Method, apparatus, device, medium and program product for quality assessment of a communication link

Abstract

The application discloses a quality evaluation method, device, equipment, medium and program product of a communication link. The method comprises the following steps: acquiring at least two bit error rates of the communication link; determining a bit error rate statistical value according to the at least two bit error rates; and performing quality evaluation on the communication link according to the bit error rate statistical value to obtain a quality evaluation result of the communication link. The evaluation process utilizes the characteristic that the bit error rate is logarithmically normally distributed, adopts the bit error rate statistical value containing the logarithmic statistical value to perform quality evaluation, can more accurately restore the actual distribution law of the bit error rate, can more truly represent the transmission quality of the communication link (especially a high-speed link) in an actual production environment, improves the accuracy of quality evaluation, and guarantees the operation stability and data processing efficiency of a high-performance computing system.

Description

Technical Field

[0001] This application relates to the field of computer technology, and in particular to a method, apparatus, device, medium, and program product for evaluating the quality of a communication link. Background Technology

[0002] High-speed links are communication links used to enable high-speed data transmission within or between electronic devices, such as Serializer (Ser) / Deserializer (Des) channels. High-speed links are an indispensable core component of high-performance computing systems, and their transmission quality determines the operational stability and data processing efficiency of the entire system.

[0003] In related technologies, the quality assessment method for high-speed links is usually a single-point test method. Specifically, during the operation of the high-speed link, the bit error rate is collected once at a preset time interval. Multiple bit error rates are collected, and the arithmetic mean of the multiple bit error rates is calculated as the quality assessment result.

[0004] However, the aforementioned single-point testing method is relatively simplistic, leading to a significant discrepancy between the quality assessment results and the actual operating status of the link, resulting in a low accuracy rate in quality assessment. Summary of the Invention

[0005] This application provides a method, apparatus, device, medium, and program product for evaluating the quality of a communication link, the technical solution of which includes the following aspects.

[0006] On the one hand, a method for assessing the quality of a communication link is provided, the method comprising: Obtain at least two bit error rates for the communication link; A bit error rate statistic is determined based on the at least two bit error rates, wherein the bit error rate statistic includes a logarithmic statistic, which is a statistic determined based on the logarithm corresponding to each of the at least two bit error rates; Based on the bit error rate statistics, a quality assessment is performed on the communication link to obtain the quality assessment result of the communication link.

[0007] On the other hand, a communication link quality assessment device is provided, the device comprising: The acquisition module is used to acquire at least two bit error rates of the communication link; The processing module is configured to determine a bit error rate statistical value based on the at least two bit error rates, wherein the bit error rate statistical value includes a logarithmic statistical value, which is a statistical value determined based on the logarithms corresponding to the at least two bit error rates respectively; The processing module is used to perform a quality assessment on the communication link based on the bit error rate statistics, and obtain the quality assessment result of the communication link.

[0008] On the other hand, a computer device is provided, the computer device including a processor and a memory, the memory storing at least one program, the at least one program being loaded and executed by the processor to implement any of the communication link quality assessment methods described above.

[0009] On the other hand, a computer-readable storage medium is provided, wherein at least one program is stored therein, the at least one program being loaded and executed by a processor to implement any of the communication link quality assessment methods described above.

[0010] On the other hand, a computer program product is provided, comprising a computer program stored in a computer-readable storage medium. A processor of a computer device reads the computer program from the computer-readable storage medium and executes the computer program, causing the computer device to perform any of the communication link quality assessment methods described above.

[0011] The beneficial effects of the technical solutions provided in this application include at least the following: By acquiring at least two bit error rates (BERs), calculating BER statistics including logarithmic values, and performing a quality assessment on the communication link based on these BER statistics, a quality assessment result is obtained. Compared to the single-point testing method in related technologies that simply collects multiple BERs and calculates their arithmetic mean, this application's assessment process utilizes the log-normal distribution characteristic of BERs and uses BER statistics including logarithmic values ​​for quality assessment. This can more accurately reproduce the actual distribution pattern of BERs, thereby more realistically characterizing the transmission quality of communication links (especially high-speed links) in actual production environments, improving the accuracy of quality assessment, and ensuring the operational stability and data processing efficiency of high-performance computing systems. Attached Figure Description

[0012] To more clearly illustrate the technical solutions in the embodiments of this application, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0013] Figure 1 This is a schematic diagram of a computer system provided in an exemplary embodiment of this application; Figure 2 This is a flowchart of a communication link quality assessment method provided in an exemplary embodiment of this application; Figure 3 This is a flowchart of a communication link quality assessment method provided in another exemplary embodiment of this application; Figure 4 This is a flowchart of a communication link quality assessment method provided in yet another exemplary embodiment of this application; Figure 5 This is a schematic diagram of a system architecture provided in an exemplary embodiment of this application; Figure 6 This is a schematic diagram of the bit error rate performance curve provided in an exemplary embodiment of this application; Figure 7 This is a flowchart of a multidimensional health scoring algorithm provided in an exemplary embodiment of this application; Figure 8 This is a flowchart illustrating the calculation of continuous anomaly penalty scores provided in an exemplary embodiment of this application; Figure 9 This is a structural block diagram of a communication link quality assessment device provided in an exemplary embodiment of this application; Figure 10 This is a structural block diagram of a computer device provided in an exemplary embodiment of this application. Detailed Implementation

[0014] To make the objectives, technical solutions, and advantages of this application clearer, the embodiments of this application will be described in further detail below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments in this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0015] In this application, the terms "first" and "second" are used to distinguish between identical or similar items that have essentially the same function. It should be understood that there is no logical or temporal dependency between "first" and "second", nor is there any limitation on the quantity or execution order.

[0016] Figure 1 A schematic diagram of a computer system provided in an exemplary embodiment of this application is shown. This computer system can implement a system architecture for a method of evaluating the quality of a communication link. The computer system includes: a terminal 110 and a server 120.

[0017] In some embodiments, terminal 110 may be a smartphone, tablet computer, laptop computer, desktop computer, smart home appliance, smart vehicle terminal, AR (Augmented Reality) device, action camera, etc., but is not limited to these. Terminal 110 has an application client installed, which has a communication link quality assessment function. This application may be a traditional application, a cloud application, a mini-program or application module within a host application, or a web platform; this embodiment does not limit its scope.

[0018] Terminal 110 is connected to server 120 via a wireless network or a wired network.

[0019] In some embodiments, server 120 provides background services for clients of applications installed on the terminal. It is worth noting that server 120 can be a standalone physical server, a server cluster or distributed system consisting of at least two physical servers, a cloud server providing basic cloud computing services, or a node in a blockchain system.

[0020] The communication link quality assessment method provided in this application embodiment can be executed by a computer device, which refers to an electronic device with data computing, processing, and storage capabilities. Figure 1 Taking the computer system shown as an example, the communication link quality assessment method can be executed by terminal 110 (e.g., the client of the application installed and running in terminal 110 executes the communication link quality assessment method), or it can be executed by server 120, or it can be executed by interaction between terminal 110 and server 120. This application does not limit this.

[0021] The following example illustrates a method for assessing the quality of a communication link, which is performed by the interaction and cooperation between terminal 110 and server 120.

[0022] like Figure 1 As shown, terminal 110 acquires bit error rate data, which includes at least two bit error rates of the communication link (i.e., Figure 1 The bit error rates (BERs) in the data are 1, ..., N, where N is an integer greater than 1. At least two BERs conform to the log-normal distribution requirement, such that the logarithms corresponding to BERs 1, ..., N respectively follow or highly approximate a normal distribution. Optionally, the terminal 110 can collect BER data from an active communication link (online collection) or directly receive BER data (offline analysis), and this application does not limit this.

[0023] Terminal 110 sends the bit error rate (BER) data to server 120. Upon receiving the BER data, server 120 determines a BER statistical value based on at least two BER rates, including at least one logarithmic statistical value and a linear statistical value. Then, server 120 performs a quality assessment of the communication link based on the BER statistical value, obtaining the communication link quality assessment result. Finally, server 120 feeds back the quality assessment result to terminal 110. Optionally, terminal 110 can visualize the quality assessment result so that technicians can view it on the terminal.

[0024] Next, the process of the communication link quality assessment method provided in the embodiments of this application will be introduced.

[0025] Figure 2 This is a flowchart illustrating a communication link quality assessment method provided in an embodiment of this application. The method is executed by a computer device, which may be... Figure 1 The terminal 110 and / or server 120 are shown. The method includes steps 210 to 230.

[0026] Step 210: Obtain at least two bit error rates for the communication link.

[0027] A communication link is a channel used for data transmission. Optionally, the communication link includes, but is not limited to, a high-speed serial link (or simply a high-speed link), a board-level interconnect link, an inter-chip interconnect link, etc. For example, the communication link is implemented as a high-speed serial computer expansion bus (Peripheral Component Interconnect Express, PCIe) link; or, the communication link is implemented as a serializer / deserializer (SerDes) channel, which is not limited in this application embodiment.

[0028] Optionally, the communication link includes at least one transmitter and at least one receiver. The transmitter converts parallel data into a serial data stream and transmits it through a physical medium, while the receiver recovers the received serial data stream into parallel data. Due to various factors such as crosstalk, inter-symbol interference, random jitter, and power supply noise in the link, bit errors will occur during transmission, which are quantified as bit error rate (BER).

[0029] At least two bit error rates refer to a set of bit error rate samples obtained through at least two independent tests or samplings under the same link configuration parameters or different link configuration parameters. This sample set is used for subsequent logarithmic statistical analysis to assess the overall health of the link.

[0030] Specifically, at least two bit error rates must conform to a log-normal distribution. Optionally, conforming to a log-normal distribution means that the obtained bit error rate sample set as a whole follows or approximately follows a log-normal distribution. A log-normal distribution is a probability distribution in which the natural logarithm of a random variable follows a normal distribution.

[0031] The following explains the process of discovering that at least two bit error rates meet the requirement of a log-normal distribution.

[0032] In traditional high-speed link quality assessment, the industry generally uses the arithmetic mean bit error rate (BER) obtained from a single test or multiple tests. However, the inventors discovered in engineering practice that the BER does not accurately reflect the typical transmission performance of a high-speed link. To address this issue, the inventors conducted repeated tests on high-speed links (such as SerDes channels) with different rates, coding modes, and hardware platforms, collecting a large number of BER samples. Through statistical methods, they verified that the natural logarithm sequence of the BER of high-speed links exhibits a highly symmetrical normal distribution.

[0033] Furthermore, by tracing back and deducing from the fundamental theories of digital communication, the inventors discovered that the signal-to-noise ratio (SNR) at the receiver of a high-speed link is measured in decibels. Affected by multiplicative factors such as thermal noise and random jitter, its statistical fluctuations conform to a normal distribution; while there is an exponential relationship between the bit error rate and the linear signal-to-noise ratio. Combining the two, it can be mathematically deduced that the natural logarithm of the bit error rate, ln(BER), approximately follows a normal distribution, that is, the bit error rate itself follows or is highly approximately log-normal (the specific proof process can be found in the core mathematical derivation section below, which will not be repeated here).

[0034] Based on this core discovery, the inventors abandoned the traditional arithmetic mean statistical approach and innovatively proposed using the geometric mean bit error rate to characterize the typical performance level of the link, and using the logarithmic standard deviation to quantify the relative fluctuation of the bit error rate. Based on this, they constructed the communication link evaluation method proposed in this application, which solves the problem of insufficient accuracy of traditional evaluation methods.

[0035] In some embodiments, the number of bit error rates required for quality assessment varies depending on the quality assessment scenario. Optionally, a recommended sample size is determined based on the quality assessment scenario. This recommended sample size indicates the recommended number of bit error rates to obtain; at least two bit error rates are then obtained based on this recommended sample size.

[0036] For example, assuming a recommended sample size of 100 to 200, 150 bit error rates can be obtained as at least two bit error rates.

[0037] The quality assessment scenarios include at least one of the following: rapid screening scenario, initial stability assessment scenario, production yield analysis scenario, and fault diagnosis scenario.

[0038] (a) Quickly filter scenarios.

[0039] Rapid screening typically occurs during the initial configuration parameter screening stage of the link design. At this point, engineers need to quickly eliminate configurations with significantly poor performance from a large number of candidate parameter combinations (such as multiple sets of TX and FIR coefficients), retaining a few promising candidate configurations for the next round of refined testing. The core requirement for this scenario is prioritizing testing speed, with statistical accuracy being a secondary consideration. For example, a sample size of 50 is recommended for rapid screening to balance testing time and statistical accuracy, suitable for initial screening.

[0040] (ii) Initialization stability assessment scenario.

[0041] Initialization stability assessment targets the performance repeatability verification of a link during multiple initializations under a fixed configuration. A typical application is SerDes link initialization stability assessment. The core metrics for this scenario are the connectivity success rate during initialization and the consistency (logarithmic standard deviation) of the bit error rate after initialization stabilization. Sufficient sample size is needed to stably estimate the logarithmic standard deviation and connectivity success rate. For example, a sample size of 50-100 is recommended for SerDes initialization stability assessment to ensure statistical accuracy and suitability for stability evaluation.

[0042] (III) Production yield analysis scenario.

[0043] Production yield analysis occurs during large-scale production testing, requiring quality assessment of each batch's sampling or full-inspection links to determine the overall production yield level. The core constraint of this scenario is that the testing time for a single link directly impacts production line throughput and testing costs, while the sample size must be sufficient to support batch-level statistical inference. For example, a sample size of 100 is recommended for production yield analysis to balance testing time and cost.

[0044] (iv) Fault diagnosis scenario.

[0045] Fault diagnosis scenarios target links where performance anomalies have been detected or intermittent faults are suspected. A larger sample size is needed to capture and locate abnormal patterns, especially continuous abnormal sequences and occasional extreme bit error spikes. Insufficient sample size may result in intermittent faults not being triggered or continuous anomalies being truncated, leading to missed detections. For example, a sample size of 100-200 is recommended for fault diagnosis scenarios to ensure accurate anomaly detection.

[0046] In the above embodiments, by subdividing the quality assessment scenario into rapid screening, initial stability assessment, production yield analysis and fault diagnosis, and providing differentiated sample size recommendations for each scenario, the different balance needs between testing time cost and statistical confidence in different engineering scenarios are addressed. Under the premise of ensuring the scientific validity of the assessment results, the overall testing efficiency is improved and the testing cost is reduced.

[0047] In some embodiments, the method for obtaining at least two bit error rates of a communication link includes at least one of the following.

[0048] Method 1: Perform at least two initialization operations on the communication link using the same link configuration parameters to obtain at least two bit error rates.

[0049] In other words, by keeping the link configuration parameters unchanged, and repeatedly initializing the link and collecting the bit error rate after each initialization, a set of samples reflecting the random fluctuations in link performance under this fixed configuration is obtained. For example, 50 initialization operations are performed on the communication link using the same link configuration parameters, resulting in 50 bit error rates.

[0050] In an optional embodiment, the step of obtaining at least two bit error rates by method one further includes steps 1 to 3 as follows.

[0051] Step 1 Under the same link configuration parameters, control the communication link to perform at least two initialization operations.

[0052] The link configuration parameters include at least the following: transmitter finite impulse response filter coefficients, link rate, and coding mode. Optionally, the link configuration parameters may also include receiver equalizer coefficients, reference clock frequency, and output drive amplitude. No limitations are specified here.

[0053] For example, controlling the communication link to perform initialization operations includes: writing specified Transmit Finite Impulse Response (TX FIR) coefficients to the link transmitter; configuring the link to operate at a preset link rate (e.g., 106.25 Gbps / 112 Gbps) and coding mode (e.g., NRZ or PAM4); triggering a link retraining or reset operation to enable the transmitter and receiver to re-handshake, equalize, and restore the clock; and considering an initialization operation as complete once the link status register indicates that initialization is complete and is locked.

[0054] Optionally, the number N (N≥2) of at least two initialization operations can be preset according to the statistical confidence level required for the evaluation. For example, in high-speed link quality evaluation scenarios, N is typically 30 to 100 times. The larger N is, the closer the obtained bit error rate sample is to the true bit error rate distribution, but the test time increases accordingly; when N is smaller, the evaluation efficiency is improved, but the statistical representativeness may decrease. This application does not limit the specific value of N, and it can be dynamically adjusted according to the test time budget and accuracy requirements in practical applications.

[0055] Step 2 After the i-th initialization operation is completed, obtain the bit error rate of the communication link corresponding to the i-th initialization operation.

[0056] i is a positive integer.

