed by multiple observation ports along the directed forwarding path from the normal message digests before the bit error event window, and use these as preceding anchor messages; and identify normal messages observed by multiple observation ports along the directed forwarding path from the normal message digests after the bit error event window, and use these as following anchor messages. Joint observation means that multiple observation ports can confirm, based on the digest identifier, that their respective normal message digests correspond to the same message.

[0040] The message sequence number reflects the sequential position of normal message digests within the corresponding observation port. For any observation port, the system defines a message sequence segment using the preceding anchor message and the following anchor message, and determines the message progress phase based on the message sequence position of each normal message digest within the segment relative to the two anchor messages. The message progress phase reflects the relative progress position of the corresponding message between the preceding and following anchor messages and does not directly depend on the local clock values ​​of different observation ports. For any sampling interval, the system determines the phase range based on the message progress phase corresponding to the normal message digest within that sampling interval; when the observation results are insufficient to directly determine the phase range, the corresponding phase range can be determined based on the continuous change relationship of the message progress phase in adjacent sampling intervals. Thus, asynchronous sampling results from different observation ports can be converted to a common message progress reference while maintaining the message sequence.

[0041] In step S130, the bit error increment is converted into a measured direct bit error distribution according to each phase range, and the normal message digests of the observation ports with upstream and downstream relationships in the directed forwarding path are compared to obtain the measured message missing quantity distribution that exists upstream but is missing downstream, so as to generate a measured bit error timing fingerprint.

[0042] The cumulative direct error count corresponds to the sampling interval, while the normal message digest corresponds to the specific message. To enable comparison between the two types of observations under a unified message progress benchmark, the system converts the error increment within each sampling interval to the corresponding message progress phase position based on the phase range corresponding to each sampling interval, thus obtaining the measured direct error distribution. This measured direct error distribution characterizes the distribution of direct errors at different observation ports and different message progress positions.

[0043] For observation ports with upstream and downstream relationships in the directed forwarding path, the system can match normal message digests according to the digest identifier and message sequence. If a normal message digest can be identified at the upstream observation port but not at the downstream observation port that it should reach according to the directed forwarding path, the system determines the corresponding message quantity as the measured message missing quantity and forms the measured message missing quantity distribution based on the message progress phase corresponding to the missing message. When performing upstream and downstream comparisons, the impact of message duplication and normal forwarding processing on the number of message observations and their correspondence can be eliminated, thus avoiding the identification of normal duplication splitting or forwarding processing as abnormal missing messages.

[0044] The system combines the measured direct bit error rate distribution and the measured message missing rate distribution according to the corresponding observation positions and message progress phases to generate a measured bit error rate timing fingerprint. This timing fingerprint can simultaneously characterize the message damage at the direct bit error observation side and the message failure to reach the downstream observation position, enabling the two types of observations to be correlated under a unified message progress benchmark.

[0045] In step S140, observation templates are constructed for multiple candidate fault sections based on the directed forwarding path and the packet replication relationship. Each observation template maps the unit damaged packet quantity of the corresponding candidate fault section to the expected direct bit error rate distribution and the expected packet missing rate distribution of each observation port.

[0046] A candidate fault section is used to represent a potential processing boundary on the directed forwarding path that may cause packet corruption or halt forwarding. A candidate fault section can correspond to a boundary in the physical link reception process, port reception process, or internal processing within a forwarding node. Based on the position of the candidate fault section in the directed forwarding path, the directed reachability relationship between the candidate fault section and each observation port, and the packet replication relationship, the system determines the expected observation results that a single corrupted packet can generate at each observation port starting from that candidate fault section.

[0047] The observation template includes the mapping relationship between the unit amount of damaged packets and the expected distribution of direct bit errors, as well as the mapping relationship between the unit amount of damaged packets and the expected distribution of missing packets. If a candidate fault segment can generate direct bit errors at the corresponding observation port, the corresponding observation template generates the expected direct bit error amount at the corresponding location; if a damaged packet cannot continue to reach the downstream observation port along the directed forwarding path, the corresponding observation template generates the expected missing packet amount within the corresponding propagation range; if a packet is replicated to form multiple downstream replicas, the corresponding observation template describes the impact of the damaged source packet on different replication branches according to the packet replication relationship. Thus, the network location, fault propagation direction, and impact range of different candidate fault segments can be converted into expected observation patterns that can be compared with the measured bit error time sequence fingerprint.

[0048] In step S150, the message progress phase is subjected to monotonically elastic mapping that preserves the message order. Under the constraint that each measured direct bit error or measured message missing amount is assigned to at most one of the candidate fault sections, the amount of damaged messages of each candidate fault section is jointly inverted based on the measured bit error time series fingerprint and each of the observation templates, and predicted time series fingerprints are generated respectively.

[0049] Different observation ports may exhibit variations in sampling period, processing delay, and local message arrival rate, causing local shifts in the phase positions of the same set of damaged messages across different observation ports. The system employs monotonic elastic mapping to adjust the local phase deviations between different observation ports without altering the message sequence. This monotonic elastic mapping maintains the start and end directions and sequence of the message progress phases without reversal, thus avoiding forced matching of measured observations through phase transformations that do not conform to the actual message propagation process.

[0050] The system assigns the measured direct bit error rate and the measured message missing rate from the measured bit error time series fingerprint to corresponding candidate fault sections, ensuring that each measured direct bit error rate or measured message missing rate is interpreted by at most one candidate fault section to avoid multiple candidate fault sections reusing the same observation evidence. Based on the assigned measured values, the observation templates of each candidate fault section, and the phase correspondence adjusted by monotonic elastic mapping, the system jointly determines the amount of damaged messages corresponding to each candidate fault section. Subsequently, the determined amount of damaged messages is input into the corresponding observation template to generate the predicted time series fingerprint for each candidate fault section. This joint inversion process allows different candidate fault sections to compete in interpreting the measured bit error time series fingerprint under unified constraints, while preserving the possibility that the bit error event involves one or more fault sections.

[0051] In step S160, the location residuals of each candidate fault section are determined based on the consistency and quantity conservation relationship between the phase position, propagation direction and propagation range of each predicted timing fingerprint and the measured bit error timing fingerprint, and the target fault section set is determined based on each location residual.

[0052] For any candidate fault segment, the system compares the corresponding predicted timing fingerprint with the measured bit error timing fingerprint assigned to that candidate fault segment. Specifically, phase position consistency characterizes whether the predicted direct bit error or message loss occurs at the message progress position corresponding to the actual observation; propagation direction consistency characterizes whether the predicted observation occurs in the correct direction along the directed forwarding path; propagation range consistency characterizes whether the downstream observation position covered by the measured message loss matches the expected impact range of the candidate fault segment; and quantity conservation relationship characterizes whether there is a corresponding quantity relationship between the direct bit error observation and the downstream message loss observation for the same batch of damaged messages.

[0053] The system combines the deviations corresponding to the above consistency relationships into the location residuals of candidate fault sections. The more consistent the predicted time-series fingerprint of a candidate fault section is with the measured bit error time-series fingerprint, the smaller the corresponding location residual, indicating that the candidate fault section explains the current bit error event to a higher degree. Based on the location residuals of each candidate fault section, the system determines one or more target fault sections from multiple candidate fault sections and forms a target fault section set. The target fault section set can be output as a fault location result to the network operation and maintenance platform, or it can be associated with and displayed with the corresponding network device, port, or link identifier.

[0054] This application's embodiments do not directly rely on absolute time synchronization between different observation ports. Instead, they utilize the pre-anchor message, post-anchor message, and message sequence number of common observations to establish a message progress phase, converting cumulative bit error observations and normal message missing observations to a unified message progress benchmark. Simultaneously, observation templates for candidate fault sections are constructed based on the actual directed forwarding path and message replication relationship. By combining phase position, propagation direction, propagation range, and quantity conservation relationship, the interpretability of different candidate fault sections on the measured bit error timing fingerprint is distinguished. This allows the direct bit error occurrence location to be associated with the subsequent missing range of damaged messages, determining the target fault section set corresponding to the bit error event.

[0055] Regarding the implementation details of obtaining bit error events in the method of this application embodiment, in some examples of this application embodiment, the collection timestamp and counter modulus of the cumulative direct bit error count of each observation port are obtained, and the sampling interval is divided accordingly; wherein, the cumulative direct bit error count includes at least one of uncorrectable codeword count and frame check error count, and each observation port uses a consistent digest generation rule and a reproducible sampling rule based on message content to generate the normal message digest.

[0056] In practice, the interval between two adjacent data collection timestamps at the same observation port can be defined as a sampling interval, and the cumulative direct bit error count read at the end of the interval can be associated and saved with the cumulative direct bit error count read at the beginning of the interval. The reproducible sampling rule can form sampling criteria based on the stable field and the preset digest field in the message header. For example, the stable field and the preset digest field can be digested, and whether to record the corresponding message can be determined based on whether the digest result falls within the preset sampling range. Different observation ports use the same fields, the same digest processing method, and the same sampling range, so that the same message can obtain consistent sampling judgment and digest identification when passing through different observation ports.

[0057] Then, the initial increment is determined based on the difference in the cumulative direct bit error counts between adjacent sampling intervals. When the cumulative direct bit error count changes in the opposite direction, counter wrap-around compensation is performed. Counter wrap-around and counter reset are distinguished based on whether the compensated increment exceeds the preset maximum count growth amount of the corresponding sampling interval. Sampling intervals where the count change cannot be recovered are determined as invalid sampling intervals. The remaining increments are converted into the equivalent of damaged messages. Based on the hierarchical response relationship of different types of bit error counts, the equivalent of damaged messages that repeatedly represent the same damaged message is counted only once to obtain the bit error increment.

[0058] For observation port Sampling interval and error counting type The initial increment or the increment after wraparound compensation is determined according to the following formula. :

[0059] Equation (1)

[0060] In the formula, For the observation port index, For sampling interval index, For error count type index, and Sampling intervals The cumulative direct error count read at the end and the beginning. For the corresponding counter modulus, This is the initial increment or the increment after wraparound compensation. The counter modulus is determined based on the counter bit width or the counter attributes reported by the device.

[0061] When the cumulative direct bit error count changes in the opposite direction, Compared with the preset maximum count increase A comparison is then made. The preset maximum count increase can be determined based on the observation port rate, sampling interval length, message length range, and the maximum count increase that a single damaged message can cause. If No more than Then the corresponding reverse change will be identified as counter rewind; if Exceed If the error occurs, it is identified as a counter reset. If the actual increase cannot be recovered from the reset record or other device status, the corresponding sampling interval is marked as an invalid sampling interval. Invalid sampling intervals are not included in the background error statistics as zero error intervals.

[0062] The damaged message equivalent is used to unify different counting units to a single damaged message count. For uncorrectable codeword counting, a conversion factor can be determined based on the port encoding method and the correspondence between codewords and messages; for frame check error counting, a conversion factor can be determined based on the semantics of error frame counting. When both error counts are available, the error increment... It can be determined according to the following formula:

[0063] Equation (2)

[0064] In the formula, For observation port In the sampling interval Error increment after conversion of internal volume equivalent and hierarchical deduplication and These represent the uncorrectable codeword count type and the frame check error count type, respectively. and These are the conversion factors for converting the corresponding counting type to the number of damaged packets. The overlap coefficient is used to represent the same damaged message repeatedly for two counting types, and The conversion factor can be determined based on the port encoding method, frame format, device counting semantics, or calibration data; the overlap factor is determined based on the inclusion relationship between the two counts in the device's receiving and processing layer, taking 1 for complete duplication and 0 for no duplication. When only one type of cumulative direct bit error count is obtained, the equivalent of the damaged message obtained by converting that count type is directly determined as... Invalid sampling intervals are not calculated. The resulting bit error increments are used as inputs in subsequent processing to determine the bit error event window and the measured direct bit error distribution.

[0065] Subsequently, the upper limit of background bit error rate is determined based on the bit error increment distribution during normal operation. The effective sampling interval that continuously exceeds the upper limit of background bit error rate is determined as the bit error burst interval, and then extended forward to the stable normal sampling interval and backward to the sampling interval where the bit error increment recovers and the normal message digest stabilizes again, thus obtaining the bit error event window.

[0066] When using quantile position to determine the upper limit of background bit error, the observation port Corresponding background error limit It can be determined according to the following formula:

[0067] Equation (3)

[0068] In the formula, The upper limit of background bit error rate, This indicates that the quantile level is taken for the set of bit error increments. The corresponding quantiles, , For observation port The set of valid sampling intervals during normal operation. Quantile level. The allowable background bit error rate (BER) is set based on the allowable background BER level or the tail distribution of historical BER increments. When the normal operating sample is insufficient to form a statistical distribution, the allowable BER level for the device can also be determined as follows: .

[0069] When the bit error increment continuously exceeds the corresponding background bit error limit within a preset number of adjacent valid sampling intervals, these sampling intervals are defined as bit error burst intervals. The stable normal sampling interval can be determined based on at least one of the following conditions: the bit error increment does not exceed the background bit error limit, the change in the number of normal message digests is within a preset range, and the message sequence remains continuous. Forward expansion is used to retain the normal forwarding state before the bit error burst; backward expansion extends to the position where the bit error increment recovers and the normal message digest stabilizes again, used to retain normal reference data after the bit error ends.

[0070] Subsequently, the effective forwarding table version, equivalent multipath and tunnel forwarding results and packet replication status within the bit error event window are obtained. When a change in the forwarding table version, link retraining or packet sequence reset occurs, the sub-event window is divided according to the corresponding change time. Otherwise, the bit error event window is treated as a single sub-event window, and the directed forwarding path and packet replication relationship corresponding to each sub-event window are determined.

[0071] The system can read the forwarding table entries, equivalent multipath selection results, tunnel encapsulation and decapsulation relationships, and packet replication status that are actually in effect at the corresponding time from the network controller, forwarding devices, or telemetry records. For equivalent multipath forwarding, the actual forwarding branch used is determined based on the packet flow identifier and hash selection results; for tunnel forwarding, the tunnel entrance, internal tunnel path, and tunnel exit are associated as a continuous directed forwarding relationship; for multicast, mirroring, or other replicated transmissions, the replication branch relationship between the source packet and its downstream copies is recorded.

[0072] Changes in forwarding table versions, link retraining, or message sequence resets may alter the forwarding position, observation order, or sequence continuity of messages. Therefore, the corresponding change time is used as the boundary of the sub-event window. Each sub-event window uses the actual effective forwarding state within its time range to restore the directed forwarding path and message replication relationship. When no such changes occur, the entire error event window is treated as a single sub-event window to avoid comparing the error increments and normal message digests formed under different forwarding states under the same propagation relationship.

[0073] Furthermore, normal message summaries with the same directed forwarding path and message replication branch sequence within each sub-event window and its preceding and following stable sampling intervals are divided into co-path message groups.

[0074] The co-path message groups can be divided according to the observation port sequence, forwarding node sequence, tunnel relationship, and replication branch sequence that the message actually passes through. When the forwarding information corresponding to the normal message digest is the same, they are grouped into the same co-path message group; when the directed forwarding paths or message replication branch sequences are different, they are grouped into different co-path message groups. The stable sampling intervals before and after the sub-event window are grouped using the same forwarding conditions as the corresponding sub-event window, so that normal messages in the same co-path message group have consistent upstream and downstream observation relationships and replication semantics.

[0075] In the embodiments of this application, the cumulative direct bit error count is processed through wraparound compensation, reset identification, damaged message equivalent conversion, and hierarchical deduplication to form a comparable bit error increment. The bit error increment and the stable normal message digest together define the bit error event window. Then, according to the actual effective forwarding status within the window, sub-event windows and co-path message groups are divided, so that each group of bit error observations and normal message digests correspond to a consistent directed forwarding path and message replication relationship.

[0076] Regarding the implementation details of phase reference construction in the method of this application embodiment, in some examples of this application embodiment, for any co-path message group within a sub-event window, the common observation normal message digests that belong to the co-path message group before and after the sub-event window and are closest to the sub-event window are respectively determined as the preceding anchor message and the following anchor message.

[0077] In this context, a common observation normal message digest refers to the same normal message digest that can be identified by multiple observation ports on the corresponding directed forwarding path based on the digest identifier. The digest identifier can be generated from at least one of the stability field in the message header and the payload digest. The system first filters normal message digests according to co-path message groups, and then matches observation records from different observation ports based on the digest identifier. When multiple common observation normal message digests with the same distance exist before or after the bit error event window, the normal message digest with complete observation records and consecutive message sequences on the corresponding directed forwarding path can be selected first.

[0078] Subsequently, the message progress phase of the preceding anchor message is set as the phase start point, and the message progress phase of the following anchor message is set as the phase end point. The message progress phase of the normal message digest is determined according to the proportion of the message sequence offset of any normal message digest relative to the preceding anchor message in the message sequence interval between the two anchor messages.

[0079] The message progress phase can be determined according to the following formula:

[0080] Equation (4)

[0081] In the formula, For the same-path message group index, For the observation port index, For the normal message digest of the message progress phase to be determined, For the same route message group The anchor message before the message, For the same route message group The anchor message after the anchor, Normal message digest At the observation port Corresponding co-path message group The message sequence number in the message. This corresponds to the message progress phase. The message progress phases of the preceding anchor message and the following anchor message are 0 and 1, respectively.

[0082] The message sequence can be continuously recorded by the observation port according to the local observation order of the normal message digest, or it can be determined based on the monotonically increasing sequence information carried in the message. When the message sequence wraps around, the message sequence is first expanded according to the corresponding sequence space, and then the message progress phase is calculated. Each observation port uses the two anchor messages observed at its own end for normalization, so that the different local message sequences are converted to the same phase range, without relying on the absolute time synchronization between observation ports.

[0083] Then, based on the order of the common observation normal message digests at each observation port, the largest consistent subsequence containing the most messages and maintaining a common order at each observation port is determined. Non-common observation normal message digests that do not change the order of the largest consistent subsequence are incorporated into the valid message sequence, and normal message digests that cause order conflicts are identified as reordered message digests.

[0084] In practice, the local message sequences of the common observation normal message digests at each observation port are extracted, and a message subsequence containing the largest number of messages that maintains the same order across all observation ports is searched. If multiple subsequences contain the same number of messages, the one with the smallest total message sequence offset across observation ports is selected as the largest consistent subsequence. For normal message digests appearing only at some observation ports, the allowed insertion sequence interval is determined based on the adjacent common observation normal message digests. If the local message sequence is within the specified sequence interval, it is incorporated into the valid message sequence; otherwise, it is identified as a reordered message digest. Reordered message digests are not involved in determining the phase endpoints of the sampling interval, but their digest identifier and message sequence are retained to exclude digest differences caused by reordering in subsequent upstream and downstream digest comparisons.

[0085] Furthermore, the phase range of the sampling interval is determined based on the minimum and maximum message progress phases corresponding to the valid message sequence within any sampling interval; when there is no valid normal message digest within the sampling interval, order-preserving interpolation is performed based on the phase endpoints of adjacent valid sampling intervals and the number of messages in the same path, and the phase uncertainty is determined based on the number of consecutive missing normal message digests, so that the phase uncertainty increases with the increase of the number of consecutive missing digests.

[0086] For sampling intervals with valid normal message digests, the minimum and maximum message progress phases corresponding to the valid message sequence are directly determined as the start and end points of the phase range, respectively. For consecutive intervals without valid normal message digests... The sampling intervals are denoted as follows: to And the nearest valid sampling interval before and after the missing segment is denoted as follows: and The cumulative allocation ratio is determined based on the number of co-path packets in each missing sampling interval. :

[0087] Equation (5)

[0088] In the formula, The order of the missing sampling intervals within the consecutive missing segments. For observation port In the sampling interval Internal corresponding co-path message group The number of messages, To end the sampling interval The cumulative distribution ratio, and make When the total number of co-channel messages is greater than 0, phase progress is allocated according to the number of co-channel messages in each sampling interval; when the total number of co-channel messages is 0, phase progress is allocated proportionally according to the order of missing sampling intervals. Prioritize based on sampling interval Internal targeting same-path message groups The recorded flow count is determined; if no corresponding flow count is recorded separately, it is estimated based on the inlet observation count or the number of normal message digests corrected for digest sampling probability; if neither of the above data is available, the corresponding flow count will be determined. Set it to 0, and allocate the phase progress proportionally according to the order of the missing sampling intervals.

[0089] Sampling interval The corresponding interpolation phase range can be determined by the following formula:

[0090] Equation (6)

[0091] In the formula, and These represent the start and end points of the phase range obtained through interpolation, respectively. This is the end point of the phase range of the previous valid sampling interval. This marks the starting point of the phase range for the next valid sampling interval. Interpolation must satisfy the following conditions: If this condition is not met, the reordering marker of the corresponding normal message digest is rechecked or an adjacent valid sampling interval is reselected. Since the cumulative allocation ratio does not decrease along the missing sampling intervals, the interpolated phase ranges will not reverse the message order.

[0092] Phase uncertainty can be determined based on the number of consecutively missing normal message digests, the phase interval between adjacent valid sampling intervals, and the change in the number of corresponding co-channel messages. Sampling interval The corresponding phase uncertainty can be determined by the following formula:

[0093] Equation (7)

[0094] In the formula, For observation port In the sampling interval Internal corresponding co-path message group Phase uncertainty; From the beginning of the continuous missing segment to the sampling interval The number of sampling intervals where no valid normal message digests exist consecutively; It is the difference between the start point of the phase range of the next effective sampling interval and the end point of the phase range of the previous effective sampling interval; and These represent the number of co-channel packets within the previous and next valid sampling intervals, respectively. , and The weights are non-negative, and the sum of the three is 1; and It is a positive stability constant.