[0057] Optionally, the bit error rate (BER) can be obtained by reading the BER counter maintained internally at the link receiver. For example, after link initialization, the BER counter and the total transmitted bits counter at the link receiver begin to accumulate. By reading the current values ​​of these two counters, the ratio of the number of erroneous bits to the total number of transmitted bits is calculated, thus obtaining the BER corresponding to this initialization. In some high-speed links (such as SerDes channels), the BER counter accumulates periodically per second or per test window. This step can involve reading the instantaneous value of the counter all at once within a stable state window after initialization, or calculating the BER within a smaller time window.

[0058] Optionally, to avoid reading inaccurate bit error rate due to the link not yet fully reaching a steady state, step 2 further includes: after the initialization operation is completed, delaying for a preset stabilization waiting time (e.g., 100 milliseconds) before performing the bit error rate acquisition operation. This stabilization waiting time is used to ensure that the clock data recovery circuit, equalizer adaptive algorithm, etc. at the link receiver have converged to a stable operating point.

[0059] For example, for the i-th initialization operation (i=1,2,...,N), the obtained bit error rate is denoted as BER(i).

[0060] Step 3 The bit error rate of the communication link corresponding to at least two initialization operations is taken as at least two bit error rates.

[0061] After completing N initializations and corresponding bit error rate acquisitions, the obtained effective bit error rates BER(1), BER(2), ..., BER(N) are arranged into an array as the raw data for subsequent processing.

[0062] Method 1 described above controls the communication link to perform at least two initialization operations under the same link configuration parameters, and obtains the bit error rate corresponding to each initialization operation as an evaluation sample. This repeated initialization process can capture the random fluctuation characteristics of the link bit error rate under fixed configuration conditions, thereby improving the statistical representativeness of the obtained bit error rate samples. This provides a reliable data foundation for subsequent quality assessment based on log-normal distribution and improves the accuracy of communication link quality assessment results.

[0063] Method 2: Perform initialization operations on the communication link using at least two link configuration parameters to obtain at least two bit error rates.

[0064] In other words, by changing the link's configuration parameters (such as the equalization coefficient, rate, and encoding method at the transmitter), the bit error rate samples of the link under different operating states are obtained. Unlike the repeated initialization under a fixed configuration (Method 1), which focuses on capturing random fluctuations, Method 2 focuses on capturing the systematic performance differences of the link caused by configuration changes.

[0065] In an optional embodiment, the step of obtaining at least two bit error rates by method two further includes steps 1 to 3 as follows.

[0066] Step 1 Under at least two link configuration parameters, control the communication link to perform initialization operations respectively.

[0067] The link configuration parameters include at least the following: transmitter finite impulse response filter coefficients, link rate, and coding mode. Optionally, the link configuration parameters may also include receiver equalizer coefficients, reference clock frequency, and output drive amplitude. No limitations are specified here.

[0068] In some embodiments, "at least two link configuration parameters" refers to taking at least two different combinations of values ​​for the same link configuration parameter (such as the TX FIR coefficient), while keeping other parameters (such as link rate and coding mode) the same. In other embodiments, "at least two link configuration parameters" refers to changing multiple configuration parameters simultaneously, such as changing both the TX FIR coefficient and the link rate. This application does not limit this approach.

[0069] Step 2 After the initialization operation corresponding to the j-th link configuration parameter is completed, obtain the bit error rate of the communication link corresponding to the j-th link configuration parameter.

[0070] j is a positive integer. For example, for the j-th link configuration parameter, after initialization is completed and stabilized, the bit error rate BER(j) is recorded.

[0071] Step 3 The bit error rate of the communication link corresponding to at least two link configuration parameters shall be used as at least two bit error rates.

[0072] After initializing all configuration parameters of the link under test and collecting the bit error rate, the bit error rates corresponding to each configuration (denoted as BER(1), BER(2), ..., BER(P), where P is the number of types of configurations under test, and P can be equal to N) are arranged into an array as the raw data for subsequent processing.

[0073] Method 2 described above performs initialization operations and obtains the corresponding bit error rate under at least two link configuration parameters. This multi-configuration parameter sampling process can cover the bit error performance of the link under different working states, so that the evaluation results can reflect the performance of the link under actual working scenarios such as different transmission rates and different equalization modes, thereby expanding the coverage of quality evaluation.

[0074] Method 3: Sample the bit error rate corresponding to at least two time windows as at least two bit error rates.

[0075] In other words, under the condition of fixed link configuration and continuous operation, by dividing the link into multiple consecutive or non-overlapping time windows, the bit error rate within each time window is statistically analyzed, thereby obtaining a set of bit error rate samples reflecting the evolution of link performance over time. Unlike the initialization-driven sampling (Method 1 and Method 2), Method 3 focuses on the time-varying characteristics of the link during actual operation (such as thermal drift, aging effect, environmental disturbances, etc.), and is suitable for long-term stability monitoring and performance trend analysis.

[0076] For example, the bit error rate is sampled for 50 time windows respectively, and 50 bit error rate data are obtained for subsequent communication link quality assessment.

[0077] In an optional embodiment, the communication link is operated under fixed link configuration parameters; during the operation of the communication link, the bit error rate corresponding to at least two time windows is sampled as at least two bit error rates.

[0078] The fixed link configuration parameters include at least the following: transmitter finite impulse response filter coefficients, link rate, and coding mode. Optionally, the fixed link configuration parameters may also include receiver equalizer coefficients, reference clock frequency, and output drive amplitude. No specific limitations are imposed here.

[0079] Sampling the bit error rate corresponding to at least two time windows means dividing the continuous running time of the link into multiple windows in chronological order (the windows can be adjacent, overlapping, or have intervals). At the end of each window, the bit error rate counter and the total number of transmitted bits counter are read, and the average bit error rate within that window is calculated as a bit error rate sample corresponding to that time window.

[0080] In some embodiments, the length of the time window remains constant throughout the sampling process; in other embodiments, a variable-length window strategy may be adopted, such as using a larger window when the link performance is stable to reduce computational overhead, and automatically switching to a smaller window to improve temporal resolution when an increase in bit error rate fluctuations is detected.

[0081] Method 3 described above effectively characterizes the time-varying characteristics of the link by running the communication link under fixed link configuration parameters and sampling the bit error rate corresponding to at least two time windows, thereby improving the accuracy of quality assessment in scenarios where the link is in continuous operation.

[0082] In some embodiments, obtaining at least two bit error rates includes at least one of online acquisition and offline analysis.

[0083] Scenario 1: Online data collection.

[0084] Optionally, during the operation of the communication link, at least two bit error rate (BER) data points are collected; a structured data file is generated based on the at least two BER data points; and the structured data file is parsed to obtain at least two BER values.

[0085] Bit error rate (BER) data refers to the smallest data unit acquired during online data collection that contains the BER value and its associated context information. Each BER data record corresponds to an independent BER sampling result. For example, BER data consists of two parts: core fields and extended fields. Core fields: Required fields used to uniquely identify a single sampling and calculate the bit error rate, including the sampling timestamp, the unique link identifier, and the calculated raw bit error rate value; Extended fields: Optional fields used to record the link and environmental status at the sampling time, including link configuration parameters (transmitter FIR coefficient, link rate, encoding mode, etc.), real-time physical parameters (temperature, power supply voltage, electromagnetic interference intensity), and link status register values ​​(clock lock status, equalizer convergence status, signal strength indication, etc.).

[0086] A structured data file is a storage file that organizes bit error rate data according to predefined data formats and syntax rules. Optionally, the structured data file is used to record bit error rate data in chronological order of acquisition time. Optionally, the format of the structured data file includes, but is not limited to: JSON (JavaScript Object Notation, a lightweight data exchange format), CSV (Comma-Separated Values), XML (Extensible Markup Language), Protocol Buffers (an efficient structured data serialization protocol), etc. This application does not limit the specific file format; in practical applications, the format can be selected based on the data volume, parsing efficiency requirements, and system integration convenience.

[0087] It should be noted that at least two bit error rates collected online can be obtained using any of the methods described in Method 1 to Method 3 above, such as: Method 1: Under the same link configuration parameters, control the communication link to perform at least two initialization operations; after the i-th initialization operation is completed, run the communication link and collect the bit error rate data corresponding to the i-th initialization operation during the communication link process; generate a structured data file based on the bit error rate data corresponding to the at least two initialization operations; parse the structured data file to obtain at least two bit error rates; Method 2: Under at least two link configuration parameters, control the communication link to perform initialization operations respectively; after the initialization operation corresponding to the j-th link configuration parameter is completed, run the communication link, and collect the bit error rate data corresponding to the j-th link configuration parameter during the communication link process; generate a structured data file based on the bit error rate data corresponding to the at least two link configuration parameters respectively; parse the structured data file to obtain at least two bit error rates; Method 3: Run the communication link under fixed link configuration parameters; during the operation of the communication link, collect bit error rate data corresponding to at least two time windows; generate a structured data file based on at least two bit error rate data; parse the structured data file to obtain at least two bit error rates.

[0088] For example, during the operation of the communication link, bit error rate (BER) data is collected according to a preset collection strategy (such as any one of methods one to three above). Each BER data point is written to a file according to a predefined data format and syntax rules to obtain a structured data file. Then, the structured data file is parsed to obtain the original BER value of each BER data point in the structured data file, which is used as the BER for quality assessment in this application.

[0089] Scenario 2: Offline analysis.

[0090] Optionally, obtain the bit error rate array or structured data file corresponding to the communication link; parse the bit error rate array or structured data file to obtain at least two bit error rates.

[0091] A bit error rate (BER) array is an array structure composed of multiple BER values ​​collected during the operation of a communication link, arranged in the sampling order. Each element in the array corresponds to the raw BER value obtained from an independent sampling.

[0092] The structured data file is defined the same as the structured data file generated during the online acquisition process mentioned above. It refers to a storage file that organizes multiple bit error rate values ​​collected during the operation of the communication link according to a predefined data format and syntax rules.

[0093] It should be noted that at least two bit error rates obtained from offline analysis can be obtained using any of the methods described in Method 1 to Method 3 above, such as: Method 1: Under the same link configuration parameters, control the communication link to perform at least two initialization operations; after the i-th initialization operation is completed, obtain the bit error rate of the communication link corresponding to the i-th initialization operation; save the bit error rate data corresponding to the at least two initialization operations as a first bit error rate array or a first structured data file; when quality assessment is required, obtain the first bit error rate array or the first structured data file; parse the first bit error rate array or the first structured data file to obtain at least two bit error rates.

[0094] Method 2: Under at least two link configuration parameters, control the communication link to perform initialization operations respectively; after the initialization operation corresponding to the j-th link configuration parameter is completed, obtain the bit error rate of the communication link corresponding to the j-th link configuration parameter; save the bit error rate data corresponding to the at least two link configuration parameters as a second bit error rate array or a second structured data file; when quality assessment is required, obtain the second bit error rate array or the second structured data file; parse the second bit error rate array or the second structured data file to obtain at least two bit error rates.

[0095] Method 3: Run the communication link under fixed link configuration parameters; during the operation of the communication link, sample the bit error rate data corresponding to at least two time windows; save the bit error rate data corresponding to at least two time windows as a third bit error rate array or a third structured data file; when quality assessment is required, obtain the third bit error rate array or the third structured data file; parse the third bit error rate array or the third structured data file to obtain at least two bit error rates.

[0096] For example, during the operation of the communication link, bit error rate (BER) data is collected according to a preset collection strategy (such as any one of methods one through three above). Each collected BER data point is written to a file according to a predefined data format and syntax rules, resulting in a structured data file or an BER array. Then, the structured data file or the BER array is saved. Subsequently, when a quality assessment of the communication link is required, the structured data file or the BER array is read and parsed to obtain at least two BER values.

[0097] Step 220: Determine the bit error rate statistics based on at least two bit error rates.

[0098] In some embodiments, invalid values ​​from at least two bit error rates are filtered out, and invalid values ​​include at least null or zero values; a bit error rate statistic is determined based on the filtered at least two bit error rates.

[0099] A null value refers to a record where the bit error rate (BER) could not be successfully collected due to reasons such as test interruption, counter read failure, or link not locking properly. A zero value refers to a record where no error bits were detected within the test window, resulting in a BER calculation value of 0.

[0100] For example, the collected raw bit error rate array is traversed, and the validity of each element is determined. If an element is empty, it is marked as invalid; if an element is the value 0, it is marked as invalid; if an element is negative, it is also marked as invalid. All elements marked as invalid are removed from the raw array, and the valid bit error rates are retained, resulting in a filtered valid array. After filtering invalid values, the retained valid bit error rates are calculated to obtain the bit error rate statistics.

[0101] In some embodiments, in addition to direct rejection, alternative strategies can be adopted for zero-value bit error rates, such as replacing zero with a positive number that represents the minimum bit error rate that the current test system can characterize.

[0102] In some embodiments, the bit error rate statistics include at least one of logarithmic statistics and linear statistics.

[0103] Linear statistics are statistical values ​​determined based on at least two bit error rates.

[0104] Optionally, linear statistical values ​​include, but are not limited to, at least one of the following: mean bit error rate (BER) value, maximum BER value, minimum BER value, fluctuation range, and median BER value. The mean BER value refers to the average of at least two BER values; the maximum BER value refers to the maximum of at least two BER values; the minimum BER value refers to the minimum of at least two BER values; the fluctuation range is used to indicate the difference between the maximum and minimum BER values; the median BER value refers to the median of at least two BER values, also known as the 50th percentile. In this application, the fluctuation range is mainly used as an example for illustration.

[0105] Logarithmic statistics are statistical values ​​determined based on the logarithms corresponding to at least two bit error rates. In some embodiments, logarithmic statistics are determined based on the natural logarithms (logarithms to the base of the natural constant e) corresponding to at least two bit error rates.

[0106] Optionally, the logarithmic statistics include, but are not limited to, at least one of the following: logarithmic mean, maximum logarithm, minimum logarithm, geometric mean bit error rate, and logarithmic standard deviation. The logarithmic mean is the average of the logarithms corresponding to at least two bit error rates; the maximum logarithm is the maximum of the logarithms corresponding to at least two bit error rates; the minimum logarithm is the minimum of the logarithms corresponding to at least two bit error rates; the geometric mean bit error rate is the geometric mean of at least two bit error rates; and the logarithmic standard deviation is the standard deviation of the logarithms corresponding to at least two bit error rates. In this application, the geometric mean bit error rate and the logarithmic standard deviation are mainly used as examples for illustration.

[0107] The methods for calculating the geometric mean bit error rate and the logarithmic standard deviation are explained below.

[0108] (a) Geometric mean bit error rate.

[0109] In some embodiments, the average of the natural logarithms corresponding to at least two bit error rates is calculated to obtain the logarithmic mean; the logarithmic mean is then subjected to an inverse natural exponential transform to restore the geometric mean bit error rate.

[0110] In other embodiments, an inverse natural exponential transform is performed on the logarithmic mean to restore the initial geometric mean bit error rate. Real-time physical parameters of the communication link's environment are obtained; the initial geometric mean bit error rate is then corrected based on these real-time physical parameters to obtain the final geometric mean bit error rate.

[0111] Real-time physical parameters refer to the environmental and operational parameters experienced by the communication link during bit error rate (BER) acquisition. Optionally, real-time physical parameters include, but are not limited to, at least one of the following: temperature, power supply voltage, and electromagnetic interference intensity. For example, the correction process refers to adjusting the initial geometric mean BER to an equivalent value under reference environmental conditions using a pre-calibrated real-time physical parameter-BER relationship model; this equivalent value is the final geometric mean BER used for calculation.

[0112] (ii) Logarithmic standard deviation.

[0113] In some embodiments, the standard deviation of the natural logarithm corresponding to at least two bit error rates is calculated to obtain the logarithmic standard deviation. It should be noted that the error of the logarithmic standard deviation decreases as the number of members n (i.e., the number of bit error rates involved in calculating the logarithmic standard deviation) increases. For example, n ≥ 30: the sample size is sufficient to make the estimation of the logarithmic standard deviation (and logarithmic mean) statistically stable; n ≥ 50: in engineering practice, a sample size of 50 or more provides sufficient statistical accuracy.

[0114] Step 230: Perform a quality assessment on the communication link based on the bit error rate statistics to obtain the quality assessment result of the communication link.

[0115] In some embodiments, a quality assessment of the communication link is performed based on logarithmic statistics to obtain a quality assessment result. Alternatively, a quality assessment of the communication link is performed based on both logarithmic and linear statistics to obtain a quality assessment result.

[0116] In some embodiments, the method for obtaining the quality assessment results of a communication link further includes the following steps: Step 1: Determine the health score of the communication link based on the bit error rate statistics.

[0117] In some embodiments, bit error rate statistics are mapped to a health score for the communication link. For example, at least one of the following—mean volatility, volatility range, logarithmic mean, geometric mean bit error rate, and logarithmic standard deviation—is mapped to a health score.