[0095] The non-negative weights can be calibrated based on historical sampling intervals with complete and normal message summaries, or preset based on the impact of consecutive summary loss, phase interval, and changes in the number of co-channel messages on the interpolation deviation. The positive stability constants are determined based on the numerical precision of the message progress phase and the range of the number of co-channel messages, respectively. According to the above relationships, Furthermore, the phase uncertainty does not decrease as the number of consecutive missing normal message digests increases. The obtained phase range is used to convert the bit error increment of the corresponding sampling interval into the message progress phase, and the obtained phase uncertainty is used to reduce the reliability weight of the observations corresponding to the interpolated phase range in subsequent fingerprint matching.

[0096] Through the embodiments of this application, the front anchor message and the back anchor message convert the local message sequence of each observation port to a unified message progress phase, and the maximum consistent subsequence eliminates the interference of reordering on the phase endpoints; when the sampling interval lacks a valid normal message digest, the phase segment is continuously allocated according to the number of co-path messages, and the reliability of the interpolation result and the direct observation result is distinguished by the phase uncertainty.

[0097] Figure 2 A flowchart illustrating an example of generating a measured bit error timing fingerprint in a method according to an embodiment of this application is shown.

[0098] like Figure 2As shown, in step S210, the message progress phase is divided into continuous and non-overlapping phase intervals. The distribution ratio of the bit error increment in each phase interval is determined according to the phase range of the sampling interval and the degree of overlap between each phase interval, so that each distribution ratio can jointly represent the total bit error increment in the sampling interval. Combined with the proportion of the number of messages in each co-channel message group in the sampling interval, the measured direct bit error of each observation port in each phase interval is determined.

[0099] The phase intervals can be continuously divided with the same phase width, or they can be divided with non-uniform widths based on the distribution density of normal message digests on the message progress phase. Each phase interval collectively covers the message progress phase range between the preceding and following anchor messages, and a left-closed-right-open or right-closed-left-open boundary rule is used to avoid adjacent phase intervals repeatedly containing common boundaries. The number and width of the phase intervals can be determined based on the number of normal messages, the number of sampling intervals, and the required phase resolution, and must remain consistent within the same sub-event window.

[0100] For the non-degenerate phase range, the allocation ratio can be determined according to the following formula:

[0101] Equation (8)

[0102] In the formula, For sampling interval index, and For phase interval index, and For the corresponding phase interval, For observation port In the sampling interval Internal corresponding co-path message group The phase range, The phase range is assigned to phase intervals. Phase allocation ratio, This represents the length of the phase interval, and the summation range covers all phase intervals. Each phase interval completely covers the message progress phase range, therefore the denominator corresponding to the non-degenerate phase range is positive, and the sum of the phase allocation ratios is 1.

[0103] When a phase range overlaps with only one phase interval, the corresponding phase allocation ratio is 1. When a phase range spans multiple phase intervals, the bit error increment is allocated according to the phase length of each overlapping segment. When a phase range degenerates into a single phase position, the length ratio is not calculated, and the corresponding bit error increment is allocated to the phase interval containing that phase position. When a single phase position is located at a common boundary, its unique assignment is determined according to the aforementioned boundary rules.

[0104] The measured direct bit error rate can be determined according to the following formula:

[0105] Equation (9)

[0106] In the formula, This refers to the set of valid sampling intervals within the corresponding sub-event window. For the same route message group At the observation port sampling interval The percentage of messages within the range For observation port In the sampling interval The error increment is obtained after internal loop compensation, damaged message equivalent conversion, and deduplication. For observation port Same route message group and phase interval The corresponding measured direct bit error rate.

[0107] The message quantity ratio can be determined based on the number of normal message digests of each co-channel message group within the corresponding sampling interval, and the number of digests of different co-channel message groups is corrected according to the digest sampling probability. The sum of the message quantity ratios of each co-channel message group within the same observation port and sampling interval is 1; when the sampling interval contains only one co-channel message group, its message quantity ratio is 1. The phase allocation ratio distributes the bit error increment between phase intervals, and the message quantity ratio distributes the bit error increment between co-channel message groups, so that the sum of the measured direct bit errors of each co-channel message group and the corresponding phase interval within the same sampling interval is consistent with the bit error increment of that sampling interval.

[0108] In step S220, normal message summaries of upstream and downstream observation ports are matched according to the summary identifier and message sequence within each phase interval. The number of normal message summaries of upstream and downstream is corrected according to the corresponding summary sampling probability. The number of normal message summaries of downstream is reverse-normalized to the number of messages before replication according to the message replication relationship. The summary differences corresponding to strategy processing and reordering are excluded. The number of messages corresponding to normal message summaries that exist upstream but not downstream is determined as the measured message missing amount.

[0109] When matching normal message summaries, observation records with the same summary identifier and whose message sequence satisfies the corresponding upstream and downstream propagation order are identified as upstream and downstream observation results of the same message. When the same summary identifier corresponds to multiple normal message summaries, disambiguation is performed by combining message sequence, co-path message group, and phase interval, and the actual occurrence count of duplicate messages is retained to avoid merging different messages into the same summary record.

[0110] Upstream observation port and downstream observation ports The number of normal message digests collected are denoted as follows: and The corresponding abstract sampling probabilities are denoted as follows: and ,and , The number of upstream normal message digests corrected for sampling probability is: The cumulative message replication factor from the upstream observation port to the downstream observation port is denoted as . ,and The number of downstream normal message digests, after sampling probability correction and inverse normalization to the number of messages before replication, is: When there is no packet replication in the corresponding forwarding segment, the cumulative packet replication coefficient is set to 1.

[0111] The digest differences corresponding to policy processing can be determined based on the hit records of access control, active dropping, sampling mirroring, traffic shaping, or other known forwarding policies; the digest differences corresponding to reordering can be determined based on the aforementioned reordered packet digests and their upstream and downstream correspondences. The digest differences corresponding to policy processing and reordering are adjusted using the same sampling probability correction and necessary replication inverse normalization to form the exclusion quantity, ensuring that the exclusion quantity and the upstream and downstream corrected digest quantities use the same source packet quantity caliber.

[0112] The measured message missing amount can be determined according to the following formula:

[0113] Equation (10)

[0114] In the formula, For upstream observation port index, For downstream observation port index, This represents the number of upstream normal message digests corrected for digest sampling probability. This represents the number of downstream normal message digests that have been corrected for digest sampling probability and inversely normalized according to message replication relationships. This refers to the exclusion amount corresponding to the strategy processing and reordering after being corrected for the same caliber. , To correspond to the upstream and downstream observation ports in the same message group and phase interval The actual number of missing messages within the system.

[0115] When the number of upstream corrected summaries does not exceed the sum of the number of downstream corrected summaries and the exclusion amount, the measured message missing amount is taken as 0; when the number of upstream corrected summaries is large, the portion that cannot be explained by downstream summaries and known exclusion reasons will be determined as the measured message missing amount. Non-negative truncation can also avoid negative missing results caused by summary sampling fluctuations or correction errors.

[0116] In step S230, the measured direct bit error rate and the measured message missing rate are arranged according to the observation type, observation location, co-path message group and phase interval to generate the measured bit error timing fingerprint composed of multiple observation units.

[0117] The measured error timing fingerprint can be represented as:

[0118] Equation (11)

[0119] In the formula, For the set of observation units, For indexing the observation units, For observation unit The corresponding measured direct bit error rate or measured message loss rate, This is the timing fingerprint of the measured bit error rate.

[0120] For measured direct bit errors, the observation location is identified by the corresponding observation port; for measured message missing quantities, the observation location is identified by observation port pairs with upstream and downstream relationships. The system numbers the observation units according to a fixed order of observation type, observation location, co-channel message group, and phase interval. The observation template and predicted time series fingerprint use the same numbering rule, so that measured and predicted quantities with the same index correspond to the same observation meaning.

[0121] In step S240, for co-path message groups with reliable transmission semantics, retransmission and reproduction messages that appear after a message is missing and have the same transmission sequence number and payload summary as the missing message are identified. The retransmission feasible interval is determined based on the historical round-trip delay and retransmission timeout range, and the corresponding message missing, retransmission time interval and the retransmission feasible interval are associated as retransmission and reproduction information.

[0122] The reliable transmission semantics can be determined based on the transmission protocol type, connection status, and message control fields. In addition to having the same transmission sequence number and payload digest, retransmitted and missing messages can also have the same connection identifier to avoid identifying the same payload in different connections as a retransmission. The observation time of the missing message at the upstream observation port is used as the missing reference time, and the difference between the observation time of the retransmitted and retransmitted message and the missing reference time is determined as the retransmission time interval.

[0123] Historical round-trip timeouts can be determined based on historical records of corresponding co-path message groups, connections, or network paths. Retransmission timeout ranges can be obtained from protocol status, device records, or historical retransmission interval distributions. The system determines a feasible retransmission interval based on the fluctuation range of historical round-trip timeouts and the retransmission timeout range. If the retransmission time interval falls within the feasible retransmission interval, it indicates that the retransmission reproduction and the corresponding message loss have a temporally interpretable relationship; if it falls outside the feasible retransmission interval, the result is recorded as retransmission time infeasibility information for subsequent causal consistency evaluation.

[0124] The retransmission and reconstruction information is used to characterize whether the corresponding missing message has been retransmitted and reconstructed within the recovery time allowed by the transmission protocol. When there is no reliable transmission semantics or no retransmission and reconstruction message is observed, the corresponding retransmission and reconstruction information may not be generated; the failure to generate retransmission and reconstruction information does not change the corresponding measured message missing amount and its observation unit in the measured bit error timing fingerprint.

[0125] In the embodiments of this application, the bit error increment within the sampling interval is fully allocated as the measured direct bit error amount according to the phase coverage relationship and the number of co-path messages. The upstream and downstream summary differences are converted into the measured message missing amount after sampling probability correction, replication reverse normalization and known difference exclusion. The two types of observations are entered into the measured bit error timing fingerprint using the same arrangement rules, and the corresponding retransmission timing information is retained.

[0126] Regarding the implementation details of the observation template construction in the method of this application embodiment, in some examples of this application embodiment, the receiving boundary of the directed physical link, the internal processing boundary of the receiving port, and the input / output boundary of the forwarding node are respectively determined as candidate fault sections.

[0127] Candidate fault profiles are used to characterize the processing boundaries at which damaged packets may cease normal propagation. The receiving boundary of a directed physical link corresponds to the boundary where a packet arrives at the receiving side via the physical medium; the internal processing boundary of a receiving port corresponds to the boundary between different receiving and processing levels within the receiving port; and the input-output boundary of a forwarding node corresponds to the boundary where a packet transitions from input processing to output processing at the node.

[0128] In practice, candidate fault sections are set segment by segment based on the physical links, receiving ports, and forwarding nodes in the directed forwarding path, and the network device, port direction, processing level, and corresponding co-path packet group of each candidate fault section are recorded. When different processing semantics exist at the same network location, candidate fault sections are set separately to distinguish between anomalies caused by physical medium transmission, internal processing of the receiving port, and node input / output processing.

[0129] Then, the cumulative message replication coefficient is determined based on the message replication relationship, and the actual subsequent forwarding range of each candidate fault section is determined based on the directed reachability relationship from each candidate fault section to each observation port, the observation type, the phase propagation relationship, and the cumulative message replication coefficient. An observation mapping relationship between the unit damaged message quantity and the expected quantity of each observation unit is established to form the observation template.

[0130] The cumulative message replication factor characterizes the number of copies formed when a source message propagates from a candidate fault section to the corresponding observation location. The observation mapping relationship arranges the expected number of damaged messages per unit according to the same observation type, observation location, co-path message group, and phase interval as the measured error timing fingerprint.

[0131] For cases where the candidate fault section passes through multiple message replication nodes sequentially from the observation location, the cumulative message replication coefficient can be determined according to the following formula:

[0132] Equation (12)

[0133] In the formula, For candidate fault aspect index, For observation location index, For the same-path message group index, For the same route message group The message is from the candidate fault section Propagation to the observation location The corresponding cumulative message replication factor at that time For candidate fault sections To the observation position It has a replication function during propagation and its copies can reach the observation location. The set of message replication nodes, For the same route message group In message replication node It forms and can continue to reach the observation position. The local message replication coefficient. When there are no message replication nodes between the candidate fault section and the observation location, let... .

[0134] The local message replication coefficient is determined based on the actual active multicast branch, mirror copy, or other replication configuration of the corresponding replication node. If only some replication branches can reach the observation location, the remaining unreachable branches are not included. The actual subsequent forwarding range of a candidate fault segment consists of the observation ports and observation port pairs reachable after that segment according to the effective forwarding status within the corresponding sub-event window. Observational units on reverse paths, unreachable paths, and inactive replication branches are not included in the expected impact range.

[0135] The observation type distinguishes between direct error observations and message missing observations; the directed reachability relation determines whether a damaged message can affect the corresponding observation location; the phase propagation relation maps the source phase interval of the damaged message to the phase interval of the downstream observation location. The phase propagation relation can be determined based on the phase correspondence results of normal message digests in the same message group at the upstream and downstream observation ports. When one source phase interval corresponds to multiple observation phase intervals, the expected quantity is allocated among the corresponding observation units according to the degree of phase overlap.

[0136] Subsequently, using the observation template, the amount of non-negative damaged messages in different co-channel message groups and phase intervals of each candidate fault section is mapped to a predicted timing fingerprint arranged in the same way as the measured bit error timing fingerprint.

[0137] The mapping process can be represented as:

[0138] Equation (13)

[0139] In the formula, For candidate fault sections The corresponding observation mapping operator, For candidate fault sections The vector of damaged packets within each co-path packet group and source phase interval This corresponds to the predicted time-series fingerprint. The observation mapping operator records the expected direct bit error rate or expected message loss rate generated by the unit of damaged message in each observation unit.

[0140] Each component of the damaged packet vector corresponds to a co-path packet group and a source phase interval, and satisfies the following conditions: The rows of the observation mapping operator correspond one-to-one with the observation units in the measured bit error timing fingerprint, and the columns correspond one-to-one with each component in the damaged message quantity vector. When a damaged message can form the expected observation in the corresponding observation unit, the corresponding position records the non-negative mapping quantity determined according to the observation type, phase propagation relationship, and cumulative message replication coefficient; when the corresponding observation position is unreachable or cannot form the corresponding observation type, the corresponding position is set to 0.

[0141] The predicted timing fingerprint and the measured bit error timing fingerprint use the same definition and arrangement order of observation units. The direct bit error observation units are arranged according to the observation port, co-channel message group, and phase interval, while the message missing observation units are arranged according to the upstream and downstream observation port pairs, co-channel message group, and phase interval, so that the expected quantities in the predicted timing fingerprint can correspond item by item with the measured quantities with the same observation semantics.

[0142] Furthermore, for the receiving boundary of the directed physical link, the corresponding observation template is configured to generate the expected direct bit error rate on the receiving side and the expected message missing rate within the actual subsequent forwarding range; for the internal processing boundary of the receiving port, the corresponding observation template is configured to generate the expected direct bit error rate based on the hierarchical response relationship between the uncorrectable codeword count and the frame check error count, and the expected message missing rate within the actual subsequent forwarding range; for the input-output boundary of the forwarding node, the corresponding observation template is configured not to generate the expected direct bit error rate but to generate the expected message missing rate within the corresponding output path and its actual subsequent forwarding range; for multiple downstream missing results generated by message replication, reverse normalization is performed according to the cumulative message replication coefficient.

[0143] For the receive boundary of a directed physical link, a damaged packet can trigger an uncorrectable codeword count or frame check error count on the receiving side, and due to the failure to continue normal forwarding, it will form an expected packet loss quantity downstream. For the internal processing boundary of the receive port, the type of direct error observation that can be formed is determined according to the order of error correction processing, frame check processing, and receive processing, and an expected packet loss quantity is formed within the actual subsequent forwarding range. For the input-output boundary of the forwarding node, if the packet has been normally observed on the input side, it will not form an expected direct error quantity on the receiving side; if the packet does not enter the expected output path, an expected packet loss quantity will be formed on the output path and its reachable downstream locations.

[0144] When a source message is replicated to form multiple downstream replicas, the expected missing amount of each replica is entered into the observation unit of the corresponding replication branch. The corresponding replica missing amount is reverse-normalized to the source message quantity according to the cumulative message replication coefficient, so as to avoid interpreting multiple missing replicas of the same damaged source message as multiple independent damaged messages.

[0145] The different response methods described above result in different combinations of direct bit errors and missing messages for physical reception anomalies, internal receiver port anomalies, and node input / output processing anomalies. Reverse normalization ensures that missing results from multiple downstream replicas still correspond to the same damaged source message volume.

[0146] Regarding the implementation details of determining the residual of each candidate fault section in the method of this application embodiment, in some examples of this application embodiment, an elastic mapping relationship is constructed for each observation port, which keeps the endpoint unchanged, maintains the message order, and has a local scaling degree within a preset range, and the phase deformation degree is determined according to the degree of deviation of the elastic mapping relationship relative to the time when no phase deformation occurs.

[0147] Specifically, an elastic mapping function is set for each observation port. This elastic mapping function keeps the phase start and phase end points unchanged and remains monotonic between adjacent phase intervals. The degree of phase deformation can be determined according to the following formula:

[0148] Equation (14)

[0149] In the formula, For observation port The corresponding elastic mapping function, and The boundary phase of adjacent phase intervals. This represents the degree of phase deformation for each elastic mapping function relative to the state without phase deformation. The boundaries of adjacent phase intervals satisfy... The summation iterates through all observation ports and all adjacent phase interval boundaries.

[0150] The elastic mapping function can be a piecewise linear function with the boundaries of each phase interval as control points, and must satisfy the following constraints:

[0151] Equation (15)

[0152] In the formula, and These are the lower and upper limits of local phase scaling, respectively. They can be determined based on the phase offset distribution of the same normal message digest from different observation ports during normal operation, or preset based on the sampling period, port processing delay, and allowable local queuing fluctuations. Endpoint constraints maintain the phase positions of preceding and following anchor messages, a positive lower limit of local slope prevents message order reversal, and an upper limit of local slope limits excessive stretching of the phase interval.

[0153] When no phase deformation occurs, each phase position is mapped to itself, and the local slope between adjacent control points is 1. The greater the deviation of the local slope from 1, the greater the resulting phase deformation. This phase deformation is not directly included in the current positioning residual, but is used to constrain the elastic mapping amplitude in subsequent joint inversion, preventing candidate fault sections from relying on excessive phase transformation to match measured observations.

[0154] Then, the reliability weight of each observation unit is determined according to the phase uncertainty and the summary sampling probability, and the reliability weight decreases as the phase uncertainty increases or the summary sampling probability decreases; the actual measurement assigned to any candidate fault section is determined as the candidate allocation quantity, the difference between the candidate allocation quantity and the corresponding predicted quantity after elastic mapping is converted into a relative deviation relative to the corresponding actual measurement, and the phase position matching residual is determined according to the reliability weight.

[0155] Among them, the observation unit Actual measurement Assigning candidate fault sections The quantity is denoted as the candidate allocation quantity. , observation unit The reliability weight is denoted as The reliability weight takes a value between 0 and 1 and is used to adjust the contribution of the corresponding relative deviation to the positioning residual.

[0156] The reliability weights can be determined through a preset mapping table or a monotonic mapping relationship calibrated based on normal operation data. For observation units whose phase range is directly determined by valid normal message digests and have a high digest sampling probability, a larger reliability weight is set; for observation units whose phase range is obtained through order-preserving interpolation, have a large number of consecutive missing normal message digests, or have a low digest sampling probability, a smaller reliability weight is set. The digest sampling probability corresponding to the observed unit with the measured message missing type can be the smaller value between the digest sampling probabilities of the upstream and downstream observation ports.

[0157] The candidate allocation and the predicted quantity after elastic mapping use the same observation type, observation location, co-path message group, and phase interval caliber. Dividing the difference between the two by the sum of the stability constants of the corresponding measured and normal measurements can reduce the impact of differences in the numerical scale of different observation units on the matching results, and maintain a limited bias calculation result when the measured quantity is zero or close to zero.

[0158] Subsequently, when the observation template of the candidate fault section includes both the expected direct bit error rate and the expected message missing rate, the total measured direct bit error rate allocated to the candidate fault section is compared with the total measured message missing rate after replication and reverse normalization, and the quantity conservation residual is determined based on the ratio of the absolute difference between the two to the sum of the two; when the corresponding observation template only includes the expected message missing rate, the quantity conservation residual is not set.

[0159] Let the set of observable units belonging to the measured direct bit error type in the measured bit error time sequence fingerprint be denoted as . The set of observation units that belong to the type of missing measured messages is denoted as Candidate Fault Section The corresponding measured total direct bit error rate is The total number of missing messages measured was The measured message missing quantity has already been reverse-normalized to the number of damaged source messages according to the cumulative message replication coefficient when generating the measured error sequence fingerprint. Here, it is directly summed without a second reverse normalization.

[0160] The quantity conservation comparison is limited to the actual subsequent forwarding range of the candidate fault section and the receiving location that can form a direct bit error observation. For the receiving boundary of the directed physical link and the internal processing boundary of the receiving port that can simultaneously form direct bit errors and downstream packet loss, the quantity conservation residual is enabled; for the input and output boundaries of the forwarding node that do not form the expected direct bit error amount, the quantity conservation residual is not enabled.