[0118] In other embodiments, the health score of the communication link is determined based on logarithmic statistics.

[0119] Optionally, at least one of a consistency score, a reliability score, and a continuous anomaly score is determined based on logarithmic statistics; a health score for the communication link is then determined based on at least one of the consistency score, reliability score, and continuous anomaly score. For example, the consistency score, reliability score, or continuous anomaly score may be used as the health score. Alternatively, a health score may be determined based on at least two of the consistency score, reliability score, and continuous anomaly score. Or, a health score may be determined based on the consistency score, reliability score, and continuous anomaly score.

[0120] In other embodiments, the health score of the communication link is determined based on logarithmic and linear statistics.

[0121] Optionally, a stability score for the communication link in the stability dimension is obtained based on at least two bit error rate fluctuation ranges, where the stability score is negatively correlated with the fluctuation range; a health score for the communication link is determined based on logarithmic statistics and the stability score. For example, at least one of a consistency score, a reliability score, and a continuous anomaly score is determined based on logarithmic statistics; and a health score for the communication link is determined based on at least one of the consistency score, reliability score, and continuous anomaly score, as well as the stability score. For example: A health score is determined based on a consistency score, a reliability score, a consecutive anomaly score, and a stability score. Alternatively, a health score is determined based on at least two of the consistency score, reliability score, consecutive anomaly score, and a stability score. Or, a health score is determined based on a consistency score, reliability score, consecutive anomaly score, and a stability score.

[0122] Step 2: Determine the quality assessment result of the communication link based on the health score.

[0123] In some embodiments, the health score is used as a quality assessment result of the communication link.

[0124] In other embodiments, a preset rating threshold table is used to map health scores to health ratings, and the health ratings are used as the quality assessment results of the communication link.

[0125] In other embodiments, health ratings and health scores are used as the results of quality assessment of the communication link.

[0126] In some embodiments, each score dimension has a corresponding score algorithm. The score algorithms (i.e., score determination methods) for each of the above score dimensions are described below. In this application, "score dimension" may also be referred to as "quantization factor," which is not limited here.

[0127] (a) Consistency score algorithm for consistency dimension.

[0128] In some embodiments, the consistency score of the communication link in the consistency dimension is obtained based on the logarithmic standard deviation. The consistency score is negatively correlated with the logarithmic standard deviation.

[0129] In an optional embodiment, a preset consistency threshold mapping table is obtained, which includes at least one log standard deviation threshold interval and its corresponding consistency score; the log standard deviation is matched with at least one log standard deviation threshold interval; and the consistency score corresponding to the successfully matched log standard deviation threshold interval is determined as the consistency score of the communication link in the consistency dimension.

[0130] Optionally, the consistency threshold mapping table is a set of predefined key-value pairs, each containing a logarithmic standard deviation threshold range and its corresponding consistency score. This mapping table can be managed through an external configuration file, supporting flexible adjustment of the correspondence between thresholds and scores based on different link rates, modulation schemes, or chip versions.

[0131] In another optional embodiment, a preset consistency score mapping function is obtained, which is used to define the conversion relationship between consistency score and logarithmic standard deviation; the logarithmic standard deviation is converted into the consistency score of the communication link in the consistency dimension through the consistency score mapping function.

[0132] Optionally, the consistency score mapping function is a monotonically decreasing function with the log-standard deviation as the independent variable and the consistency score as the dependent variable. The function form can be chosen from linear decay functions, exponential decay functions, sigmoid decay functions, etc., based on practical engineering experience; no limitation is made here. For example, an exponential decay function can be used as the mapping function to better reflect the tail characteristics of the log-normal distribution of the bit error rate.

[0133] In the above embodiments, a consistency dimension is introduced, and a consistency score is generated based on the logarithmic standard deviation. The logarithmic standard deviation is used to quantify the dispersion of the logarithmic bit error rate. Since the bit error rate follows a log-normal distribution, the logarithmic standard deviation can accurately and objectively quantify the dispersion of the bit error rate data, enabling the consistency score to accurately reflect the consistency of the communication link's performance under multiple tests. This dimension can effectively identify links with drastic performance fluctuations and poor consistency. Such links are more prone to sudden transmission failures in actual operation, and early identification can significantly reduce system operation risks.

[0134] In some embodiments, after obtaining the consistency score, a health score for the communication link is determined based on the consistency score. Optionally, the consistency score is used as the health score; or, the health score is determined based on at least one of the stability score, reliability score, and continuous anomaly score, as well as the consistency score.

[0135] (ii) Stability score algorithm for stability dimension.

[0136] In some embodiments, the stability score of the communication link in the stability dimension is obtained based on at least two bit error rate fluctuation ranges.

[0137] The stability score is negatively correlated with the fluctuation range. That is, the larger the fluctuation range, the more drastic the span between the best and worst states of the link performance, the worse the stability, and the lower the corresponding score.

[0138] Optionally, the fluctuation range is used to characterize the difference between the maximum and minimum bit error rate values ​​among at least two bit error rates. For example, the fluctuation range is determined based on the ratio of the maximum to the minimum bit error rate value. In high-speed link bit error rate testing, since the bit error rate itself can fluctuate across orders of magnitude, defining the fluctuation range as a ratio can more accurately reflect the relative dispersion of extreme values. It should be noted that when the minimum bit error rate is zero, the zero value must first be replaced with a positive number representing the minimum bit error rate characterized by the current test system, following the aforementioned invalid value handling strategy, before the ratio calculation.

[0139] In an optional embodiment, a preset stability threshold mapping table is obtained, the stability threshold mapping table including at least one fluctuation range threshold interval and its corresponding stability score; the fluctuation range is matched with at least one fluctuation range threshold interval; the stability score corresponding to the successfully matched fluctuation range threshold interval is determined as the stability score of the communication link in the stability dimension.

[0140] Optionally, the stability threshold mapping table is a set of predefined key-value pairs, each containing a fluctuation range threshold interval and its corresponding stability score. This mapping table is managed through an external configuration file and can be flexibly adjusted according to factors such as link rate, modulation scheme, and application scenario.

[0141] In another optional embodiment, a preset stability score mapping function is obtained, which is used to define the conversion relationship between stability score and fluctuation range; the fluctuation range is converted into the stability score of the communication link in the stability dimension through the stability score mapping function.

[0142] Optionally, the stability score mapping function is a monotonically decreasing function with the fluctuation range as the independent variable and the stability score as the dependent variable.

[0143] Alternatively, a piecewise function strategy can be introduced: a denser score mapping is used in the high stability range with a smaller fluctuation range to finely distinguish performance differences, while a sparser score mapping is used in the low stability range with a larger fluctuation range, so that the score distribution is more in line with the focus of engineering practice.

[0144] In the above embodiments, a stability dimension is introduced, and a stability score is generated based on the fluctuation range of the bit error rate samples. The fluctuation range can directly reflect the difference between the extreme maximum and minimum values ​​in the bit error rate samples. This allows the evaluation method to not only capture the overall dispersion trend, but also to provide warnings for individual extreme deviations, thereby further strengthening the ability of quality assessment to characterize the stability of link operation.

[0145] In other embodiments, at least two bit error rates are arranged in the order of acquisition to obtain a bit error rate sequence. The bit error rate sequence is divided into multiple continuous subsequences; the ratio of the local fluctuation range to the global fluctuation range of each continuous subsequence is calculated; a fluctuation uniformity score is obtained based on the dispersion of the fluctuation range ratio; and a stability score is determined based on the global fluctuation range and the fluctuation uniformity score.

[0146] The acquisition order refers to the order in which the bit error rates are acquired. In the initialization driver sampling method (Method 1 and Method 2), it corresponds to the execution round of the initialization operation, and in the time window sampling method (Method 3), it corresponds to the timing of each time window.

[0147] Optionally, the method of dividing the bit error rate sequence into multiple consecutive subsequences includes at least one of the following: Method 1: Fixed-length partitioning. Divide the bit error rate sequence into K consecutive subsequences according to a preset subsequence length. If the length of the bit error rate sequence is not divisible by the subsequence length, the remaining samples at the end can form a separate subsequence or be merged into the last complete subsequence.

[0148] Method 2: Equal division. Divide the bit error rate sequence into K consecutive subsequences.

[0149] Method 3: Time window division. When using Method 3 to obtain at least two bit error rates, each time window can be naturally mapped to a subsequence, and the length of the subsequence is the number of bit error rates collected within each time window.

[0150] Optionally, the local fluctuation range is used to characterize the difference between the maximum and minimum bit error rate values ​​in the subsequence. For example, the local fluctuation range is determined based on the ratio of the maximum to the minimum bit error rate value in the subsequence.

[0151] Optionally, the global fluctuation range is used to characterize the difference between the maximum and minimum bit error rate values ​​in the bit error rate sequence. For example, the global fluctuation range is determined based on the ratio of the maximum to the minimum bit error rate value in the bit error rate sequence.

[0152] After calculating the global fluctuation range, the ratio between the local fluctuation range and the global fluctuation range of each continuous subsequence is calculated. For example, the fluctuation range ratio of subsequence m is: local fluctuation range of subsequence m ÷ global fluctuation range. This fluctuation range ratio measures the proportion of the local fluctuation amplitude of each subsequence relative to the global fluctuation amplitude.

[0153] Then, based on the fluctuation range ratio of each continuous subsequence, a fluctuation uniformity score is determined. This score quantifies the dispersion of the fluctuation range ratios of each subsequence, reflecting the uniformity of the link performance fluctuations over time. Smaller dispersion indicates closer fluctuation amplitudes across stages, better fluctuation uniformity, and a higher score. Larger dispersion indicates significantly deviated fluctuations from other periods, poorer fluctuation uniformity, and a lower score. Optionally, the standard deviation of the fluctuation range ratios of all subsequences is calculated as the dispersion; this dispersion is then converted into a fluctuation uniformity score using a pre-defined mapping table or function.

[0154] After calculating the fluctuation uniformity score, the basic stability score is determined based on the global fluctuation range. Finally, the stability score is determined based on both the fluctuation uniformity score and the basic stability score. The method for "determining the basic stability score based on the global fluctuation range" can be found in the explanation of the method for "determining the stability score based on the fluctuation range" described above, and will not be repeated here.

[0155] Optionally, the method for determining the final stability score based on the fluctuation uniformity score and the basic stability score includes: assigning weights to the fluctuation uniformity score and the basic stability score respectively, and obtaining the final stability score by weighted summation. Alternatively, using the basic stability score as a benchmark, when the fluctuation uniformity score is lower than a preset threshold, the basic stability score is lowered to obtain the final stability score; when the fluctuation uniformity score is higher than or equal to the preset threshold, the basic stability score is used as the final stability score.

[0156] In some embodiments, after obtaining the stability score, the health score of the communication link is determined based on the logarithmic statistics and the stability score.

[0157] Optionally, at least one of a consistency score, a reliability score, and a continuous anomaly score is determined based on logarithmic statistics; and a health score of the communication link is determined based on at least one of the consistency score, the reliability score, and the continuous anomaly score, as well as a stability score.

[0158] (iii) Reliability score algorithm for reliability dimension.

[0159] In some embodiments, a reliability score for the communication link in the reliability dimension is obtained based on the geometric mean bit error rate. The steps for determining the reliability score include steps 1 to 3 as follows.

[0160] Step 1: Determine at least one bad bit error rate from at least two bit error rates.

[0161] In some embodiments, at least one candidate bad bit error rate is determined from at least two bit error rates based on the quantiles of at least two bit error rates; and the bit error rate that is greater than the geometric mean bit error rate among the at least one candidate bad bit error rate is determined as the bad bit error rate.

[0162] Optionally, at least one quantile corresponding to at least two bit error rates is determined; at least two quantile intervals are determined by dividing the at least two quantiles, and the at least two quantile intervals include a target interval that meets the requirements for indicating a poor bit error rate; the bit error rate in the target interval among the at least two bit error rates is determined as a candidate poor bit error rate. For example, a first quantile, a second quantile, and a third quantile corresponding to at least two bit error rates are determined; a first interval, a second interval, a third interval, and a fourth interval are determined by dividing the first quantile, the second quantile, and the third quantile; the first interval is the interval where the bit error rate is less than or equal to the first quantile, the second interval is the interval where the bit error rate is greater than the first quantile and less than or equal to the second quantile, the third interval is the interval where the bit error rate is greater than the second quantile and less than or equal to the third quantile, and the fourth interval is the interval where the bit error rate value is greater than the third quantile; the bit error rate in the fourth interval among the at least two bit error rates is determined as a candidate poor bit error rate.

[0163] The following explanation uses the example of the first quantile being implemented as the 25th quantile, the second quantile as the 50th quantile, and the third quantile as the 75th quantile.

[0164] The first interval (or Excellent interval): BER ≤ 25th percentile, representing the best performance level; The second interval (or Good interval): 25th percentile < BER ≤ 50th percentile, indicating a good performance level; The third interval (or Fair interval, acceptable interval): 50th percentile < BER ≤ 75th percentile, representing an acceptable level of performance; The fourth interval (also known as the Poor candidate interval or bad candidate interval, which is the interval that meets the requirements of the bad bit error rate indication): BER > 75th percentile, may be an outlier.

[0165] The bit error rate that falls within the Poor candidate interval among at least two bit error rates is taken as the candidate poor bit error rate.

[0166] After determining all candidate bad bit error rates, it is further determined whether they are greater than the geometric mean BER. If they are greater, they are taken as bad bit error rates.

[0167] In other embodiments, the bit error rate that is greater than the target quantile and greater than the geometric mean bit error rate among at least two bit error rates is determined as the bad bit error rate.

[0168] Optionally, the target quantile is determined based on at least one quantile corresponding to at least two bit error rates. For example, the target quantile is the 75th quantile of at least two bit error rates; or, the target quantile is the 50th quantile of at least two bit error rates; or, the target quantile is the average or weighted average of at least two quantiles of at least two bit error rates, such as the target quantile being the 75th and 50th quantiles of at least two bit error rates, or the target quantile being the weighted average of the 75th and 50th quantiles of at least two bit error rates. The weighting of each quantile is used to quantify the contribution of different quantiles in defining the anomaly threshold and can be dynamically configured according to different quality scenarios, such as: The weighting configuration for the quick filtering scenario is: 75th percentile weight 0.7, 50th percentile weight 0.3. Under this configuration, the weighted quantile is closer to the 75th percentile, which can effectively filter out configurations with poor tail performance. The initial weighting configuration for the stability assessment scenario is: 75th percentile weight 0.5, 50th percentile weight 0.5. Under this configuration, the weighted quantile is the arithmetic mean of the two quantiles, which can balance the sensitivity to central trends and tail fluctuations.

[0169] The weighting configuration for the production yield analysis scenario is as follows: 75th percentile weight 0.4, 50th percentile weight 0.6. Under this configuration, the weighted quantile is closer to the median, which can reduce misjudgments caused by individual extreme values ​​and improve the accuracy of yield statistics.

[0170] The weighting configuration for fault diagnosis scenarios is as follows: 75th percentile weight 0.9, 50th percentile weight 0.1. Under this configuration, the weighted quantile is almost equivalent to the 75th percentile, which can sensitively identify any high bit error rate samples that exceed the normal range.

[0171] After determining the target quantile, the bit error rate that is greater than the target quantile and greater than the geometric mean bit error rate among at least two bit error rates is determined as the bad bit error rate.

[0172] It should be noted that the number of at least two bit error rates used to determine the bad bit error rate is greater than or equal to 4, that is, at least 4 bit error rates need to be obtained to determine the bad bit error rate.

[0173] In the above embodiments, a dual screening mechanism is constructed by introducing quantiles (such as at least one of the 75th quantile and 50th quantile) as a threshold for screening defective bit error rates, combined with the condition that the bit error rate is greater than the geometric mean. This mechanism uses the statistical distribution characteristics (quantiles) of the data itself to dynamically define anomalies, rather than using fixed or subjective thresholds. It can adaptively exclude normal dispersion caused by the overall volatility of the data (characterized by the logarithmic standard deviation), and identify extreme abnormal bit errors that are statistically significantly higher than the average level. This design avoids misjudging normal fluctuations at the tail of the distribution as defective events, and improves the anti-interference and robustness of the reliability dimension assessment.

[0174] In other embodiments, the bit error rate that deviates from at least two bit error rates to a predetermined deviation requirement is designated as the poor bit error rate. Optionally, the deviation is used to indicate the difference between the poor bit error rate and the geometric mean bit error rate. If the deviation of a certain bit error rate is greater than k × logarithmic standard deviation, then that bit error rate is determined as the poor bit error rate, where k represents a multiple, such as k equal to 2.

[0175] Step 2 The first penalty score is determined based on the degree of deviation between at least one bad bit error rate and the geometric mean bit error rate.