[0161] Furthermore, a causal penalty is determined based on whether there are propagation direction conflicts, propagation range discrepancies, or retransmission time intervals outside the corresponding retransmission feasible intervals among the candidate allocation quantities, and the phase position matching residual, the quantity conservation residual, and the causal penalty are combined into the positioning residual; when there is no retransmission reproduction information, no causal penalty is set for the retransmission time.

[0162] A propagation direction conflict occurs when the candidate allocation is located upstream of the candidate fault section, or when its phase order is inconsistent with the propagation order of the directed forwarding path. A propagation range mismatch occurs when the candidate allocation is outside the actual subsequent forwarding range of the candidate fault section, or when the downstream missing range indicated by the predicted timing fingerprint is inconsistent with the measured missing range of the message. A retransmission time infeasibility occurs when the retransmitted message and the missing message have the same transmission sequence number and payload digest, but their retransmission time interval is outside the corresponding retransmission feasible interval.

[0163] The system statistically analyzes the candidate allocation quantities corresponding to propagation direction conflicts, propagation range mismatches, and infeasible retransmission times. It then converts each statistical result into a ratio of the conflict observation to the total number of corresponding candidate allocations, and generates a normalized non-negative causal penalty value based on preset non-negative combination coefficients. When no conflicts exist, Set to 0; when there is no retransmission reproduction information, do not calculate the component corresponding to the infeasibility of retransmission time.

[0164] The positioning residual can be determined according to the following formula:

[0165] Equation (16)

[0166] In the formula, For the set of observation units, The phase mapping operator is composed of the elastic mapping functions of each observation port. For predicting time-series fingerprints After elastic mapping in the observation unit The predicted quantity, To measure the stability constant, For candidate fault sections The positioning residual. For observation units with direct bit error type, phase mapping is performed using the elastic mapping function of the corresponding observation port; for observation units with message loss type, phase mapping is performed using the elastic mapping function of the corresponding downstream observation port.

[0167] in, To enable quantity conservation, a flag is set to 1 if the corresponding observation template contains both the expected direct bit error rate and the expected message missing rate, and 0 otherwise. The positive message volume stability constant, The quantity conservation constraint coefficient is positive. , and The meaning is as described above.

[0168] The positive observation measurement stability constant and the positive message quantity stability constant can be determined based on the minimum effective value of the observation unit and the minimum resolution of the damaged message equivalent, respectively. The quantity conservation constraint coefficient can be calibrated based on the numerical range of the phase position matching residual and the quantity conservation residual in historical fault samples. The positioning residual increases with the increase of phase position deviation, quantity conservation deviation, or causal conflict degree, and the obtained... Used for subsequent competitive allocation solutions and target fault profile filtering.

[0169] Through the embodiments of this application, the elastic mapping relationship absorbs finite phase shifts without changing the message order; the reliability weight reduces the impact of observations with high summary sparsity and interpolation uncertainty on the matching results; the quantity conservation residual and causal penalty respectively constrain the quantity relationship of damaged messages and the propagation feasibility.

[0170] Figure 3 A flowchart illustrating an example of determining the target fault profile set in a method according to an embodiment of this application is shown.

[0171] like Figure 3 As shown, in step S310, a candidate fault section allocation flag is set for each observation unit, and an enabled state is set for each candidate fault section, so that any observation unit has at most one valid allocation flag, the actual measurement allocated to the candidate fault section does not exceed the actual measurement of the corresponding observation unit, and the actual measurement can only be allocated to the candidate fault section in the enabled state.

[0172] The candidate fault section assignment marker is used to record the exclusive attribution relationship between observation units and candidate fault sections. When an observation unit has a valid assignment marker, the assigned measurements are used to evaluate the observation interpretation of the corresponding candidate fault section, and the remaining measurements are treated as uninterpreted observations. The enabled status of a candidate fault section is used to control whether the candidate fault section participates in joint adjustment; candidate fault sections in the disabled status do not receive measurements.

[0173] The candidate allocation amount can be smaller than the actual measurement of the corresponding observation unit to retain the portion that cannot be explained by sampling error, background bit error, or existing candidate fault sections. Since the same observation unit has at most one valid allocation tag, any measured direct bit error or measured message missing amount will not be reused by multiple candidate fault sections.

[0174] In step S320, the amount of damaged messages in each candidate fault segment is constrained to be non-negative and not exceed the preset upper limit of the amount of damaged messages. The amount of damaged messages in the candidate fault segments that are not enabled is constrained to be zero, and the sum of the allocated actual measurements is not less than the preset coverage ratio of all actual measurements.

[0175] The above allocation and enable constraints can be broken down as follows:

[0176] Equation (17)

[0177] Equation (18)

[0178] In the formula, For observation unit Relative to candidate fault section binary assignment variables, For candidate fault sections Candidate enable variables, The sum of the components of the damaged message vector. The upper limit of the number of damaged messages is positive. For the lowest observation coverage ratio and ;about The summation iterates through all candidate fault sections, regarding Summation of the set of observation units .

[0179] when When the value is 0, the above constraints can make the corresponding , and All are 0; when When the value is 1, the amount of damaged messages can be adjusted within a non-negative range. It can be determined based on the total number of messages in the sub-event window or the interpretable upper bound of all actual measurements. It can be set according to the proportion of interpretable observations in normal operation data or the inversion results of historical fault samples, and is used to exclude all or most of the degradation results that have not been explained by actual measurements.

[0180] In step S330, a competitive allocation evaluation relationship is constructed based on the location residuals of each enabled candidate fault section, the phase deformation degree of the elastic mapping relationship, the proportion of actual measurements not explained by the candidate fault section relative to the corresponding actual measurements, and the number of enabled candidate fault sections.

[0181] The competitive allocation evaluation relationship can be represented by the following competitive allocation objective function:

[0182] Equation (19)

[0183] In the formula, Assign an objective function to the competition. The phase deformation constraint coefficient is positive. Positive unexplained observation constraint coefficients The candidate number constraint coefficient is non-negative; , and The meaning is as described above.

[0184] The competitive allocation objective function's first term summarizes the location residuals of activated candidate fault sections; the second term restricts phase distortion caused by matching measured bit error timing fingerprints; the third term evaluates measured data not explained by candidate fault sections; and the fourth term suppresses the activation of unnecessary candidate fault sections. Each constraint coefficient can be calibrated based on the value range of the corresponding evaluation item in historical fault samples, ensuring that different evaluation items have comparable influence in the competitive allocation objective function. When it is 0 or close to 0, Used to keep the evaluation items for unexplained observations limited.

[0185] In step S340, the allocation results of the observation units, the activation status of the candidate fault sections, the amount of damaged messages, and the elastic mapping relationship are jointly adjusted to reduce the comprehensive deviation represented by the competitive allocation evaluation relationship, and the predicted time-series fingerprint of each candidate fault section is generated based on the adjusted amount of damaged messages and the corresponding observation template.

[0186] In practice, under the constraints of allocation, activation, damaged packet volume, and elastic mapping local scaling, the binary allocation variables, candidate allocation quantities, candidate activation variables, damaged packet volume vector, and elastic mapping function are jointly solved. First, the binary allocation variables and candidate allocation quantities can be adjusted while keeping the candidate activation state and elastic mapping relationship unchanged. Then, the non-negative damaged packet volume is solved based on the adjusted candidate allocation quantities. Subsequently, the elastic mapping relationship is adjusted under the conditions of unchanged endpoints, monotonicity, and limited local scaling, and the location residual is recalculated.

[0187] After completing one round of adjustments, the candidate activation status is adjusted based on the updated positioning residuals and the amount of damaged messages. The result of this round is accepted only if the adjustment result satisfies all constraints and the competitive allocation objective function does not increase. Joint adjustment ends when the change in the objective function value between two adjacent rounds does not exceed a preset convergence threshold, the allocation result remains unchanged within a preset number of rounds, or the preset maximum number of adjustment rounds is reached. Subsequently, the damaged message quantity vector is input into the corresponding observation template to generate an optimized prediction time-series fingerprint and update the corresponding positioning residuals.

[0188] In step S350, candidate fault sections that are in the enabled state and whose total number of damaged messages is greater than a preset lower limit of damage are determined as initial target fault sections. Each initial target fault section is then deactivated in turn and a joint adjustment is performed. The independent interpretation contribution of each initial target fault section to the measured bit error timing fingerprint is determined based on the comprehensive deviation increment before and after deactivation. The initial target fault sections whose positioning residual is less than a preset upper limit of residual and whose independent interpretation contribution is greater than a preset lower limit of contribution are formed into the target fault section set.

[0189] When any initial target fault section is deactivated, the corresponding candidate activation variable is set to 0, and the competitive allocation objective function is resolved while keeping other constraints unchanged. The contribution residual increment of this initial target fault section can be determined according to the following formula:

[0190] Equation (20)

[0191] In the formula, The optimal value of the competitive allocation objective function obtained from the original joint adjustment is given. To cut the candidate fault surface The constraint is the optimal value of the objective function obtained after readjustment from the inactive state. For candidate fault sections The larger the contribution residual increment, the more difficult it is for other candidate fault sections to replace this candidate fault section in interpreting the measured bit error timing fingerprint.

[0192] A preset lower limit for damage is used to exclude candidate fault sections that, although enabled, have a total number of damaged packets close to zero. A preset upper limit for residuals is used to constrain the interpretation bias of candidate fault sections towards allocated actual measurements. A preset lower limit for contribution is used to exclude candidate results that can be replaced by other candidate fault sections. For concurrent faults, each initial target fault section can interpret non-overlapping observation units. After one of the initial target fault sections is deactivated, its original candidate allocation can be re-competed for by other enabled candidate fault sections, or it can be converted into uninterpreted observations. The corresponding objective function increment reflects the independent interpretation contribution of that initial target fault section.

[0193] In step S360, when the difference in the location residuals of at least two target fault sections is less than a preset ambiguity threshold, an acknowledgment detection path is generated based on the difference segment between the actual subsequent forwarding ranges of the corresponding target fault sections. The detection results are converted into supplementary observations. The measured bit error timing fingerprint is updated based on the supplementary observations. The corresponding observation template is expanded based on the acknowledgment detection path, the observation position of the supplementary observations, and the message replication relationship. The joint adjustment is then performed to update the target fault section set.

[0194] Specifically, the detection path identifies path segments whose actual subsequent forwarding range differs across different target fault sections, enabling the detection results to distinguish between multiple approximate candidate interpretations. These differing segments can be path segments affected only by some target fault sections, or path segments where different target fault sections are expected to generate different observation types. The detection can employ controlled detection messages or temporary enhanced summary observations of existing service messages.

[0195] The detection results are converted into supplementary observations according to the observation type, observation location, co-path message group, and phase interval consistent with the original measured bit error timing fingerprint. After the supplementary observations are added to the measured bit error timing fingerprint, observation units with the same observation semantics are added to the corresponding observation template based on the directed reachability of the confirmed detection path, the observation location of the supplementary observations, and the message replication relationship, so that the measured side and the prediction side maintain a one-to-one correspondence.

[0196] After readjustment, if one target fault section can explain the supplementary observations while another target fault section cannot generate corresponding predictions, their localization residuals and independent interpretation contributions will differ, and the target fault section set will be updated accordingly. If no target fault section meets the preset ambiguity conditions, the current target fault section set is directly output.

[0197] Through the embodiments of this application, exclusive allocation constraints restrict the repeated use of the same observation evidence, minimum coverage ratio excludes degradation results that are basically unexplained by actual measurements, and deactivate the contribution residual increment obtained from recalculation to eliminate explanations that can be replaced by other candidate fault sections; when existing observations still cannot distinguish the target fault section, the confirmation probe adds corresponding observations in the differential forwarding segment to update the target fault section set.

[0198] Figure 4 A schematic diagram illustrating the system operation mechanism of an example of a packet link fault location method based on bit error timing fingerprinting according to an embodiment of this application is shown.

[0199] like Figure 4 As shown, the system acquires the cumulative direct bit error count and normal message digests from multiple observation ports, and obtains the corresponding forwarding status and message replication relationship. The system determines the preceding and following anchor messages from the normal message digests before and after the bit error event window, uses these anchor messages to define message sequence segments, constructs a normalized message progress phase based on the message sequence position of each observation port between the two anchor messages, and associates the sampling interval and its bit error increment with the corresponding phase range to reduce the impact of asynchronous sampling offset between different observation ports on observation alignment. The forwarding status and message replication relationship are used to determine the actual propagation direction of the message, the subsequent forwarding range, and the replication branch.

[0200] The system allocates the incremental increase of the cumulative direct bit error count to the corresponding phase interval based on the phase range of each sampling interval, thus obtaining the measured direct bit error rate distribution. Simultaneously, it compares the normal message digests collected by observation ports with upstream and downstream relationships in the directed forwarding path, and after digest sampling correction and reverse normalization of replication, obtains the measured message missing rate distribution. The measured direct bit error rate distribution and the measured message missing rate distribution are arranged according to a unified observation caliber to form a measured bit error rate time sequence fingerprint. Furthermore, for different candidate fault sections, the system constructs observation templates based on directed reachability, observation type, phase propagation relationship, and message replication relationship, mapping the damaged message quantity of each candidate fault section to a predicted time sequence fingerprint with the same observation unit arrangement as the measured bit error rate time sequence fingerprint.

[0201] Under the constraint that any measured direct bit error rate or measured message missing rate can be assigned to at most one candidate fault section, the system matches the measured bit error rate timing fingerprint with the predicted timing fingerprint of each candidate fault section. While maintaining message order and limiting the degree of local phase scaling, the system adjusts the phase correspondence, observation assignment, and the amount of damaged messages in each candidate fault section through monotonically elastic joint inversion, and applies a quantity conservation constraint based on the relationship between the total measured direct bit error rate and the total measured message missing rate after replication and inverse normalization. The system determines the localization residual based on the consistency between the predicted timing fingerprint of each candidate fault section and the corresponding measured observation in terms of phase position, propagation direction, propagation range, and quantity relationship. Combining this with the independent interpretation contribution of each candidate fault section to the measured bit error rate timing fingerprint, the system obtains the target fault section set.

[0202] To verify the effectiveness of the monotonic elasticity, propagation causality, and quantity conservation mechanisms in the proposed method, experiments were conducted using a parameterized high-performance packet network discrete event simulation platform. The simulation topology employed a two-layer Clos interconnect architecture for a data center, comprising 16 leaf switches and 8 spine switches. Link bandwidth was configured with a mix of 100 Gbps and 400 Gbps. Switch nodes were configured with either 16 MB or 32 MB of dynamic shared buffer pools, and ECMP (Equal-Cost Multi-Path) hash forwarding was enabled between leaf and spine switches. The traffic model employed a mix of long and short data center traffic, encompassing services based on RoCEv2 (RDMA over Converged Ethernet version 2) and TCP (Transmission Control Protocol). Background link load dynamically varied between 35% and 85%. The default sampling rate for normal message digests was set to 1 / 1024. In robust response surface experiments, the sampling rates for normal message digests were further set to 1 / 256, 1 / 512, 1 / 1024, 1 / 2048, 1 / 4096, and 1 / 8192, respectively. Port-level telemetry data included frame check error counts, uncorrectable codeword counts, and dropped codeword counts. The polling period was randomly set between 100 ms and 1000 ms, and an initial polling phase was independently set for different devices to simulate asynchronous sampling and observations lacking valid normal message digests within continuous sampling intervals.

[0203] The fault and interference event set includes four scenarios: short-term physical link errors, with 2 to 60 frame check errors occurring suddenly within 20 ms to 400 ms; receiver port channel anomalies, causing the uncorrectable codeword count and frame check error count to increase sequentially according to the receiving processing level; silent dropping within forwarding nodes, resulting in the existence of normal message digests on the input side but the corresponding output digest being missing, without generating direct error counts; and background congestion packet loss, accompanied only by an increase in queue occupancy and output drop counts, serving as a control event without physical errors. In the fault events, the ratio of single-fault events to concurrent double-fault events was set to 70% and 30%, respectively. Each test case also randomly superimposed device clock skew from −2 s to +2 s, clock drift of ±80 ppm (parts per million), local packet reordering from 0 to 8%, and ECMP path switching within the fault window to evaluate the localization results under different asynchronous sampling, packet reordering, and forwarding state change conditions. Each test condition was run repeatedly using 30 random seeds, generating a total of 3600 independent events.

[0204] Four comparison methods were set up in the experiment, including Counter-Max based on single-point count peaks, Loss-Tomography based on end-to-end statistical relationships, Path-Vote based on path voting, and Temporal-Correlation based on physical time window correlation. Each comparison method used the same simulation topology, fault and disturbance events, telemetry data, and event windows as the method in this application to avoid the influence of different input data ranges on the comparison results.

[0205] The evaluation metrics include five dimensions: Fault location accuracy (for single-fault events, the proportion of events where the first candidate fault segment matches the actual injected fault segment; for dual-fault events, the proportion of events where the output target fault segment set contains two actual injected fault segments and no non-actual fault segments); Anomaly type discrimination accuracy (the proportion of events that correctly distinguishes between physical link reception anomalies, internal receiver port anomalies, forwarding node input / output processing anomalies, and pure congestion packet loss); Candidate ranking (recording the order of the actual fault segment in the candidate list; for dual-fault events, recording the order of the two actual fault segments separately); Location latency and computational overhead (the time from the first bit error event to the output of a stable location result, recording the algorithm execution time, processor utilization, and memory usage); and False alarm rate (the proportion of events that incorrectly classify pure congestion events without physical bit errors as physical link bit errors or receiver port bit errors).

[0206] Figure 5 The diagram shows a comparative simulation of the Top-1 fault location accuracy of the method of this application and the timing correlation method under different device clock offsets and normal message digest sampling rates.

[0207] like Figure 5 As shown, the horizontal axis represents the device clock offset, ranging from −2 s to +2 s; the vertical axis represents the normal message digest sampling rate, including 1 / 256, 1 / 512, 1 / 1024, 1 / 2048, 1 / 4096, and 1 / 8192 respectively. A larger denominator in the sampling rate indicates sparser message digest sampling; the vertical axis represents the Top-1 fault location accuracy. The solid grid lines labeled "Method of this Application" represent the experimental results of this application's method, while the dashed grid lines represent the experimental results of the timing correlation method. The default normal message digest sampling rate is 1 / 1024.

[0208] Under the default normal message digest sampling rate, when the device clock offset is 0 s, the Top-1 fault location accuracy of the method in this application is approximately 96.8%; when the device clock offset is -2 s and +2 s, the accuracy is approximately 94.9%. As the device clock offset increases, the location accuracy of the time-series correlation method decreases significantly, forming a low-accuracy region in the area where device clock offset and sparse digest sampling work together. The time-series correlation method associates the bit error increment with message missing observations based on the physical time window. When the device clock offset increases, the two types of observations may be assigned to different physical time windows. The method in this application uses the preceding anchor message and the following anchor message to construct the message progress phase, and adjusts the phase correspondence between different observation ports through a monotonic elastic mapping that maintains the message order, so that the matching of direct bit error observations and message missing observations does not directly depend on the physical clock alignment results of different devices.

[0209] When the normal message digest sampling rate decreases to 1 / 4096 or below, the experimental results corresponding to the method in this application begin to decline significantly, and approach the performance degradation boundary in the region where the normal message digest sampling rate is 1 / 8192 and the device clock offset is large. When the normal message digest sampling is too sparse, the number of normal message digests that can be jointly identified by multiple observation ports before and after the bit error event window decreases, and the phase range of some sampling intervals needs to rely more on order-preserving interpolation, which increases the phase uncertainty accordingly, making it more difficult to distinguish the predicted timing fingerprints of different candidate fault sections. The experimental results show that within the range of device clock offset and normal message digest sampling rate set in this experiment, the method in this application maintains a high fault location accuracy under most parameter combinations; when the normal message digest is so sparse that it cannot stably determine the preceding anchor message, the following anchor message, or the phase range, the fault location accuracy decreases accordingly.

[0210] Figure 6 The diagram illustrates a comparative simulation of the Pareto distribution between diagnostic computation delay and Top-1 fault location accuracy of different methods, as well as an example of the candidate fault section location residual spectrum of the proposed method.

[0211] like Figure 6As shown in the left subplot, the horizontal axis represents the diagnostic computation delay, and the vertical axis represents the Top-1 fault location accuracy. Circles, squares, triangles, and hollow diamonds represent the experimental results of the peak counting method, time-series correlation method, end-to-end statistical method, and path voting method, respectively; solid diamonds represent the experimental results of the proposed method. Dashed lines represent the Pareto optimal front formed by the experimental results of different methods. The experimental results of the proposed method are located near the Pareto optimal front, indicating that the proposed method achieves a higher fault location accuracy compared to the comparative methods, even with a certain increase in diagnostic computation delay. Based on the statistical results of 3600 independent events, the proposed method achieves a Top-1 accuracy of 96.4% for single faults, 91.7% for dual fault sets, 93.8% for anomaly type discrimination, and 94.6% for pure congestion event discrimination.

[0212] like Figure 6 As shown in the right subplot, the horizontal axis represents the candidate fault section number, and the vertical axis represents the location residual of each candidate fault section. The location residuals of adjacent candidate fault sections are connected by orthogonal broken lines. Candidate fault sections numbered 3 and 7 correspond to the real fault sections, and their location residuals are below 0.1 and are located in the target candidate low residual region; the location residuals of the remaining non-target candidate fault sections are above 0.4 and are located in the non-target candidate residual region. Under the constraint that each observation unit can be assigned to at most one candidate fault section, the proposed method jointly adjusts the candidate allocation amount, the amount of damaged messages, and the phase correspondence, so that the predicted time series fingerprint of the real fault section can simultaneously conform to the phase position, propagation direction, propagation range, and quantity conservation relationship of the actual observation, while non-target candidate fault sections that do not conform to the above relationships produce large location residuals.