[0176] Optionally, for each defective bit error rate, the ratio of that defective bit error rate to the geometric mean bit error rate is calculated as the degree of deviation of that defective bit error rate. Then, based on the degree of deviation corresponding to each defective bit error rate, a first penalty score is determined.

[0177] Optionally, the first penalty score is determined based on at least one of the deviation penalty score and the proportion penalty score.

[0178] (1) Deviation penalty points.

[0179] Optionally, a deviation penalty score is determined based on the degree of deviation between at least one defective bit error rate and the geometric mean bit error rate. The deviation penalty score is positively correlated with the degree of deviation. The deviation penalty score quantifies the severity of the numerical deviation of the defective bit error rate from the geometric mean bit error rate; the greater the deviation, the more severe the damage to link reliability, and the heavier the penalty.

[0180] Optionally, the method for determining the deviation penalty score includes: obtaining a preset deviation penalty mapping table, which includes at least one deviation degree threshold interval and its corresponding deviation penalty score; for each defective bit error rate, matching the deviation degree of the defective bit error rate with at least one deviation degree threshold interval to obtain the single deviation penalty score corresponding to the defective bit error rate; and accumulating the single deviation penalty scores corresponding to all defective bit error rates to obtain the total deviation penalty score. For example, a communication link identifies three defective bit error rates with deviation degrees of 3.5 times, 8 times, and 45 times, respectively. Looking up the table, the single deviation penalty scores are 2, 3, and 5 points, respectively, with a total deviation penalty score of 10 points.

[0181] (2) Proportional penalty points.

[0182] The proportional penalty score quantifies the frequency of defective bit error rates across the entire sample, and is positively correlated with the proportion. Even if the deviation of a single defective bit error rate is small, frequent occurrences of defective bit error rates indicate a systemic reliability vulnerability in the link, warranting additional penalty.

[0183] Optionally, a proportional penalty score is determined based on the proportion of at least one bad bit error rate to at least two bit error rates.

[0184] Optionally, the method for determining the deviation penalty score includes: calculating the ratio of the number of defective bit error rates to the total number of at least two bit error rates; obtaining a preset proportional penalty mapping table, which includes at least one proportional threshold interval and its corresponding proportional penalty score; and matching the ratio with at least one proportional threshold interval to obtain the proportional penalty score. For example, in a link, 3 defective bit error rates are identified out of 50 bit error rate samples, with a ratio R_poor of 6%. A table lookup matches to the interval 5% < R_poor ≤ 15%, resulting in a proportional penalty score of 2 points.

[0185] Optionally, the deviation penalty score can be used as the first penalty score. Alternatively, the proportional penalty score can be used as the first penalty score. Or, the first penalty score can be determined based on the deviation penalty score and the proportional penalty score; for example, the deviation penalty score and the proportional penalty score can be added together to obtain the first penalty score.

[0186] In the above embodiments, the first penalty score is decomposed into a deviation penalty score and a proportion penalty score. Corresponding penalty scores are generated based on the degree of deviation and the proportion of the defective bit error rate. The degree of deviation quantifies the severity of a single anomaly, while the proportion quantifies the frequency of anomalies. This design achieves a dual quantitative assessment of the defective bit error rate: it focuses on penalizing severe single anomalies with large deviations, while also cumulatively penalizing frequently occurring anomalies. This allows for a more accurate matching of the actual risk level posed to the system by different anomalies, further improving the accuracy and rationality of reliability dimension assessment.

[0187] Step 3 Based on the first penalty score, determine the reliability score of the communication link in the reliability dimension.

[0188] The reliability score is negatively correlated with the first penalty score. For example, the reliability score is a preset starting baseline score minus the first penalty score, and the reliability score is not lower than zero. Optionally, the starting baseline score is associated with a reliability weight; for example, a higher reliability weight results in a higher starting baseline score.

[0189] For example, the aforementioned link calculates a deviation penalty score of 10 points and a proportional penalty score of 2 points. Summing these together yields a first penalty score of 12 points. The initial baseline score is 25 points. The reliability score is 25 - 12 = 13 points.

[0190] In the above embodiments, a reliability dimension is introduced to calculate the deviation of the bit error rate from the geometric mean bit error rate and determine the first penalty score accordingly. Finally, a reliability score negatively correlated with the penalty score is obtained. Since the bit error rate follows a log-normal distribution, the penalty mechanism based on deviation can accurately and objectively quantify the impact of extreme abnormal bit errors on actual transmission, enabling the reliability score to accurately reflect the abnormal interference capability of the communication link during actual data transmission. This dimension can sensitively capture individual abnormally prominent bit error samples, effectively identifying links with poor resistance to random disturbances and prone to sudden high bit error rates, avoiding the risk of under-judgment due to extreme defects being masked by the arithmetic mean evaluation method.

[0191] In some embodiments, after determining the reliability score, a health score for the communication link is determined based on the reliability score. Optionally, the reliability score is used as the health score; or, the health score is determined based on at least one of a stability score, a consistency score, and a continuous anomaly score, as well as the reliability score.

[0192] (iv) Algorithm for continuous anomaly scores in the continuous anomaly dimension.

[0193] In some embodiments, the continuous anomaly score of the communication link in the continuous anomaly dimension is obtained based on the geometric mean bit error rate and the logarithmic standard deviation. The steps for determining the continuous anomaly score include steps 1 to 4 as follows.

[0194] Step 1 The anomaly threshold is determined based on the geometric mean bit error rate and the logarithmic standard deviation.

[0195] Optionally, a preset anomaly multiple is obtained; the anomaly threshold = geometric mean bit error rate × e 异常倍数×对数标准差 e represents the natural constant.

[0196] Step 2 Based on the abnormal threshold, at least one abnormal sequence among at least two bit error rates is identified.

[0197] An abnormal sequence includes at least two consecutive abnormal bit error rates, where the abnormal bit error rate is the bit error rate that meets the abnormal threshold requirement among the at least two bit error rates.

[0198] Optionally, each bit error rate is traversed according to the acquisition time order of at least two bit error rates. When the bit error rate is greater than the abnormal threshold, the current consecutive abnormal count is incremented by 1. When the bit error rate is not greater than the abnormal threshold, it is determined whether the current consecutive abnormal count has reached the preset minimum consecutive abnormal length. If it has, it is recorded as an abnormal sequence and added to the abnormal sequence list. Then, the current consecutive abnormal count is cleared. After the traversal is completed, the longest abnormal sequence and the number of abnormal sequences are determined based on the recorded abnormal sequence list. Here, the longest abnormal sequence refers to the length of the longest abnormal sequence in the abnormal sequence list (that is, the number of abnormal bit error rates contained in the longest abnormal sequence), and the number of abnormal sequences refers to the total number of abnormal sequences in the abnormal sequence list.

[0199] Step 3 A second penalty score is determined based on at least one abnormal sequence.

[0200] The second penalty score is used to quantify the negative impact of consecutive anomalies on the overall quality of the link. Optionally, the second penalty score is determined based on at least one of the first sub-score and the second sub-score.

[0201] (1) First subfraction.

[0202] Optionally, a first sub-score is determined based on the longest anomalous sequence among at least one anomalous sequence. The first sub-score is positively correlated with the number of anomalous bit error rates in the longest anomalous sequence.

[0203] Optionally, a preset continuous anomaly penalty mapping table is obtained, which includes at least one anomaly sequence length threshold and its corresponding first sub-score; the number of abnormal bit error rates in the longest anomaly sequence is matched with at least one anomaly sequence length threshold to obtain the first sub-score.

[0204] (2) Second subfraction.

[0205] Optionally, a second sub-score is determined based on the number of sequences in at least one anomalous sequence. The second sub-score is positively correlated with the number of sequences.

[0206] Optionally, a preset multi-sequence penalty mapping table is obtained, which includes at least one sequence number threshold and its corresponding second sub-score; the number of abnormal sequences is matched with at least one sequence number threshold to obtain the second sub-score.

[0207] Optionally, the first sub-score can be used as the second penalty score; or, the second sub-score can be used as the second penalty score; or, the second penalty score can be determined based on the first sub-score and the second sub-score, for example, the one with the larger absolute value between the first sub-score and the second sub-score can be determined as the second penalty score; or, the weighted sum of the first sub-score and the second sub-score can be calculated as the second penalty score.

[0208] In the above embodiments, the second penalty score is decomposed into a first sub-score positively correlated with the length of the longest anomalous sequence and a second sub-score positively correlated with the number of anomalous sequences. This design covers two core dimensions of continuous anomalies: the length of the longest anomalous sequence reflects the most severe nature of a single continuous failure, directly determining the maximum potential data loss; the number of anomalous sequences reflects the frequency of continuous failures, embodying the long-term stability of the link. Combining these two aspects allows for a comprehensive and accurate quantification of the overall risk posed by continuous anomalies, ensuring that the evaluation results for the continuous anomaly dimension are objective and reliable.

[0209] Step 4 Based on the second penalty score, the continuous anomaly score of the communication link in the continuous anomaly dimension is determined, and the continuous anomaly score is negatively correlated with the second penalty score.

[0210] Optionally, the consecutive anomaly score is the negative of the second penalty score, that is, consecutive anomaly score = -second penalty score. For example, if the aforementioned link's second penalty score is 4 points, then the consecutive anomaly score = -4 points.

[0211] In the above embodiments, a continuous anomaly dimension is introduced. An anomaly threshold is determined based on the geometric mean bit error rate and the logarithmic standard deviation. Continuous anomaly sequences are identified, and a second penalty score is calculated, ultimately yielding a continuous anomaly score. The anomaly threshold distinguishes between normal and abnormal bit errors, and the continuous anomaly sequence quantifies the characteristics of continuous faults. Since the bit error rate follows a log-normal distribution, anomaly determination based on statistical distribution can accurately and objectively quantify the destructive impact of continuous abnormal bit errors on the system. This allows the continuous anomaly score to accurately reflect the communication link's resilience to continuous faults in long-term continuous data transmission scenarios. This dimension effectively identifies links with multiple consecutive abnormal bit error sequences, ensuring system stability in long-term transmission scenarios.

[0212] In some embodiments, after obtaining the continuous anomaly score, a health score for the communication link is determined based on the continuous anomaly score. Optionally, the continuous anomaly score is used as the health score; or, the health score is determined based on at least one of a stability score, a consistency score, and a reliability score, as well as the continuous anomaly score.

[0213] In other embodiments, logarithmic statistics are input into a multimodal evaluation model to obtain a quality assessment result, wherein the multimodal evaluation model is an artificial intelligence model obtained through training. For example, in this embodiment, the multimodal evaluation model can be implemented as a fully connected deep neural network, encoding the total bit error rate, geometric mean bit error rate, logarithmic standard deviation, maximum bit error rate, minimum bit error rate, fluctuation range, number and proportion of defective bit error rates, etc., into a feature vector. This feature vector is then input into the fully connected deep neural network to obtain a health score or health rating as the quality assessment result.

[0214] In other embodiments, signal integrity parameters of the communication link are obtained; logarithmic statistics and signal integrity parameters are input into a multimodal evaluation model to obtain a quality evaluation result. Signal integrity parameters include, but are not limited to, at least one of: equalizer convergence parameters, clock data recovery parameters, and channel characteristic parameters. For example, in this embodiment, the multimodal evaluation model can be implemented as an attention-based multimodal fusion network. In this attention-based multimodal fusion network, logarithmic statistics are encoded as a first feature vector; signal integrity parameters are encoded as a second feature vector; the interaction relationship between the first and second feature vectors is learned through a cross-attention mechanism to obtain a fused feature vector; the fused feature vector is input into the output head of the fully connected layer to obtain a health score or health rating as the quality evaluation result.

[0215] In summary, the communication link quality assessment method provided in this application obtains at least two bit error rates (BERs), calculates BER statistics including logarithmic values, and performs a quality assessment on the communication link based on the BER statistics to obtain the quality assessment result. Compared with the single-point testing method in related technologies that simply collects multiple BERs and calculates their arithmetic mean, the assessment process in this application utilizes the log-normal distribution of BERs and uses BER statistics including logarithmic values ​​for quality assessment. This can more accurately reproduce the actual distribution pattern of BERs, thereby more realistically characterizing the transmission quality of communication links (especially high-speed links) in actual production environments, improving the accuracy of quality assessment, and ensuring the operational stability and data processing efficiency of high-performance computing systems.

[0216] In some embodiments, the communication link described above includes a high-speed link.

[0217] Optionally, a high-speed link is a communication link whose communication rate meets a preset high-speed requirement. For example, a high-speed link is a communication link whose communication rate is greater than or equal to a preset rate.

[0218] Optionally, the high-speed link includes a serializer / deserializer SerDes channel.

[0219] In the above embodiments, by applying the communication link quality assessment method to high-speed links, especially SerDes channels, the problem of traditional arithmetic mean-based quality assessment methods failing to accurately characterize the bit error rate (BER) of high-speed SerDes channels, which naturally follows a log-normal distribution, leading to significant bias in assessment results and hindering practical debugging, is addressed. This application obtains at least two BER values ​​and performs quality assessment based on logarithmic statistics, adapting to the log-normal distribution of BER caused by crosstalk, jitter, and inter-symbol interference in SerDes links. This significantly improves the effectiveness and reliability of high-speed SerDes channel quality assessment.

[0220] In some embodiments, long-term trend monitoring of communication links can be performed, and predictive maintenance of communication links can be carried out based on periodically acquired scoring data, thereby improving the availability of communication links.

[0221] Optionally, the above at least two bit error rates are bit error rate data acquired periodically. The acquisition methods for periodic bit error rate data include at least one of the following: Method 1: Periodic sampling based on time windows. Under the condition that the communication link is fixed and continuously running, according to a preset sampling period (e.g., hourly, daily, weekly), time windows are automatically divided during the operation of the communication link, and the bit error rate within the time window is sampled as the bit error rate sample for this period.

[0222] Method 2: Periodic sampling based on traffic volume. A fixed number of transmitted bits is used as the sampling window, and the bit error rate within that window is calculated.

[0223] Method 3: Event-triggered supplementary sampling. When a preset trigger condition is detected (such as a change in connectivity, a momentary bit error rate exceeding a preset alarm threshold, or an ambient temperature exceeding the normal operating range), a supplementary sampling cycle is automatically triggered to capture bit error rate data before and after the abnormal event occurs.

[0224] Optionally, the number of bit error rates acquired in each sampling period can be set with reference to the aforementioned recommended sample size for specific scenarios. In trend monitoring scenarios, at least 50 bit error rate samples can be acquired in each period to ensure that the estimation of the logarithmic statistics within the period has sufficient statistical stability, thereby ensuring the reliability of subsequent trend analysis.

[0225] At the end of each sampling period, a complete quality assessment process is performed on at least two bit error rates acquired in that period, namely steps 210 to 230 above, which will not be elaborated here.

[0226] Optionally, after obtaining the quality assessment results of the communication link, the method further includes: determining the scoring trend corresponding to at least two periods; and performing predictive maintenance on the communication link based on the scoring trend.

[0227] The scoring trend refers to the directional characteristics of the evolution of various evaluation scores over time within at least two consecutive periods for the communication link. Specifically, the scoring trend includes at least one of the following: the trend of changes in the health score, the trend of changes in the consistency score, the trend of changes in the stability score, the trend of changes in the reliability score, and the trend of changes in the continuous anomaly score for the corresponding communication link.

[0228] Predictive maintenance refers to proactively taking maintenance measures based on the results of scoring trend analysis, before a substantial service interruption or serious performance failure occurs in the communication link, in order to avoid or reduce the impact of the failure. Optionally, predictive maintenance includes at least one of link performance degradation early warning and predictive communication link failure.

[0229] (a) Early warning of link performance degradation.

[0230] Optionally, in response to a scoring trend indicating a downward trend in consistency and / or reliability scores over at least two periods, a link performance degradation warning is output. For example, the warning information may include: the dimension triggering the warning and details of the trend, suggested troubleshooting directions (e.g., checking the clock recovery circuit lockout status, monitoring power ripple changes, checking for oxidation or loosening of physical connection interfaces, etc.), and suggested maintenance window time. With early warnings, maintenance personnel can perform preventative checks and parameter optimizations on the link within the planned maintenance window, avoiding sudden link failures during peak business periods.

[0231] (ii) Fault prediction based on the trend of continuous abnormal score changes.

[0232] Optionally, communication link failures can be predicted based on the trend of continuous abnormal scores over at least two periods in the scoring trend (i.e., the trend of continuous abnormal scores).

[0233] For example, analyze the changing trend of continuous abnormal scores over at least two periods to determine the characteristic patterns of the continuous abnormal score changing trend. For instance, Pattern 1: The continuous abnormal score suddenly shows a large negative value in a certain period (e.g., a sudden change from 0 to -8), and then returns to 0 or a smaller negative value in the subsequent period. This pattern usually indicates that the link has encountered occasional external interference (such as electromagnetic interference, power transient drops) or a single abnormal event, with a medium risk of failure. It is recommended to pay attention to and record the event time for subsequent correlation analysis. Another example is Pattern 2: The continuous abnormal score alternates between negative and zero values ​​over multiple periods, exhibiting intermittent abnormality characteristics. This pattern indicates that the link has a potential intermittent fault source, with a high risk of failure. It is recommended to focus on investigating during the period of fluctuation. Yet another example is Pattern 3: The continuous abnormal score remains negative for several consecutive periods, and the absolute value of the negative value shows an increasing trend, indicating that the length and / or frequency of continuous abnormalities is continuously increasing. This pattern indicates that the link is undergoing a continuous degradation process, with an extremely high risk of failure. It is recommended to arrange a shutdown for inspection or switch to a backup link as soon as possible.

[0234] Then, based on the feature pattern matching and preset fault prediction rules, fault prediction information is output. Fault prediction information may include: the predicted fault type (persistent or intermittent fault), the predicted fault occurrence time window, and suggested emergency measures (such as link speed reduction, switching to a backup link, enabling enhanced error correction mode, etc.). For example, if a link's continuous anomaly score is 0, -4, -8, and -8 in four consecutive monitoring periods, matching pattern 3, the system outputs a fault prediction alarm: It predicts that this link may experience persistent data transmission failure in the future, and recommends switching to a redundant link in the next planned maintenance window.

[0235] In the above embodiments, based on the obtained quality assessment results, a scoring trend analysis of at least two consecutive periods is further introduced, and predictive maintenance is performed based on the trend characteristics. By tracking the dynamic characteristics of the communication link's scores over time in multiple dimensions such as health, consistency, stability, reliability, and continuous anomalies, the originally isolated assessment results can be transformed into a temporal pattern that characterizes the performance degradation process. By utilizing trend information (such as continuous decline, increased fluctuations, and increasing frequency of anomalies), link performance problems can be identified in advance, and link performance degradation warnings or fault predictions can be proactively triggered before substantial service interruptions or serious failures occur, thereby improving system availability.

[0236] In some embodiments, at least two dimensions include a consistency dimension, a reliability dimension, a continuous anomaly dimension, and a stability dimension. In this case, step 230 can also be implemented as steps 231 and 232.

[0237] Step 231: Determine the health score of the communication link based on the consistency score of the consistency dimension, the stability score of the stability dimension, the reliability score of the reliability dimension, and the continuous anomaly score of the continuous anomaly dimension.

[0238] Optionally, the health score of the communication link is obtained by calculating the weighted sum of the consistency score of the consistency dimension, the stability score of the stability dimension, the reliability score of the reliability dimension, and the continuous anomaly score of the continuous anomaly dimension.

[0239] Optionally, the consistency dimension corresponds to a consistency weight (a%), the stability dimension corresponds to a stability weight (b%), and the reliability dimension corresponds to a reliability weight (c%), where a% + b% + c% = 100%. Therefore, the health score = consistency score × consistency weight + stability score × stability weight + reliability score × reliability weight + continuous anomaly score. The continuous anomaly score is negative, and the continuous anomaly score × reliability weight + continuous anomaly score ≥ 0.

[0240] It should be noted that the health score is typically calculated using the consistency score, stability score, reliability score, and continuous anomaly score simultaneously. However, it is also possible to select only some of these four scores to calculate the health score, depending on actual needs. For example, the health score can be calculated using only the consistency score, stability score, and reliability score (e.g., Health Score = Consistency Score × Consistency Weight + Stability Score × Stability Weight + Reliability Score × Reliability Weight), or only the consistency score, reliability score, and continuous anomaly score (e.g., Health Score = Consistency Score × Consistency Weight + Reliability Score × Reliability Weight + Continuous Anomaly Score). No limitation is imposed here. Optionally, the specific dimensions for calculating the health score can be selected through the configuration file corresponding to the communication link.

[0241] In the above embodiments, the quality assessment result is obtained by mapping logarithmic statistics to scores of at least two evaluation dimensions and combining them. On the one hand, scores of different dimensions can independently reflect the specific performance of the link in terms of consistency, reliability, etc.; on the other hand, the assessment result formed by the fusion of multi-dimensional scores makes the quality assessment more comprehensive and in-depth, thus improving the quality assessment effect.

[0242] In other embodiments, a weighted sum of the consistency score (consistency dimension), the reliability score (reliability dimension), the stability score (stability dimension), and the continuous anomaly score (continuous anomaly dimension) is calculated to obtain the basic health score of the communication link. If at least two of the consistency score, reliability score, and stability score fail to meet the threshold requirements for their respective dimensions, an additional penalty score is determined. The additional penalty score is then subtracted from the basic health score to obtain the final health score.

[0243] For example, after calculating the basic health score, the system further checks whether the scores of the three key dimensions (consistency, reliability, and stability) meet their respective preset threshold requirements. The threshold for each dimension can be an absolute lower limit or a percentage threshold relative to the baseline. If two or three dimensions fail to meet the threshold requirements for their respective dimensions, additional penalty points are obtained to further deduct points from the basic health score, such as Health Score = Basic Health Score - Additional Penalty Points.

[0244] Optionally, the method for determining additional penalty points includes: If the number of dimensions that do not meet the threshold requirement is 2, the additional penalty score is the preset first score; If the number of dimensions that do not meet the threshold requirement is 3, the additional penalty score is the preset second score; the absolute value of the first score is less than the second score.

[0245] Step 232: Determine the quality assessment result of the communication link based on the health score.

[0246] In some embodiments, after determining the health score, the health score is mapped to a corresponding health rating according to a preset rating threshold table. The rating threshold table contains multiple sets of mapping relationships between health score ranges and corresponding ratings. Optionally, the health rating includes seven levels: Excellent, Very Good, Very Good, Average, Poor, and Unacceptable.

[0247] Optionally, the dimension scores, health ratings, and health scores used to determine the health score can be used as the results of the quality assessment.

[0248] Optionally, after obtaining the quality assessment results of the communication link, the method further includes generating a link health report for the communication link. The link health report includes, but is not limited to, at least one of the following: a test configuration summary, core statistical results, details of the dimensional scores used to determine the health score, health score and health rating, a list of poor bit error rates, optimization suggestions based on the health rating, and a bit error rate performance trend chart. The bit error rate performance trend chart includes at least one of the following: a measured bit error rate curve, a geometric mean bit error rate curve, and a median bit error rate curve.

[0249] In summary, the method provided in this application generates a health score by combining consistency score, reliability score, stability score, and continuous anomaly score, and determines the quality assessment result accordingly. This quality assessment result covers key quality characteristics of communication link performance, such as central tendency, dispersion, extreme anomaly risk, continuous failure resistance, and long-term operational stability, thereby comprehensively, accurately, and objectively characterizing the overall quality of the communication link.

[0250] In some embodiments, at least two candidate bit error rates of the communication link are subjected to log-normal distribution fit detection, and outlier bit error rates that deviate from the log-normal distribution are removed to obtain at least two bit error rates for quality assessment. Then, step 210 above can also be implemented as steps 211 to 214 below.

[0251] Step 211: Obtain at least two candidate bit error rates for the communication link.

[0252] It should be noted that the specific method for obtaining at least two candidate bit error rates of the communication link can be found in the introduction to obtaining at least two bit error rates of the communication link in step 210 above, and will not be repeated here.

[0253] Step 212: Obtain the log-normal distribution fit of at least two candidate bit error rates.

[0254] The log-normal distribution fit is a quantitative indicator used to measure how well a set of candidate bit error rate samples statistically follows a log-normal distribution. Since the quality assessment method in this application is based on the theoretical premise that the bit error rate follows a log-normal distribution, directly calculating the geometric mean bit error rate and log standard deviation may produce significant errors when the sample set deviates significantly from this distribution. Therefore, by using a fit test, it can be determined whether the sample data satisfies the distribution assumption; if not, further processing is required.

[0255] In some embodiments, a method for obtaining the log-normal distribution fit includes: Step 1: Take the natural logarithm of at least two candidate bit error rates to obtain a natural logarithm sequence.

[0256] Step 2: Calculate the quantiles of the natural logarithm sequence.

[0257] Quantiles include the upper quartile (75th quartile), the lower quartile (25th quartile), and the median (50th quartile).

[0258] Step 3: Calculate the quantile ratio index.

[0259] Alternatively, the quantile ratio index = (upper quartile - median) ÷ (median - lower quartile).

[0260] The standard normal distribution is symmetric about the median. Theoretically, the distances from the quartiles and lower quartiles to the median are equal, meaning the theoretical value of the quantile ratio index is 1. If at least two candidate bit error rates approximately follow a normal distribution, then their corresponding quantile ratio index should be close to 1.

[0261] Step 4: Determine the log-normal distribution fit based on the quantile ratio index.

[0262] For example, the goodness of fit for the log-normal distribution is F = 1 - |quantile ratio index - 1|. The closer F is to 1, the more symmetrical the distribution of ln(BER) is, that is, the closer it is to a normal distribution.

[0263] Step 213: If the goodness of fit to the log-normal distribution is less than the preset goodness of fit, determine the outlier bit error rate that deviates from the log-normal distribution from at least two candidate bit error rates.

[0264] The preset fit is a pre-defined threshold used to determine whether the candidate bit error rate sample set meets the log-normal distribution requirement. For example, the preset fit can be set to 0.85.

[0265] Outlier bit error rate (BER) refers to the bit error rate that causes the entire sample set to deviate from a log-normal distribution. For example, outlier determination boundaries are defined as follows: the lower bound of the outlier determination boundary = lower quartile - k × (upper quartile - lower quartile); the upper bound of the outlier determination boundary = lower quartile + k × (upper quartile - lower quartile). Here, k is a configurable determination coefficient. The natural logarithm value of each candidate BER is iterated. If the natural logarithm value of a candidate BER is less than the lower bound or greater than the upper bound, then that candidate BER is determined to be an outlier.

[0266] Step 214: Remove outlier bit error rates from at least two candidate bit error rates to obtain at least two bit error rates.

[0267] After identifying outlier bit error rates, they are removed from the candidate sample set, and only samples that are not considered outliers are retained, forming at least two bit error rates for subsequent logarithmic statistical analysis.

[0268] In summary, the method provided in this application solves the problem that the log-normal distribution fit of the bit error rate is disrupted by extreme outliers introduced during the sampling process, which leads to large errors in the calculation of key statistical values ​​such as geometric mean bit error rate and log standard deviation. This is achieved by first obtaining at least two candidate bit error rates of the communication link, calculating their log-normal distribution fit, and then removing outliers that deviate from the distribution when the fit is insufficient. This ensures the accuracy of subsequent logarithmic statistical analysis and improves the accuracy and reliability of the quality assessment results.

[0269] In some embodiments, the evaluation rules in the communication link quality assessment process can also be configured by obtaining a configuration file. Therefore, the above... Figure 2 The embodiment also includes the following step 240. Optionally, step 240 is performed before step 210, or before step 220, or before step 230, and no limitation is made here.

[0270] Step 240: Obtain the configuration file corresponding to the communication link.

[0271] The configuration file (or external configuration file) is used to configure the evaluation rule parameters in the quality evaluation process of the communication link.

[0272] Optionally, the configuration file format may include, but is not limited to, JSON, CSV, XML, etc. This application does not limit the specific format of the configuration file.

[0273] Optionally, the evaluation rule parameters include at least one of the following: sample size recommendation parameters, invalid value handling parameters, dimension weight parameters, score mapping parameters, penalty rule parameters, and rating mapping parameters.

[0274] The sample size recommendation parameter indicates the recommended number of bit error rates to be obtained in different quality assessment scenarios. Optionally, the sample size recommendation parameter includes a set of key-value pairs between scenario identifiers and corresponding recommended sample size values.

[0275] Invalid value handling parameters are used to define the handling strategy for invalid values ​​(such as null values, zero values, and negative values) in the bit error rate sample. Optionally, invalid value handling parameters include: whether to remove zero values, zero value replacement strategy, replacement value setting, etc.

[0276] The dimensional weight parameter defines the weight allocation for each dimension in the quality assessment. Optionally, the dimensional weight parameter includes the weights for consistency, stability, reliability, and continuous anomalies. Optionally, the continuous anomaly dimension is used as a negative penalty, with its weight fixed at 0%.

[0277] Optionally, the dimension weight parameter also supports configuring multiple sets of weights according to different quality assessment scenarios, such as automatically selecting the corresponding weight reorganization based on the current quality assessment scenario.

[0278] The rating mapping parameters are used to define the mapping rules from statistics to scores for each evaluation dimension. Optionally, the rating mapping parameters include consistency threshold mapping tables, stability threshold mapping tables, deviation penalty mapping tables, proportional penalty mapping tables, etc.

[0279] The penalty rule parameters are used to define the detection and penalty rules for consecutive anomalies. Optionally, penalty rule parameters include: anomaly multiplier, minimum consecutive anomaly length, consecutive anomaly penalty mapping table, multi-sequence penalty mapping table, etc.

[0280] The rating mapping parameters define the mapping relationship between health scores and ratings. Optionally, the rating mapping parameters include multiple score thresholds and their corresponding health ratings.

[0281] The connectivity success rate rating parameter defines the mapping relationship between connectivity success rate and rating. Optionally, the connectivity success rate rating parameter includes multiple success rate thresholds and their corresponding connectivity ratings.

[0282] It should be noted that the examples of evaluation rule parameters above are merely illustrative and this application does not limit them.

[0283] In some embodiments, during the communication link quality assessment process, dynamic detection of changes to the configuration file is supported. When a change to the configuration file is detected, the quality assessment process is performed according to the changed configuration file (i.e., steps 210 to 230 above).

[0284] In the above embodiments, a configuration-based architecture with external parameters is adopted. All adjustable parameters (i.e., evaluation rule parameters) related to business rules and scenario adaptation are separated from the evaluation logic and uniformly managed by an external configuration file. This decouples the evaluation logic from the evaluation rules, allowing adaptation to the evaluation needs of different application scenarios without modifying the core algorithm code, thus improving the system's flexibility, maintainability, and scalability. In some embodiments, different configuration files correspond to different application scenarios. Optionally, the configuration file corresponding to the communication link is determined according to the application scenario. Optionally, the scenario parameters of the application scenario include, but are not limited to, at least one of link rate, modulation method, and chip version.

[0285] Link rate refers to the transmission rate of a communication link. Optionally, different link rates correspond to different configuration files.

[0286] Modulation method refers to the modulation and coding method of the signal, such as NRZ (Non-Return-to-Zero), PAM4 (4-Level Pulse Amplitude Modulation), PAM8, etc. Optionally, different modulation methods correspond to different configuration files.

[0287] Chip version refers to the type of chip used in the communication link, such as the type of SerDes chip. Optionally, different chip versions correspond to different configuration files.

[0288] In the above embodiments, the corresponding configuration file is dynamically determined according to the application scenario (rate, modulation method, chip version, etc.), so that the quality assessment process can match the actual physical characteristics of the link, thereby improving the accuracy and fairness of the quality assessment results.

[0289] In some embodiments, at least two bit error rates are obtained after performing at least two initialization operations on the communication link. For example... Figure 3 As shown above, Figure 2 The embodiment can also be implemented as follows: steps 310 to 340.

[0290] Step 310: After performing at least two initialization operations on the communication link, obtain at least two bit error rates of the communication link.

[0291] At least two bit error rates must conform to the log-normal distribution requirement.

[0292] Optionally, at least two initialization operations refer to the initialization process that is repeatedly executed on the same communication link under the same link configuration parameters. For example, under the same link configuration parameters, the communication link is controlled to perform at least two initialization operations; after the i-th initialization operation is completed, the bit error rate of the communication link corresponding to the i-th initialization operation is obtained, where i is a positive integer; the bit error rates corresponding to the at least two initialization operations are used as at least two bit error rates.

[0293] Optionally, the number N of at least two initialization operations can be preset according to the quality assessment scenario. For example, in the initialization stability assessment scenario, the recommended sample size of N is 50 to 100 to ensure that the estimates of logarithmic standard deviation and geometric mean bit error rate reach a statistically stable level, while taking into account the testing time cost.

[0294] Step 320: Determine the bit error rate statistics based on at least two bit error rates.

[0295] Step 330: Determine the health score of the communication link based on the bit error rate statistics.

[0296] Bit error rate statistics include logarithmic statistics.

[0297] Among them, the health score is a rating obtained based on the bit error rate statistics, which is used to intuitively represent the overall quality of the communication link.