[0213] The Pareto distribution on the left and the localization residual spectrum on the right both indicate that our proposed method reduces the repeated interpretation of the same bit error or missing message evidence by multiple candidate fault sections through exclusive allocation, and uses localization residuals to distinguish between true fault sections and non-target candidate fault sections. Compared to comparative methods that rely solely on single-point counting or physical time correlation for localization, our proposed method introduces joint inversion calculations, resulting in higher diagnostic computational overhead, but it maintains high fault section discrimination capability under concurrent faults and congestion interference conditions.

[0214] This application's method is primarily applicable to gray faults where the link still maintains forwarding capability but intermittently experiences short-term bit errors. When sufficient common observation normal message digests to determine the preceding and following anchor messages cannot be obtained before and after the bit error event window, or when the actual effective forwarding state at the time of the fault cannot be restored, the system does not force the output of a unique target fault profile and can supplement the judgment by combining link state detection, bidirectional forwarding detection, or segment-by-segment active testing. For key service paths where normal message digests are too sparse, the digest sampling rate can be temporarily increased or the ingress observation sequence number can be enabled after the anomaly is triggered; for end-to-end encrypted services, digest generation does not require parsing the plaintext payload. The method locates the physical link receiving boundary, the internal processing boundary of the receiving port, or the input / output boundary of the forwarding node. It cannot rely solely on frame check error counts to further determine specific device causes such as optical fiber, optical module, connector, or receiving chip. Device-level causes can be further diagnosed by combining optical power, forward error correction distribution, temperature, or device replacement results.

[0215] For scenarios where the receiving end identifies and prevents the normal subsequent forwarding of failed messages, this application constructs a dual-anchor message progress phase using common observations of normal message summaries before and after the error event window. This transforms asynchronously acquired error increments and missing upstream and downstream message observations to a unified phase reference. Furthermore, candidate fault profile observation templates are constructed based on the actual directed forwarding path and message replication relationships. Under the constraints of monotonic elastic mapping that maintains message order and exclusive allocation of observations, the system jointly inverts the amount of damaged messages in each candidate fault profile and determines the location residual and the target fault profile set based on phase position, propagation direction, propagation range, and quantity conservation relationships. Thus, the dispersed port error counts and normal message summaries are organized into event-level evidence with propagation causality, reducing the repeated interpretation of the same measured observation by multiple candidate fault profiles and enabling the location of intermittent error faults without requiring full packet-by-packet tracing or black-box classification models.

[0216] It should be noted that, for the sake of simplicity, the foregoing method embodiments are all described as a series of combined actions. However, those skilled in the art should understand that this application is not limited to the described order of actions, as some steps may be performed in other orders or simultaneously according to this application. Secondly, those skilled in the art should also understand that the embodiments described in the specification are preferred embodiments, and the actions and modules involved are not necessarily essential to this application. In the above embodiments, the descriptions of each embodiment have their own emphasis; for parts not described in detail in a certain embodiment, please refer to the relevant descriptions of other embodiments.

[0217] Figure 7 A structural block diagram of an example of a packet link fault location system based on bit error timing fingerprinting according to an embodiment of this application is shown.

[0218] like Figure 7 As shown, the packet link fault location system 700 based on bit error timing fingerprinting includes a bit error event acquisition module 710, a phase reference construction module 720, a measured fingerprint generation module 730, an observation template construction module 740, a damaged message inversion module 750, and a fault section location module 760.

[0219] The bit error event acquisition module 710 is used to acquire the cumulative direct bit error count and normal message digest of multiple observation ports, determine the bit error increment based on the cumulative direct bit error count, determine the bit error event window based on the bit error increment, and acquire the directed forwarding path and message replication relationship.

[0220] The phase reference construction module 720 is used to determine the preceding anchor message and the following anchor message jointly observed by multiple observation ports on the directed forwarding path from the normal message digests before and after the bit error event window, respectively, and to construct the message progress phase and the phase range of each sampling interval according to the message sequence of each observation port between the preceding anchor message and the following anchor message.

[0221] The measured fingerprint generation module 730 is used to convert the bit error increment into a measured direct bit error distribution according to each phase range, and compare the normal message digests of the observation ports with upstream and downstream relationships in the directed forwarding path to obtain the measured message missing quantity distribution that exists upstream but is missing downstream, so as to generate a measured bit error timing fingerprint.

[0222] The observation template construction module 740 is used to construct observation templates for multiple candidate fault sections based on the directed forwarding path and the packet replication relationship. Each observation template maps the unit damaged packet quantity of the corresponding candidate fault section to the expected direct bit error rate distribution and the expected packet missing rate distribution of each observation port.

[0223] The damaged message inversion module 750 is used to perform monotonically elastic mapping on the message progress phase to maintain the message order. Under the constraint that each measured direct bit error or measured message missing amount is assigned to at most one of the candidate fault sections, the damaged message amount of each candidate fault section is jointly inverted based on the measured bit error time sequence fingerprint and each of the observation templates, and predicted time sequence fingerprints are generated respectively.

[0224] The fault section localization module 760 is used to determine the localization residual of each candidate fault section based on the consistency and quantity conservation relationship between the phase position, propagation direction and propagation range of each predicted timing fingerprint and the measured bit error timing fingerprint, and to determine the target fault section set based on each localization residual.

[0225] In some embodiments, this application provides a non-volatile computer-readable storage medium storing one or more programs including execution instructions. The execution instructions can be read and executed by an electronic device (including but not limited to a computer, server, or network device) to perform the steps of any of the above-described packet link fault location methods based on bit error timing fingerprints.

[0226] In some embodiments, this application also provides a computer program product, the computer program product including a computer program stored on a non-volatile computer-readable storage medium, the computer program including program instructions, which, when executed by a computer, cause the computer to perform the steps of any of the above-described packet link fault location methods based on bit error timing fingerprints.

[0227] In some embodiments, this application also provides an electronic device, comprising: at least one processor, and a memory communicatively connected to the at least one processor, wherein the memory stores instructions executable by the at least one processor, the instructions being executed by the at least one processor to enable the at least one processor to perform the steps of a packet link fault location method based on bit error timing fingerprints.

[0228] The above-described product can perform the methods provided in the embodiments of this application, and has the corresponding functional modules and beneficial effects for performing the methods. Technical details not described in detail in this embodiment can be found in the methods provided in the embodiments of this application.

[0229] The electronic devices in this application can exist in various forms, including but not limited to: mobile communication devices, ultra-mobile personal computer devices, portable entertainment devices, or other airborne electronic devices with data interaction functions.

[0230] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs.

[0231] Through the above description of the embodiments, those skilled in the art can clearly understand that each embodiment can be implemented using software plus a general-purpose hardware platform, or of course, using hardware. Based on this understanding, the above technical solutions, in essence or the parts that contribute to the related technology, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods described in the various embodiments or some parts of the embodiments.

[0232] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit them. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of this application.

Claims

1. A method for locating packet link faults based on bit error timing fingerprints, characterized in that, The method includes: Obtain the cumulative direct bit error count and normal message digest of multiple observation ports, determine the bit error increment based on the cumulative direct bit error count, determine the bit error event window based on the bit error increment, and obtain the directed forwarding path and message replication relationship; The preceding anchor message and the following anchor message observed by multiple observation ports on the directed forwarding path are determined from the normal message digests before and after the bit error event window, respectively. Based on the message sequence of each observation port between the preceding anchor message and the following anchor message, the message progress phase and the phase range of each sampling interval are constructed. The bit error increment is converted into a measured direct bit error distribution according to each phase range, and the normal message digests of the observation ports with upstream and downstream relationships in the directed forwarding path are compared to obtain the measured message missing quantity distribution that exists upstream but is missing downstream, so as to generate a measured bit error timing fingerprint. Based on the directed forwarding path, the receiving boundary of the directed physical link, the internal processing boundary of the receiving port, and the input / output boundary of the forwarding node are respectively determined as candidate fault sections; the cumulative message replication coefficient is determined based on the message replication relationship, and the actual subsequent forwarding range of each candidate fault section is determined based on the directed reachability relationship from each candidate fault section to each observation port, the observation type, the phase propagation relationship, and the cumulative message replication coefficient; and an observation mapping relationship is established between the unit damaged message quantity of the candidate fault section and the expected direct bit error rate distribution and the expected message missing rate distribution of each observation port to form the observation template of each candidate fault section. The message progress phase is subjected to monotonically elastic mapping that preserves the message order. Under the constraint that each measured direct bit error or measured message missing amount is assigned to at most one of the candidate fault sections, the amount of damaged messages of each candidate fault section is jointly inverted based on the measured bit error time series fingerprint and each of the observation templates, and predicted time series fingerprints are generated respectively. Based on the consistency and quantity conservation relationship between the phase position, propagation direction and propagation range of each predicted timing fingerprint and the measured bit error timing fingerprint, the location residual of each candidate fault section is determined, and the target fault section set is determined based on each location residual.

2. The method according to claim 1, characterized in that, The step of determining the error increment based on the cumulative direct error count, determining the error event window based on the error increment, and obtaining the directed forwarding path and packet replication relationship includes: The acquisition timestamp and counter modulus of the cumulative direct bit error count for each observation port are obtained, and the sampling interval is divided accordingly. The cumulative direct bit error count includes at least one of uncorrectable codeword count and frame check error count. Each observation port generates the normal message digest using a consistent digest generation rule and a reproducible sampling rule based on message content. The initial increment is determined based on the difference between the cumulative direct bit error counts of adjacent sampling intervals. When the cumulative direct bit error count changes in the opposite direction, counter wrap-around compensation is performed. Counter wrap-around and counter reset are distinguished based on whether the compensated increment exceeds the preset maximum count growth amount of the corresponding sampling interval. Sampling intervals where the count change cannot be recovered are determined as invalid sampling intervals. The remaining increments are converted into the equivalent of damaged messages. Based on the hierarchical response relationship of different types of bit error counts, the equivalent of damaged messages that repeatedly represent the same damaged message is counted only once to obtain the bit error increment. The background bit error limit is determined based on the bit error increment distribution during normal operation. The effective sampling interval that continuously exceeds the background bit error limit is determined as the bit error burst interval. The error event window is then extended forward to the stable normal sampling interval and backward to the sampling interval where the bit error increment recovers and the normal message digest stabilizes again. Obtain the effective forwarding table version, equivalent multipath and tunnel forwarding results and packet replication status within the bit error event window. When a forwarding table version change, link retraining or packet sequence reset occurs, divide the sub-event window according to the corresponding change time. Otherwise, treat the bit error event window as a single sub-event window and determine the directed forwarding path and packet replication relationship corresponding to each sub-event window. Normal message digests with the same directed forwarding path and message replication branch sequence within each sub-event window and its preceding and following stable sampling intervals are divided into co-path message groups.