[0298] Step 340: Determine the quality assessment result of the communication link based on the health score.

[0299] Once the health score is obtained, intuitive quality assessment results can be generated to guide link configuration selection, status monitoring, and maintenance decisions.

[0300] In some embodiments, the health score is directly used as the quality assessment result of the communication link. In other embodiments, the health score is mapped to a corresponding health rating according to a preset rating threshold table, and the health rating is used as the quality assessment result. For example, the health rating can be divided into multiple levels such as excellent, very good, good, average, poor, and unacceptable, with each level corresponding to a health score range.

[0301] In other embodiments, the quality assessment result of the communication link is determined based on the health score and the connectivity success rate. The connectivity success rate refers to the proportion of times the communication link successfully establishes a stable physical connection and is able to transmit data normally in at least two initialization operations. The method for determining the connectivity success rate further includes steps 1 and 2 below.

[0302] Step 1 After the initialization of the communication link is completed, determine the connectivity status of the communication link.

[0303] Link connectivity status indicates whether a stable physical connection has been successfully established and data transmission can proceed normally after each initialization operation. In some embodiments, link connectivity status is recorded as a binary variable: 1 for successful connection and 0 for connection failure.

[0304] Optionally, the criteria for determining successful connection may include: the link status register indicating that initialization is complete and locked, the receiver clock data recovery circuit being locked, and the error counter starting normally. If the above states are not achieved within the preset timeout period, the connection is determined to have failed.

[0305] Step 2 The success rate of communication link connectivity is determined based on the connectivity status of the communication links corresponding to at least two initialization operations.

[0306] For example, if 50 initialization operations are performed on a communication link, and the connection is established successfully 47 times and fails 3 times, then the connection success rate is 94%.

[0307] Optionally, the connectivity success rate is also mapped to the corresponding connectivity rating, and an example of the mapping rule (configurable) is as follows: Connection success rate ≥ 95% → Rating "Excellent"; Connection success rate ≥85% and <95% → rating "Good"; Connection success rate ≥75% and <85% → rating "Fair"; Connection success rate <75% → rating "Poor", triggering a warning.

[0308] Optionally, the scores of the dimensions used to determine the health score, the health rating, the health score, the connectivity success rate, and the connectivity rating can be used as the quality assessment results.

[0309] Optionally, after determining the connectivity success rate: if the connectivity success rate is less than a preset success rate threshold, the quality assessment result of the communication link is determined to be unqualified; if the connectivity success rate is greater than or equal to the preset success rate threshold, the quality assessment result of the communication link is determined based on the health score and connectivity success rate. For example, the health score, health rating, connectivity success rate, and connectivity rating can be output together as the quality assessment result, providing multi-dimensional reference information for operations and maintenance personnel.

[0310] In summary, the communication link quality assessment method provided in this application determines the link connectivity after the communication link initialization operation is completed, calculates the connectivity success rate corresponding to at least two initialization operations, and then combines the health score obtained based on logarithmic statistics to jointly determine the quality assessment result. This improves the dimensions of quality assessment and enables the assessment result to more comprehensively and realistically reflect the overall availability of the communication link.

[0311] In some embodiments, the bit error rate corresponding to at least two time windows is sampled as at least two bit error rates. For example... Figure 4 As shown above, Figure 2 The embodiment can also be implemented as follows: steps 410 to 440.

[0312] Step 410: During the operation of the communication link, sample the bit error rate corresponding to at least two time windows.

[0313] At least two bit error rates must conform to the log-normal distribution requirement.

[0314] Optionally, sampling the bit error rate corresponding to at least two time windows means: under the condition that the communication link has a fixed configuration and operates continuously, the continuous operation time of the link is divided into multiple time windows according to a preset rule. At the end of each time window, the bit error rate counter and the total number of transmitted bits counter within that window are read, and the average bit error rate within the window is calculated as a bit error rate sample. The time windows can be adjacent, overlapping, or spaced apart, and the window length can be set to a constant or variable value according to application requirements. For example, in long-term stability monitoring scenarios, a constant-length time window can be used, such as one window per hour or per day; when an increase in bit error rate fluctuation is detected, it can automatically switch to a smaller window to improve time resolution.

[0315] Optionally, sampling at least two time windows can include periodic sampling based on a time period, window sampling based on the number of transmitted bits, and supplementary sampling based on event triggering. For example, in periodic sampling based on a time period, time windows are automatically divided during the operation of the communication link according to a preset sampling period (e.g., hourly, daily), and the bit error rate within that time window is sampled as the bit error rate sample for that period. This method is suitable for long-term trend monitoring and link health tracking.

[0316] Step 420: Determine the bit error rate statistics based on at least two bit error rates.

[0317] Bit error rate statistics include logarithmic statistics. Logarithmic statistics are determined based on the logarithms corresponding to at least two bit error rates.

[0318] In some embodiments, the logarithmic statistics are determined directly based on the logarithms corresponding to at least two bit error rates. Specifically, after filtering out invalid values ​​from the at least two sampled bit error rates, the natural logarithms are taken to obtain a natural logarithm sequence; the mean and standard deviation of the natural logarithm sequence are calculated; the mean is subjected to an inverse natural exponential transform to restore the geometric mean bit error rate; and the standard deviation is used as the logarithmic standard deviation.

[0319] In other embodiments, time weights are determined for at least two bit error rates; based on the time weight for the w-th bit error rate, a weighted average is applied to the logarithm corresponding to the w-th bit error rate to obtain a weighted logarithm for the w-th bit error rate; and the geometric mean bit error rate and logarithmic standard deviation are calculated based on the weighted logarithms corresponding to the at least two bit error rates. Specifically, after filtering out invalid values ​​from the weighted logarithms corresponding to the at least two bit error rates, the natural logarithm is taken for each, resulting in a natural logarithm sequence; the mean and standard deviation of this natural logarithm sequence are calculated; an inverse natural exponential transform is performed on the mean to restore the geometric mean bit error rate; and the standard deviation is used as the logarithmic standard deviation.

[0320] Optionally, the time weight corresponding to the bit error rate is negatively correlated with the interval between the bit error rate sampling time window and the current moment. That is, the closer the time window is to the current moment, the greater the reference value of its bit error rate sample in the evaluation, and the higher its weight is assigned; the further back the historical sample is from the current moment, the less indicative its role in the current link status gradually diminishes, and the lower its weight is assigned. This weighting mechanism can more sensitively reflect the recent performance change trend of the link and is suitable for scenarios that require real-time or near-real-time evaluation of the link status.

[0321] By introducing time weights, logarithmic statistics can more accurately characterize the typical performance level and fluctuation characteristics of the link in the current operating phase, providing a more timely data foundation for subsequent health assessments.

[0322] Step 430: Obtain the health score of the communication link based on the bit error rate statistics.

[0323] In this embodiment, the health score is a rating obtained based on logarithmic statistics, which is used to intuitively characterize the overall quality of the communication link during continuous operation.

[0324] Step 440: Determine the quality assessment result of the communication link based on the health score.

[0325] In some embodiments, the health score is directly used as the quality assessment result of the communication link. In other embodiments, the health score is mapped to a corresponding health rating according to a preset rating threshold table, and the health rating is used as part of the quality assessment result. For example, the health rating can be divided into multiple levels such as excellent, very good, good, average, poor, and unacceptable, with each level corresponding to a health score range.

[0326] Optionally, when sampling covers at least two consecutive periods, a scoring trend can be determined based on the health score for each period. For example, the trends in health score changes, consistency score changes, stability score changes, reliability score changes, or continuous anomaly score changes can be determined for at least two periods. Based on the scoring trend, predictive maintenance can be performed on the communication link. For instance, when the consistency and reliability scores show a continuous downward trend over multiple periods, a link performance degradation warning is output; when the continuous anomaly score frequently shows negative values ​​over multiple periods and the absolute value shows an increasing trend, a persistent link failure is predicted, and a switchover or investigation is recommended within the planned maintenance window.

[0327] Furthermore, the scores of the dimensions used to determine the health score, the health rating, the health score, and the scoring trend analysis results can all be combined as quality assessment results to generate a link health report for the communication link. The link health report may include: a test configuration summary, core statistical results, multi-dimensional health score details, health score and health rating, a list of poor bit error rates, optimization suggestions based on the health rating, and a bit error rate performance trend chart. The bit error rate performance trend chart may include at least one of the measured bit error rate curve, the geometric mean bit error rate curve, and the median bit error rate curve, visually presenting the performance evolution of the link during continuous operation.

[0328] In summary, the communication link quality assessment method provided in this application samples the bit error rate (BER) for at least two time windows during the operation of the communication link, calculates a health score based on logarithmic statistics, and determines the quality assessment result accordingly. This method fully utilizes the log-normal distribution of BER, thereby accurately and sensitively assessing the real-time status and trends of link quality in continuous operation scenarios, providing objective decision support for predictive maintenance and long-term reliability monitoring.

[0329] In some embodiments, the above-mentioned at least two time windows are divided into light load periods, moderate load periods, and heavy load periods. Optionally, the above-mentioned... Figure 4 The embodiment can also be implemented as follows: steps 1 to 4.

[0330] Step 1: During the operation of the communication link, sample the bit error rate data corresponding to at least two time windows.

[0331] Among them, at least two time windows are divided into light load period, medium load period and heavy load period.

[0332] Load periods refer to the intervals of continuous operation time divided into different load intensities based on the service traffic carried by the communication link during operation. By distinguishing load periods, the bit error rate performance of the link under different operating pressures can be characterized more precisely, providing a data foundation for evaluating the transmission quality of the link under various real-world operating conditions.

[0333] In some embodiments, a light load period refers to the operating period when the traffic carried by the link is lower than a first preset threshold. During this period, the amount of data transmitted by the link is small, the signal-to-noise ratio environment is relatively relaxed, and the bit error rate is usually at a low level. A moderate load period refers to the operating period when the traffic is between the first and second preset thresholds, and the link is under moderate workload. A heavy load period refers to the operating period when the traffic is higher than the second preset threshold. The link operates at near full load or under high concurrency, and the effects of crosstalk, power supply noise, and other factors may be aggravated, resulting in a higher or more volatile bit error rate.

[0334] The bit error rate (BER) data includes at least two BER rates corresponding to lightly loaded periods, at least two BER rates corresponding to moderately loaded periods, and at least two BER rates corresponding to heavily loaded periods. That is, at least two time windows need to be defined within each load period, and BER rates need to be sampled within each window to ensure that each load period can independently provide a sufficient set of BER samples that meet the log-normal distribution requirement. The number of BER samples within each load period can refer to the aforementioned recommended sample size values ​​for specific scenarios, for example, obtaining at least 50 BER samples per period to ensure that the estimation of the logarithmic statistics has sufficient statistical stability.

[0335] Optionally, the length of the time window can be the same or different across different load periods to accommodate the duration and statistical accuracy requirements of each load period. For example, heavy load periods are typically short in duration but have a significant impact on link quality, so a shorter window length can be used to increase sampling density; light load periods are longer in duration, so a longer window length can be used to reduce computational overhead.

[0336] Step 2: Determine the first bit error rate (BER) statistic based on at least two BER rates corresponding to light load periods. Determine the second BER statistic based on at least two BER rates corresponding to moderate load periods. Determine the third BER statistic based on at least two BER rates corresponding to heavy load periods.

[0337] This step involves performing logarithmic field statistical analysis on the bit error rate samples under different load periods to independently characterize the typical performance and fluctuation characteristics of the link under various load conditions.

[0338] Step 3: Obtain the first health score of the communication link based on the first bit error rate statistic. Obtain the second health score of the communication link based on the second bit error rate statistic. Obtain the third health score of the communication link based on the third bit error rate statistic.

[0339] The first health score, second health score, and third health score represent the link's health rating under light, medium, and heavy load conditions, respectively. The method for generating each health score is consistent with the scheme for calculating the health score based on the bit error rate statistics described in the preceding steps, and will not be repeated here.

[0340] In some embodiments, the weighting of consistency score, stability score, reliability score and continuous anomaly score under each load period can be kept consistent, using a globally unified dimensional weighting configuration.

[0341] In other embodiments, considering that different load periods have different focuses in link quality assessment, differentiated weights can be configured for different load periods. For example, during periods of heavy load, since the link is closer to its performance limit, the risk and harm of continuous anomalies are greater. The penalty coefficient for continuous anomalies can be appropriately increased or the starting baseline score for reliability can be reduced to make the score more sensitive to anomalies under heavy load. During periods of light load, the weight of consistency can be appropriately increased because the signal-to-noise ratio is better at this time, and consistency can better reflect the inherent characteristics of the link itself rather than environmental interference.

[0342] Step 4: Determine the quality assessment result of the communication link based on the first health score, the second health score, and the third health score.

[0343] The quality assessment results reflect the communication quality of the communication link under different load conditions. By integrating the health scores from the three load periods, a comprehensive evaluation of the link's performance under all operating conditions can be formed, avoiding the bias of evaluation under a single load scenario.

[0344] In some embodiments, determining the quality assessment result of a communication link based on a first health score, a second health score, and a third health score includes weighting and summing the three health scores according to preset load weights to obtain a comprehensive health score. The load weights can be set based on the time proportion or importance of each load period in actual business operations. For example, if the link mainly operates during periods of heavy load and the business criticality is highest during those periods, the health score for the heavy load period can be assigned the highest weight, such as a light load weight of 20%, a medium load weight of 30%, and a heavy load weight of 50%.

[0345] In other embodiments, the minimum of the three health scores is selected as the conservative evaluation score for the link under extreme conditions, or a load-health profile containing three sets of scores is output, displaying the quality differences of the link under different loads in the form of a structured array or a visual chart. For example, the quality evaluation results may include: a health score and rating for light load, a health score and rating for medium load, a health score and rating for heavy load, and a comprehensive health score and rating. When the health score during a certain load period (especially a heavy load period) falls below a preset threshold, an alarm can be directly triggered even if the scores are higher in other periods, indicating that the link has a performance bottleneck or potential risk under high load.

[0346] Optionally, the performance degradation trend of the link as the load increases can be analyzed based on the health score for each load period. For example, the difference or attenuation rate between the first and third health scores can be calculated. When the attenuation exceeds the allowable range, it indicates that the stability of the link deteriorates significantly under increased pressure, and it is recommended to optimize the link's drive capability, balancing parameters, or heat dissipation conditions accordingly.

[0347] In summary, by dividing the runtime window into three load periods—light, moderate, and heavy—and sampling the bit error rate (BER) for each period, and independently calculating the BER statistics and health score for each load period, a comprehensive evaluation result reflecting the link's communication quality under all operating conditions is obtained. This method incorporates the impact of load status on BER performance into the quality evaluation system, effectively addressing the evaluation blind spots under single load conditions. This makes the quality evaluation results closer to the comprehensive performance of the link in real-world business scenarios, providing a decision-making basis for link optimization and resource scheduling under different load pressures.

[0348] In some embodiments, the communication link quality assessment method provided in this application can be implemented as a "link health analysis method and system based on multi-dimensional quantitative scoring". This link health analysis method and system based on multi-dimensional quantitative scoring is particularly suitable for multi-dimensional link quality assessment systems based on bit error rate (BER) statistics.

[0349] This application proposes a link health analysis method and system based on multi-dimensional quantitative scoring. This method comprehensively evaluates the performance stability, consistency, reliability, and continuous abnormal BER (Breakpoint Error) of high-speed links under specific configurations, generating intuitive health scores and reports. This provides a scientific and objective basis for configuration selection, quality assessment, and status monitoring. This application achieves the following technical effects: (1) Provide multi-dimensional quantitative evaluation: comprehensively score high-speed links from multiple dimensions such as consistency, stability, reliability and continuous abnormal BER; (2) To achieve an objective comparison of multiple TX FIR (Transmit Finite Impulse Response) configurations: to provide quantifiable health scores for different TX FIR configurations to assist in selection decisions; (3) Support for long-term monitoring and trend analysis: Identify link performance drift trends by statistically modeling historical BER data; (4) Enhance confidence in system deployment: Verify the stability of the link in a real environment through multiple rounds of initialization tests and long-term monitoring.

[0350] The key features of the link health analysis method and system based on multi-dimensional quantitative scoring proposed in this application include: 1. A BER data analysis method based on geometric mean and logarithmic standard deviation solves the distortion problem of traditional arithmetic mean when dealing with data spanning orders of magnitude; 2. A multi-dimensional (consistency, stability, reliability, and continuous anomaly BER) scoring system to comprehensively assess the health of the link; 3. Dual-threshold Real Poor BER identification mechanism: Combining the 75th percentile and geometric mean BER, it accurately distinguishes between real abnormal BER and BER that is at the tail but still performs better than the typical level, avoiding misjudgment by the single threshold method; 4. Multi-scenario applicability: Supports both online data acquisition and offline analysis input modes, supports multiple rounds of initialization comparison, multiple sets of TX FIR selection, and long-term monitoring trend analysis; 5. Configurable architecture with external parameters: Scoring thresholds, weights, and penalty rules are managed through external configuration files, supporting flexible adaptation to different application scenarios (different rates, modulation methods, and chip versions); 6. Dual-mode data input: Supports online acquisition (parsing from structured data files) and offline analysis (directly receiving BER arrays or structured files), sharing the same set of statistical analysis and scoring algorithms. Regardless of the actual scenario, link status assessment can be performed as long as BER data is available.

[0351] The following section will elaborate on the above-mentioned link health analysis method and system based on multi-dimensional quantitative scoring.

[0352] (a) Log-normal distribution of BER.

[0353] The cornerstone of this application's technical approach is based on a profound understanding of the statistical characteristics of bit error rate (BER) in high-speed serial links: BER data essentially conforms to or highly approximates a log-normal distribution. This characteristic determines that traditional analysis methods based on arithmetic mean have fundamental flaws, while the log-scale statistics (geometric mean, log-standard deviation) adopted in this application are the scientific and optimal choice.

[0354] 1. Theoretical origin: From SNR normal distribution to BER log-normal distribution.

[0355] The performance of high-speed links (such as SerDes channels) is ultimately reflected in the bit error rate, but its statistical characteristics are rooted in the physical layer at the front end.

[0356] (1) Communication system performance model: According to digital communication theory, under high signal-to-noise ratio (SNR) operating conditions, the system's bit error rate (BER) and linear signal-to-noise ratio (LNR) are related. There are definite approximate relationships between them: This relationship reveals the core mechanism by which BER decays exponentially with SNR (Signal-to-Noise Ratio).

[0357] (2) Statistical Model of Channel and Measurement: In practical systems, the signal-to-noise ratio (SNR) at the receiver is multiplicatively affected by a large number of independent random factors (such as thermal noise, random jitter, interference, and power supply noise) in the transmitter, channel, and receiver. According to the multiplicative central limit theorem, the logarithm (i.e., decibel value) of a physical quantity determined by the product of a large number of independent random variables will approach a normal distribution. Therefore, in engineering practice, the decibel value of the signal-to-noise ratio (SNR) is... dB The statistical fluctuations are widely modeled as a normal distribution, i.e. .

[0358] Combining the two models above—the exponential relationship of performance and the normal statistics of measurement—constitutes a model from... The normal distribution leads to the complete logical starting point for deriving the BER log-normal distribution.

[0359] 2. Core mathematical derivation.

[0360] Let the linear signal-to-noise ratio of the high-speed link be... (That is, the ratio of signal power to noise power), its decibel value is According to digital communication theory, under high signal-to-noise ratio (SNR) conditions, the bit error rate (BER) and the linear SNR have the following approximate relationship: Step 1: Establish the proportional relationship.

[0361] "Proportional to" (∝) means that there exists a positive constant A, determined by factors such as the system modulation method and coding, such that: (1).

[0362] Step 2: Standardize the units (decibel to linear).

[0363] In engineering practice, signal-to-noise ratio (SNR) is often measured and statistically analyzed in decibels (dB). The conversion relationship between linear values ​​and decibel values ​​is given by the definition: (2).

[0364] Mathematical transformations are performed on both sides of equation (2) to solve for... : (3.1); (3.2).

[0365] Step 3: Convert the exponent base to introduce the natural logarithm.

[0366] To derive the distribution characteristics later, the exponent with a base of 10 needs to be converted to a base of the natural constant e. Using the formula... Let b=1, then ,make Transform equation (3.2) as follows: (4).

[0367] Using formula Let a = 10, b = SNR dB / 10, transform equation (4): (5).

[0368] Let k = ln(10) / 10 (this is a definite mathematical constant, approximately equal to 0.23026), then the above equation simplifies to: (6).

[0369] Step 4: Substitute into the BER formula and take the natural logarithm.

[0370] Substituting the key relation (6) into the initial BER formula (1): (7).

[0371] Taking the natural logarithm (ln) of both sides of equation (7) transforms the analysis of the log-normal distribution into the analytical framework of the additive normal distribution: (7).

[0372] Using the formula ln(xy)=ln(x)+ln(y), we can transform equation (7): (8).

[0373] use Transform equation (8) as follows: (9).

[0374] Let the constant C = ln(A), and transform equation (9): (10).

[0375] Step 5: First-order Taylor expansion.

[0376] In practical engineering, The fluctuation range is usually limited (e.g., ±3dB). Let: .

[0377] Where μSNR is The mean is given, and δ represents random fluctuations. Based on engineering observations and theoretical analysis, It can be modeled as a normal distribution: .

[0378] For the nonlinear term in equation (10) Perform a first-order Taylor expansion at μSNR: (12).

[0379] When the fluctuation range of δ is limited (in actual high-speed links, The typical fluctuation range is usually within ±3dB, i.e. (|k Since δ∣<0.7), we have an approximation: (13); Substituting into formulas (10) and (12), we get:

[0380] make: ; .

[0381] but: .

[0382] because therefore ,but: .

[0383] ln(BER)≈C'-β δ approximately follows a normal distribution: ( ).

[0384] By definition, if the natural logarithm of a random variable follows a normal distribution, then the variable itself follows a log-normal distribution. Therefore, BER follows (or is highly approximately) a log-normal distribution.

[0385] Final conclusion: Based on the classic model of communication systems and The normal statistical model, through the analysis of... By performing a first-order Taylor expansion linear approximation, it can be rigorously derived that ln(BER) approximately follows a normal distribution, that is, BER follows (or is highly approximating) a log-normal distribution. This forms the theoretical foundation for the use of logarithmic scale statistics such as geometric mean and log-standard deviation in this application for health status analysis.

[0386] The above first-order approximation is in The accuracy is sufficient when the fluctuation range is limited. In practical engineering, The fluctuations are mainly caused by factors such as thermal noise and random jitter, and their standard deviation is usually much smaller than the mean, making the linear approximation highly accurate. When When the fluctuations are large, a second-order or higher-order Taylor expansion can be used to improve the approximation accuracy, but the first-order approximation used in this application can already meet the needs of engineering analysis.

[0387] (II) System Overall Architecture.

[0388] Figure 5 A schematic diagram of the system architecture provided in this application is shown. (For example...) Figure 5 As shown, the system includes the following modules: 1. Data acquisition and control module 510.

[0389] Users input data such as TX FIR coefficients, link speed, and encoding mode into the data acquisition and control module 510 according to the scenario.

[0390] The data acquisition and control module 510 is responsible for interacting with the hardware. Depending on the scenario, it can perform multiple initializations of the same TX FIR, initializations of different TX FIRs, or long-term monitoring of a specific TX FIR. It controls the target link to initialize according to user-specified parameters (such as TX FIR coefficients, link rate, and coding mode), and after initialization stabilizes, it reads link status information such as BER through the device driver interface.

[0391] This application mainly uses the example of multiple initializations of the same TX FIR to illustrate the concept. Figure 6 A schematic diagram of the bit error rate performance curves for multiple rounds of initialization acquisition using the same TXFIR is shown. Figure 6 As shown in Figure 600, the test was conducted at a high-speed link rate of vGbps, and the TX FIR coefficients were configured as [a, b, c, d, e]. In Figure 600, the horizontal axis represents the initialization round, and the vertical axis represents the bit error rate (BER) measured after the link stabilized in the corresponding round. The figure contains three curves: The solid line represents the measured BER (v Gbps): it shows the measured instantaneous bit error rate of the link after each round of initialization. The dotted line represents the geometric mean BER: a central trend indicator calculated from the logarithmic mean based on all valid measured BER data, representing the typical bit error rate level of the downlink under this TX FIR configuration; The dotted line represents the median BER: the median value is taken after sorting all measured BER values ​​by size.

[0392] Optionally, during the multi-round initialization process, the success status of the BIST Lock on the RX side is also recorded after each initialization, and the BIST Lock success rate (i.e., the connection success rate) is calculated: BIST_Lock_Rate = BIST_Lock_Count / Total_Run_Count × 100%.

[0393] This is mapped to ratings: Excellent (≥95%), Good (≥85%), Fair (≥75%), and Poor (<75%). This metric reflects the reliability of the link initialization and complements the BER health score.

[0394] Optionally, the data acquisition and control module 510 performs initialization, acquires BER, or directly receives the input BER. That is, the data acquisition and control module 510 supports two data input modes: online acquisition mode, which first acquires BER data to generate a structured data file, and then parses the BER and related status information block by block from the structured data file; and offline analysis mode, which directly receives the BER array or structured file as input for statistical analysis.

[0395] 2. Core statistical calculation module 520.

[0396] Receive the original BER dataset, perform core statistical calculations based on the log-normal distribution model, and output core statistics.

[0397] 3. Multidimensional health score module 530.

[0398] It receives core statistics, executes an innovative four-dimensional scoring algorithm, calculates scores for consistency, stability, reliability, and continuous anomaly BER, as well as a comprehensive total score, and outputs multi-dimensional scores and a total score.

[0399] 4. Report generation and visualization module 540.

[0400] Integrate all intermediate and final results to generate a text report containing detailed data, scores, ratings, and recommendations, and optionally generate BER trend charts, then output a health report and charts.

[0401] (III) Detailed implementation of the core statistical calculation module 520.

[0402] The processing flow of the core statistical calculation module 520 is based on a key understanding: high-speed link BER data follows or approximately follows a log-normal distribution. Therefore, the following steps are adopted: Input: An array BERs=[ber1, ber2, ..., berN] containing N valid BER values ​​(e.g., N=50).

[0403] Data preprocessing: Filter out invalid values ​​(such as "NaN", 0.00 [in which case the link is optimal, there are no errors, and no evaluation is needed]). Take the natural logarithm of each valid BER value to obtain a new array ln_BERS=[ln(ber1), ln(ber2),..., ln(berN)].

[0404] Calculate the core statistic: Logarithmic mean: mu_ln = average(ln_BERs); Geometric Mean BER: Geo_Mean_BER = exp(mu_ln). This value is a central tendency indicator that measures "typical" performance; Logarithmic standard deviation: sigma_ln = std_dev(ln_BERs). This value is a core indicator for measuring the relative volatility of BER and directly reflects consistency; Linear statistics (auxiliary reference): Calculate the minimum BER (Min_BER), maximum BER (Max_BER), arithmetic mean BER (Avg_BER), and median BER (Median_BER); Fluctuation Range: Fluctuation_Ratio = Max_BER / Min_BER, and this value reflects the extreme differences in performance.

[0405] Output: All the above calculated statistics.

[0406] (4) Detailed implementation of the multi-dimensional health score module 530.

[0407] Figure 7 The flowchart of the multi-dimensional health score algorithm provided by this application is shown. <{0001000}>The scoring algorithms for each dimension include: 1. Consistency score calculation 710 (weight a%): The consistency score is used to measure the predictability and repeatability of system behavior, and it reflects: the high repeatability of the system's performance in different tests, and its purpose: to measure the predictability of link behavior in repeated tests.

[0409] As Figure 7 shown, the process of consistency score calculation 710 includes: Input: Logarithmic standard deviation; Processing: Look up table / function mapping; Output: Consistency score.

[0410] The method of consistency score calculation 710 is completely based on the logarithmic standard deviation (sigma_ln), and uses hierarchical threshold mapping (a total of 10-level mapping), and matches in ascending order of the logarithmic standard deviation.

[0411] Mapping rule: Match level by level starting from the first threshold, and return the corresponding score when sigma_ln ≤ threshold; if all thresholds are not met, return the lowest score of 5. The above thresholds can all be adjusted through an external configuration file. An example of a mapping rule (configurable, the following is just an example and does not represent the actual algorithm) is as follows: If sigma_ln ≤ 0.5, then the score S_con = 45 (excellent); If 0.5 < sigma_ln ≤ 1.0, then S_con = 20; If sigma_ln > 2.0, then S_con = 5.

[0412] 2. Stability score calculation 720 (weight b%): The stability score is used to measure the degree of difference between the best and worst cases, and it reflects: the extreme fluctuations of system performance, and its purpose: to measure the extreme cases of link performance. [[ID=​​​As shown, the process of calculating the stability score 720 includes: Input: Fluctuation range; Process: Look up table / function mapping; Output: Stability score.

[0414] The method of calculating the stability score 720 is completely based on the fluctuation range (Fluctuation_Ratio = Max_BER / Min_BER), and uses hierarchical threshold mapping (a total of 9 levels of thresholds).

[0415] An example of a mapping rule (configurable, the following is just an example, not representing the actual algorithm) is as follows: If Fluctuation_Ratio ≤ 5, then the score S_sta = 30 (extremely stable); If 5 < Ratio ≤ 20, then S_sta = 18; If Ratio > 100, then S_sta = 5.

[0416] 3. Calculation of reliability score 730 (weight c%): The reliability score is used to measure the ability of the system to operate stably within an acceptable range, which reflects: the reliable degree of the system in actual use, and its purpose: to measure the ability of the system to operate stably within an acceptable range, and focuses on abnormal situations of relatively poor BER in performance.

[0417] Such as Figure 7 As shown, the process of calculating the reliability score 730 includes: Input: PoorBER list, geometric mean BER; Process: Traverse the PoorBER list, calculate the deviation of PoorBER relative to the geometric mean BER; look up table / function mapping; obtain the first penalty score according to the degree of deviation; Output: Reliability score.

[0418] The method of calculating the reliability score 730 includes: (1) Identify "PoorBER": Perform quartile analysis on the BER data, and divide all valid BERs into four grades according to performance: Excellent interval: BER ≤ P25 (25% quantile), indicating the best performance level; Good interval: P25 < BER ≤ P50 (median), indicating a good performance level; Fair interval: P50 < BER ≤ P75 (75% quantile), indicating an acceptable performance level; Poor candidate: BER > P75, which may be an outlier.

[0419] (2) The dual threshold recognition mechanism of "PoorBER": Calculate the 75th percentile for all BERs (the 75th percentile is denoted as Q3 below); For the Poor candidate BER, further determine whether it is greater than the geometric mean BER: RealPoorBER: BER > Q3 and BER > Geo_Mean_BER (geometric mean BER) are both at the tail end and worse than the typical level, and are outliers that really need attention. FalsePoorBER: BER > Q3 but BER ≤ Geo_Mean_BER (geometric mean BER). Although such BER is in the bottom 25%, its performance is still better than or equal to the typical level and should not be judged as abnormal.

[0420] This dual-threshold mechanism effectively avoids the misjudgment problem of the single-threshold method: when the geometric mean BER is poor (close to Q3), some of the BERs in the bottom 25% are actually not worse than the typical level and should not be penalized.

[0421] (3) The reliability score adopts a two-level penalty mechanism: Measurement: Detect and penalize consecutive anomalies in the BER of a time series. Such patterns may indicate persistent or intermittent stability problems in the system and serve as an early warning signal of potential failures.

[0422] Reflection: Warning of continuous anomalies.

[0423] Level 1: Deviation penalty. For each RealPoorBER, calculate its deviation from the geometric mean by the factor d = BER / Geo_Mean_BER.

[0424] The deviation is matched progressively from largest to smallest, based on the degree of deviation (6 levels in total). The deviation penalties for all RealPoorBERs are accumulated, with each deviation level corresponding to a certain deduction score. When the total deviation penalty reaches the upper limit (reliability score weight), the calculation of deviation penalties for subsequent BERs stops. This is because, at this point, the already calculated RealPoorBERs have already penalized the reliability score to zero in the deviation score category.

[0425] Level 2: Quantity-based penalty. Calculate the ratio of Real_Poor_Count to Total_Tests.

[0426] The proportion of Realpoorber in the test is matched from largest to smallest according to the degree (4 levels in total), and the corresponding score is deducted for each proportion.

[0427] Quantity ratio penalty = min(penalty value, maximum penalty limit - total deviation penalty), ensuring that deviation penalty + quantity ratio penalty ≤ maximum penalty limit (reliability weighted).

[0428] Reliability score = max(0, initial baseline score - total deviation penalty - quantity ratio penalty), where the initial baseline score is the weight of reliability.

[0429] 4. Calculation of continuous anomaly penalty score: 740 (weight 0% - negative penalty items): The continuous anomaly penalty score is used to detect and penalize the BER of continuous anomalies in a time series. Such patterns may indicate persistent or intermittent stability problems in the system and serve as an early warning signal of potential faults. It reflects the warning of continuous anomalies.