3. The method according to claim 2, characterized in that, The step of determining the preceding and following anchor messages jointly observed by multiple observation ports on the directed forwarding path from the normal message digests before and after the bit error event window, and constructing the message progress phase and the phase range of each sampling interval based on the message sequence position of each observation port between the preceding and following anchor messages, includes: For any co-path message group within a sub-event window, the common observation normal message digests that belong to the co-path message group and are closest to the sub-event window before and after the sub-event window are respectively determined as the preceding anchor message and the following anchor message. The message progress phase of the preceding anchor message is set as the phase start point, and the message progress phase of the following anchor message is set as the phase end point. The message progress phase of the normal message digest is determined according to the proportion of the message sequence offset of any normal message digest relative to the preceding anchor message in the message sequence interval between the two anchor messages. Based on the order of common observation normal message digests at each observation port, determine the largest consistent subsequence containing the most messages and maintaining a common order at each observation port. Incorporate non-common observation normal message digests that do not change the order of the largest consistent subsequence into the valid message sequence, and identify normal message digests that cause order conflicts as reordered message digests. The phase range of the sampling interval is determined based on the minimum and maximum message progress phases corresponding to the valid message sequence within any sampling interval. When there is no valid normal message digest within the sampling interval, order-preserving interpolation is performed based on the phase endpoints of adjacent valid sampling intervals and the number of messages in the same path. The phase uncertainty is determined based on the number of consecutive missing normal message digests, so that the phase uncertainty increases with the number of consecutive missing messages.

4. The method according to claim 3, characterized in that, The step of converting the bit error increment into a measured direct bit error rate distribution based on each phase range, and comparing the normal message digests of observation ports with upstream and downstream relationships in the directed forwarding path to obtain the measured message missing rate distribution where the upstream is present but the downstream is missing, in order to generate a measured bit error rate timing fingerprint, includes: The message progress phase is divided into continuous and non-overlapping phase intervals. The distribution ratio of the bit error increment in each phase interval is determined according to the phase range of the sampling interval and the degree of overlap between each phase interval. The distribution ratios together represent the total bit error increment in the sampling interval. Combined with the proportion of the number of messages in each co-channel message group in the sampling interval, the measured direct bit error of each observation port in each phase interval is determined. Within each phase interval, normal message summaries of upstream and downstream observation ports are matched according to the summary identifier and message sequence. The number of normal message summaries of upstream and downstream is corrected according to the sampling probability of the corresponding summary. The number of normal message summaries of downstream is reverse-normalized to the number of messages before replication according to the message replication relationship. The summary differences corresponding to strategy processing and reordering are excluded. The number of messages corresponding to normal message summaries that exist upstream but not downstream is determined as the measured message missing amount. The measured direct bit error rate and the measured message missing rate are arranged according to the observation type, observation location, co-path message group and phase interval to generate the measured bit error time sequence fingerprint composed of multiple observation units; For co-path message groups with reliable transmission semantics, retransmission and reproduction messages that appear after a message is missing and have the same transmission sequence number and payload summary as the missing message are identified. The retransmission feasible interval is determined based on the historical round-trip delay and retransmission timeout range, and the corresponding message missing, retransmission time interval and the retransmission feasible interval are associated as retransmission and reproduction information.

5. The method according to claim 4, characterized in that, The generation of predicted temporal fingerprints includes: Using the observation template, the amount of non-negative damaged messages in different co-channel message groups and phase intervals of each candidate fault section is mapped to a predicted timing fingerprint arranged in the same way as the measured bit error timing fingerprint. For the receiving boundary of a directed physical link, the corresponding observation template is configured to generate the expected direct bit error rate on the receiving side and the expected packet loss rate within the actual subsequent forwarding range. For the internal processing boundary of the receiving port, the corresponding observation template is configured to generate the expected direct bit error rate based on the hierarchical response relationship between the uncorrectable codeword count and the frame check error count, and the expected packet loss rate within the actual subsequent forwarding range. For the input-output boundary of a forwarding node, the corresponding observation template is configured not to generate the expected direct bit error rate but to generate the expected packet loss rate within the corresponding output path and its actual subsequent forwarding range. For multiple downstream loss results caused by packet replication, reverse normalization is performed according to the cumulative packet replication coefficient.

6. The method according to claim 5, characterized in that, The determination of the localization residual for each candidate fault profile based on the consistency and quantity conservation relationship between the phase position, propagation direction, and propagation range of each predicted time-series fingerprint and the measured bit error time-series fingerprint includes: An elastic mapping relationship is constructed for each observation port, which keeps the endpoints unchanged, maintains the message order, and the local scaling degree is within a preset range. The degree of phase deformation is determined based on the degree of deviation of the elastic mapping relationship from the state without phase deformation. The reliability weight of each observation unit is determined based on the phase uncertainty and the summary sampling probability, and the reliability weight decreases as the phase uncertainty increases or the summary sampling probability decreases. The actual measurement assigned to any candidate fault section is determined as the candidate allocation quantity. The difference between the candidate allocation quantity and the corresponding predicted quantity after elastic mapping is converted into a relative deviation relative to the corresponding actual measurement. The phase position matching residual is determined based on the reliability weight. When the observation template of a candidate fault segment includes both the expected direct bit error rate and the expected message missing rate, the total measured direct bit error rate allocated to the candidate fault segment is compared with the total measured message missing rate after replication and reverse normalization, and the quantity conservation residual is determined based on the ratio of the absolute difference between the two to the sum of the two; when the corresponding observation template only includes the expected message missing rate, the quantity conservation residual is not set. The causal penalty is determined based on whether there is a propagation direction conflict, a propagation range mismatch, or a retransmission time interval outside the corresponding retransmission feasible interval among the candidate allocation quantities. The phase position matching residual, the quantity conservation residual, and the causal penalty are combined into the positioning residual. When there is no retransmission reproduction information, no causal penalty is set for the retransmission time.

7. The method according to claim 6, characterized in that, The monotonically elastic mapping of the message progress phase to preserve message order, under the constraint that each measured direct bit error or measured message missing amount is assigned to at most one of the candidate fault sections, involves jointly inverting the damaged message amount of each candidate fault section based on the measured bit error time series fingerprint and each of the observation templates, and generating predicted time series fingerprints respectively, including: Set a candidate fault section assignment flag for each observation unit and set an enabled state for each candidate fault section, so that any observation unit has at most one valid assignment flag, the actual measurement assigned to the candidate fault section does not exceed the actual measurement of the corresponding observation unit, and the actual measurement can only be assigned to the candidate fault section in the enabled state. The amount of damaged packets in each candidate fault segment is constrained to be non-negative and does not exceed the preset upper limit of the amount of damaged packets. The amount of damaged packets in the candidate fault segment that is not enabled is constrained to be zero, and the sum of the allocated actual measurements is not less than the preset coverage ratio of all actual measurements. Based on the location residuals of each activated candidate fault section, the phase deformation degree of the elastic mapping relationship, the proportion of actual measurements not explained by the candidate fault section relative to the corresponding actual measurements, and the number of activated candidate fault sections, a competitive allocation evaluation relationship is constructed. The allocation results of the observation units, the activation status of the candidate fault sections, the amount of damaged messages, and the elastic mapping relationship are jointly adjusted to reduce the comprehensive deviation represented by the competitive allocation evaluation relationship, and the predicted time series fingerprint of each candidate fault section is generated based on the adjusted amount of damaged messages and the corresponding observation template. The step of determining the target fault profile set based on each of the positioning residuals includes: Candidate fault sections that are in the enabled state and whose total number of damaged messages is greater than the preset lower limit of damage are identified as initial target fault sections. Each initial target fault section is then deactivated in turn and a joint adjustment is performed. The independent interpretation contribution of each initial target fault section to the measured bit error timing fingerprint is determined based on the comprehensive deviation increment before and after deactivation. The initial target fault sections with positioning residuals less than the preset upper limit of residuals and independent interpretation contributions greater than the preset lower limit of contributions are formed into the target fault section set. When the difference in the location residuals of at least two target fault sections is less than a preset ambiguity threshold, an acknowledgment detection path is generated based on the difference segment between the actual subsequent forwarding ranges of the corresponding target fault sections. The detection results are converted into supplementary observations. The measured bit error timing fingerprint is updated based on the supplementary observations. The corresponding observation template is expanded based on the acknowledgment detection path, the observation position of the supplementary observations, and the message replication relationship. The joint adjustment is then performed to update the target fault section set.

8. A packet link fault location system based on bit error timing fingerprinting, characterized in that, The system includes: The bit error event acquisition module is used to acquire the cumulative direct bit error count and normal message digest of multiple observation ports, determine the bit error increment based on the cumulative direct bit error count, determine the bit error event window based on the bit error increment, and acquire the directed forwarding path and message replication relationship. The phase reference construction module is used to determine the front anchor message and the back anchor message jointly observed by multiple observation ports on the directed forwarding path from the normal message digests before and after the bit error event window, respectively, and to construct the message progress phase and the phase range of each sampling interval according to the message sequence of each observation port between the front anchor message and the back anchor message. The measured fingerprint generation module is used to convert the bit error increment into a measured direct bit error distribution according to each phase range, and compare the normal message digests of the observation ports with upstream and downstream relationships in the directed forwarding path to obtain the measured message missing quantity distribution that exists upstream but is missing downstream, so as to generate a measured bit error timing fingerprint. The observation template construction module is used to determine the receiving boundary, the internal processing boundary of the receiving port, and the input / output boundary of the forwarding node as candidate fault sections according to the directed forwarding path; determine the cumulative message replication coefficient according to the message replication relationship; and determine the actual subsequent forwarding range of each candidate fault section according to the directed reachability relationship from each candidate fault section to each observation port, the observation type, the phase propagation relationship, and the cumulative message replication coefficient; and establish the observation mapping relationship between the unit damaged message quantity of the candidate fault section and the expected direct bit error rate distribution and the expected message missing rate distribution of each observation port to form the observation template of each candidate fault section. The damaged message inversion module is used to perform monotonically elastic mapping on the message progress phase to maintain the message order. Under the constraint that each measured direct bit error or measured message missing amount is assigned to at most one of the candidate fault sections, the module jointly inverts the damaged message amount of each candidate fault section based on the measured bit error time series fingerprint and each of the observation templates and generates predicted time series fingerprints respectively. The fault section localization module is used to determine the localization residual of each candidate fault section based on the consistency and quantity conservation relationship between the phase position, propagation direction and propagation range of each predicted timing fingerprint and the measured bit error timing fingerprint, and to determine the target fault section set based on each localization residual.

9. An electronic device, comprising: At least one processor; as well as A memory communicatively connected to the at least one processor, characterized in that the memory stores instructions executable by the at least one processor, the instructions being executed by the at least one processor to enable the at least one processor to perform the steps of the method as described in any one of claims 1-7.

10. A storage medium having a computer program stored thereon, characterized in that, When the program is executed by the processor, it implements the steps of the method as described in any one of claims 1-7.

Images