[0430] like Figure 7 As shown, the process of calculating the continuous anomaly penalty score of 740 includes: Input: BER list, geometric mean BER, logarithmic standard deviation; Processing: Calculate the outlier threshold = geometric mean BER + n times the logarithmic standard deviation; iterate through the BER list; determine if the BER is greater than the threshold, and if so, increment the temporary consecutive outlier count by 1, otherwise reset the temporary consecutive outlier count to zero; calculate the maximum number of consecutive outliers and the number of outlier sequences. Output: Second penalty score.

[0431] The methods for calculating the continuous anomaly penalty score of 740 include: (1) Calculate the abnormal threshold.

[0432] Anomaly threshold = geometric mean BER × e^(anomaly multiple × log standard deviation).

[0433] This threshold, based on the geometric mean and n times the log-standard deviation, is used to identify statistically significant outlier BER values. The outlier factor is configurable. If a BER is significantly higher than the overall geometric mean and exceeds the exponential range corresponding to n times the log-standard deviation, it is considered a statistically significant outlier. This method does not simply use a fixed BER threshold to determine anomalies; instead, it judges "relative anomalies" based on the statistical distribution of the current dataset itself, dynamically determining the outlier boundary using the geometric mean and log-standard deviation, thus adapting to differences in BER distribution under different testing conditions and link quality levels.

[0434] (2) Detect continuous abnormalities.

[0435] The BER test sequence is traversed according to the test round or time order. For each valid BER value, if the BER value is greater than the anomaly threshold, the point is determined to be an anomaly point, and the current consecutive anomaly count is incremented by 1; if the BER value is not greater than the anomaly threshold, the current consecutive anomaly segment is considered to have ended, and it is determined whether a valid anomaly sequence is formed based on the length of the consecutive anomalies.

[0436] The variables recorded during the detection process are shown in Table 1.

[0437] By default, a sequence of consecutive anomalies must be ≥3 (configurable) to be counted as a valid anomaly sequence. Therefore, while a single or two isolated anomalies may reflect local fluctuations, they do not directly constitute a penalty for consecutive anomalies.

[0438] (3) Calculate the penalty for consecutive anomalies.

[0439] A negative penalty score is calculated based on the severity of consecutive anomalies. The penalty strategy prioritizes the maximum number of consecutive anomalies; if the maximum consecutive anomaly penalty is not triggered, then it is determined whether multiple anomaly sequences exist.

[0440] If the maximum consecutive exception is not triggered, but multiple exception sequences are triggered, it indicates that: (1) Although the link did not show any serious continuous anomalies (such as n consecutive anomalies), it showed multiple scattered anomaly sequences. (2) This indicates that the link quality is unstable. Although the duration of a single anomaly is not long, the frequency of anomalies is high. Therefore, a multi-sequence penalty (-a points) needs to be applied to reflect this unstable state.

[0441] The current penalty rules (conditional and trend penalty scores are configurable) are shown in Table 2 below:

[0442] The penalty levels are matched from highest to lowest severity, and the most severe level is selected. For example, when max_consecutive_poor ≥ 5, a penalty of -8 points is triggered directly; when max_consecutive_poor is less than 5 but ≥ 3, a penalty of -4 points is triggered; when the maximum consecutive anomaly length does not reach 3, but the condition of multiple valid anomaly sequences is met, a penalty of -2 points is triggered.

[0443] Note: The continuous anomaly penalty score is not a positive scoring dimension, but a negative penalty item, directly deducted from the overall score. Its weight is 0%, but it reflects anomaly patterns in the time series. The continuous anomaly penalty score allows the scoring algorithm to not only determine whether the overall BER distribution is stable (the previous three factors), but also further identify whether the anomalies have temporal continuity. If anomalies appear only sporadically, the risk is low; if anomalies appear consecutively 3 or 5 times, or repeatedly in multiple segments, it indicates that the link may have persistent degradation, intermittent jitter, environmental sensitivity, or configuration boundary issues. Therefore, the negative penalty item lowers the overall health score in advance.

[0444] In summary, the following is through Figure 8 This section details the calculation method for continuous anomaly penalty scores. Figure 8 A flowchart illustrating the calculation of continuous anomaly penalty scores is shown. Figure 8 As shown, the method includes the following: Step 1. Begin.

[0445] The process starts with three types of data as input: BER data array (raw bit error rate sequence); Geometric mean of BER; Log-standard deviation of BER.

[0446] Step 2. Check if the array size is less than 3? Determine if the number of input BER data points is less than 3: If (insufficient data): Execute step 3 and directly return the default value "0:0:0" (no valid data, process terminated); If not (sufficient data): Proceed to step 4 to continue processing.

[0447] Step 3. Return to "0:0:0".

[0448] When the number of BER data points is less than 3, the default result is returned directly because the amount of data is insufficient to support effective statistics, and the process ends.

[0449] Step 4. Extract parameters.

[0450] Extract three types of key parameters from the input: Geometric mean BER: The geometric mean of the BER sequence, representing the overall bit error rate; Logarithmic standard deviation: The standard deviation of the logarithm of the BER sequence reflects the degree of fluctuation in bit error rate. Actual BER array: The raw bit error rate data sequence to be processed.

[0451] Step 5. Calculate the anomaly threshold.

[0452] The BER anomaly detection threshold is calculated based on statistical features, using the following formula: threshold = geo_mean × exp(multiplier × log_stddev).

[0453] Step 6. Is the threshold valid and >0? Determine whether the threshold calculated in step 5 is a valid positive number (to avoid situations where the threshold is 0 / negative due to abnormal input data, making it impossible to identify abnormalities): If not (invalid threshold ≤ 0), proceed to step 7 and use the preset default threshold; If the threshold is valid, proceed to step 8 and initialize the statistics counter.

[0454] Step 7. Use the default threshold.

[0455] When the threshold calculated in step 5 is invalid, a preset empirical fixed threshold is used to ensure that subsequent anomaly detection can proceed normally.

[0456] Step 8. Initialize the counter.

[0457] Initialize three core counters for subsequent traversal and statistics: max_consecutive_poor = 0: Maximum consecutive anomaly constant, which records the length of the longest consecutive anomaly sequence detected; current_streak = 0: Current length of consecutive exceptions, the current count of consecutive exceptions being accumulated; poor_sequences = 0: The number of consecutive anomalous sequence points that meet the minimum length requirement.

[0458] Step 9. Traverse the BER array.

[0459] Enter the loop and process each data point in the BER array in turn until all data points have been traversed.

[0460] Step 10. Is the current BER valid and not 0.00? Determine if the current BER data point is valid (non-empty, non-zero value, to avoid invalid data interfering with statistics): If not (invalid data / data is 0): skip the current point and return to step 9 to process the next data point; If (data is valid): Proceed to step 11 and convert to numeric format.

[0461] Step 11. Convert to numbers.

[0462] Convert the current BER value (such as a string or scientific notation) into a comparable numeric type to prepare for subsequent anomaly detection. The conversion formula is: ber_num = convert(ber).

[0463] Step 12.ber_num>threshold? Determine if the current BER value exceeds the anomaly threshold: If (BER anomaly) occurs, proceed to step 16 and update the current consecutive anomaly count; If not (BER is normal), proceed to step 13 to check whether the previously accumulated consecutive anomalies meet the minimum length requirement.

[0464] Step 13. current_streak ≥ min_length? When the BER value is normal, check whether the previous consecutive abnormal sequences meet the minimum length requirement: If the length meets the requirement, proceed to step 14; If not (length not up to standard), proceed to step 15.

[0465] Step 14. poor_sequences++.

[0466] If the minimum length is met, the number of valid abnormal sequences is increased by 1.

[0467] Step 15. current_streak=0.

[0468] Reset the current consecutive anomaly count. After resetting, proceed to step 19.

[0469] Step 16. Current_streak++.

[0470] BER anomaly increments the current consecutive anomaly length, which is equivalent to incrementing the current consecutive anomaly count by 1, accumulating the length of the current consecutive anomaly sequence.

[0471] Step 17.current_streak>max_consecutive_poor? Determine whether the current accumulated length of consecutive anomalies exceeds the longest previously recorded length of consecutive anomalies: If you proceed to step 18; If not, proceed directly to step 19.

[0472] Step 18.max_consecutive_poor=current_streak.

[0473] Set the current continuous length as the new longest abnormal length.

[0474] Step 19. The last BER? Determine if all BER data points have been processed: If not (not fully traversed), return to step 9 and continue processing the next data point; If (the traversal is complete) proceed to step 20, check the last consecutive abnormal sequence (there may still be unprocessed abnormal sequences after the loop ends).

[0475] Step 20. current_streak ≥ min_length? Check if the length of the last consecutive abnormal segment meets the standard: If the length meets the requirement, proceed to step 21; If not (length not up to standard), proceed directly to step 22.

[0476] Step 21. poor_sequences++.

[0477] Mark the last consecutive anomaly as a valid sequence, increment the count by 1, and then proceed to step 22.

[0478] Step 22. Iterate through trend_penalities.

[0479] Format: Threshold 1 (points) Threshold 2 (points) format, with values ​​threshold_val and corresponding penalty_val.

[0480] For example, the preset trend_penalities configuration table is traversed. This configuration is a "threshold-penalty score" key-value pair (format: threshold1(score) threshold2(score)), which means that when max_consecutive_poor≥threshold, the corresponding penalty score is applied.

[0481] Step 23. max_consecutive_poor ≥ threshold_val.

[0482] Determine whether the threshold_val of the currently traversed penalty rule is satisfied by max_consecutive_poor (longest consecutive exception length): If the step is (matching rule), proceed to step 24; If the step does not match, proceed to step 25.

[0483] Step 24. trend_penalty=penalty_val.

[0484] If the maximum number of consecutive anomalies reaches or exceeds the threshold, find the matching penalty rule, exit the loop, and apply the corresponding penalty score. For example, when max_consecutive_poor reaches or exceeds the threshold of a certain penalty rule, exit the traversal, assign the penalty_val corresponding to the rule to trend_penalty (trend penalty score, reflecting the stability decrease of the link due to long consecutive anomalies), and proceed to step 28.

[0485] Step 25. The last trend_penalities configuration? Determine if all penalty rules have been traversed: If not (not fully traversed), return to step 22 and continue traversing the next penalty rule; If (traversal complete, no matching rules) proceed to step 26.

[0486] Step 26. trend_penalty==0 and poor_sequences≥multi_sequence_min_count? If max_consecutive_poor fails to capture any long consecutive exception penalty (trend_penalt is still initialized to 0), determine if multiple valid exception sequences exist: Condition: poor_sequences (number of valid outlier sequences) ≥ multi_sequence_min_countun (minimum count threshold for multiple outlier sequences, e.g., ≥ 2 sequences); If it is a (multiple segment exception), proceed to step 27; otherwise, proceed directly to step 28.

[0487] Step 27. trend_penalty=g_multi_sequence_penalty(-2).

[0488] Multi-stage exception penalty assignment: When there are multiple short anomalies in the link (no long continuous anomalies, but anomalies occur frequently), assign the value trend_penalty= g_multi_sequence_penalty(-2).

[0489] Explanation of trend_penalty == 0: The maximum number of consecutive outliers (max_consecutive_poor) has not reached any of the thresholds configured in g_trend_penalities, meaning that no single outlier sequence is long enough to reach the threshold for consecutive outlier penalties.

[0490] poor_sequences ≥g_multi_sequence_min_count Description: There are multiple anomalous sequences that meet the minimum length requirement.

[0491] 1) Although the link did not show any serious continuous anomalies (such as n consecutive anomalies), it did show multiple scattered anomaly sequences. 2) This indicates that the link quality is unstable. Although the duration of a single anomaly is short, the frequency of anomalies is high. Therefore, a multi-sequence penalty (-a points) needs to be applied to reflect this unstable state.

[0492] Step 28. Return the results.

[0493] The process has ended and the result is returned: max_consecutive_poor: The length of the longest consecutive sequence of exceptions; trend_penalty: Trend penalty score (reflects link stability; a negative score indicates decreased stability). poor_sequences: The number of valid abnormal sequences that meet the minimum length requirement.

[0494] 5. Overall score and rating.

[0495] Total score = Consistency score + Volatility score + Reliability score + Continuous anomaly penalty (negative number).

[0496] The total score is mapped to an intuitive rating (7 levels in total). The mapping relationship can be found in Table 3 below:

[0497] (v) Report generation and visualization module 540.

[0498] This module organizes all the aforementioned results into a structured report and optionally generates charts.

[0499] The report includes: Test configuration summary (TXFIR, rate, mode, etc.); Core statistical results (minimum / maximum / average / median / geometric mean BER, range of variation, log standard deviation); Multidimensional health score details (scores and explanations for multiple dimensions); Overall score and rating; List of "Real PoorBER"; Optimization recommendations based on ratings (e.g., "Excellent consistency, configuration can be deployed directly" or "Poor reliability, it is recommended to try other FIR configurations").

[0500] (vi) Configuration management.

[0501] This system adopts an external parameter design; all scoring thresholds, weight allocations, and penalty rules are defined through external configuration files, allowing for strategy adjustments without code modification. The configuration management module includes the following parameter groups: The scoring weights are configured as follows: consistency weight (a%), stability weight (b%), and reliability weight (c%), where a+b+c=100.

[0502] Consistency scoring threshold table: defines the mapping relationship from logarithmic standard deviation to score (10-level threshold).

[0503] Stability rating threshold table: defines the mapping relationship between fluctuation range and score (9-level threshold).

[0504] Reliability penalty parameters: deviation penalty table (5-level deviation threshold and corresponding deduction), quantity penalty table (3-level ratio threshold and corresponding deduction), baseline score (25) and maximum penalty (25).

[0505] Continuous anomaly detection parameters: anomaly multiple (2), minimum continuous anomaly length (3), trend penalty table.

[0506] Rating and Recommendation Configuration: Rating level threshold (7 levels), consistency / stability / reliability conclusion threshold, recommendation level threshold (5 levels).

[0507] This design allows users to adjust the scoring strategy according to specific application scenarios (such as different rates, different modulation methods, and different chip versions) without modifying the core algorithm code.

[0508] (vii) Supplementary explanations regarding the above technical solutions.

[0509] 1. Adjustment of Scoring Weights and Dimensions: The weighting of multiple dimensions (consistency, stability, reliability, and continuous anomaly BER) (a%, b%, c%, 0%) can be adjusted according to specific application scenarios. For example, in data center interconnect scenarios with extremely high consistency requirements, the consistency weight can be increased to 60%; in some scenarios more sensitive to peak performance, the stability weight can be appropriately increased. It's even possible to consider adding more dimensions, such as "performance excellence" (an absolute value score based on the geometric mean BER).

[0510] 2. Replacement of specific scoring mapping functions: The mapping of scores for each dimension from statistics (such as sigma_ln, Fluctuation_Ratio) to scores can be replaced by continuous functions (such as exponential decay functions, sigma functions) instead of piecewise lookup tables to achieve a smoother scoring transition.

[0511] 3. Alternatives to the “Poor BER” identification algorithm: In addition to using the rule of “greater than Q3 and greater than the geometric mean”, a method based on logarithmic standard deviation can also be adopted. For example, BERs that deviate from the geometric mean by more than 2×sigma_ln can be considered as outliers, and the degree of deviation can be calculated and penalized.

[0512] 4. Alternatives to integration and delivery methods: This application can exist not only as a standalone software tool, but also as an advanced diagnostic module in existing link test equipment firmware, chip debugging software (such as chip debugging SDK, Software Development Kit) or network device management system.

[0513] 5. Expanding the input data source: The input of this method is not limited to BER data obtained through multiple initializations. It is also applicable to analyzing time-series BER data obtained from real-time monitoring of online devices to assess the health trend of the link during operation.

[0514] (viii) Usage scenarios.

[0515] 1. After finding the most stable TX FIR through multiple rounds of TX FIR scanning, the method provided in this application can be used to initialize SerDes for the same TX FIR through multiple rounds (the number of test rounds can be configured). This method will calculate the BER after the test, provide BER statistical data analysis, evaluate the BER stability under the TX FIR configuration, and generate a BER curve.

[0516] 2. Find several TX FIR sets with similar BER values, perform multiple rounds of initialization on each set, and evaluate which set to use based on the health score.

[0517] 3. Product environment: Conduct long-term BER monitoring and assess link health.

[0518] 4. Based on the health score, assess whether the link quality is showing a deteriorating trend and issue a fault warning.

[0519] In summary, this application achieves the following technical effects: 1. It provides a multi-dimensional and quantifiable comprehensive evaluation system: through quantitative scoring of multiple dimensions such as consistency, stability, reliability and continuous abnormal BER, it comprehensively and objectively reflects the overall health of the link, making the evaluation results more instructive.

[0520] 2. Based on scientific statistical models: The system adopts a log-normal distribution model that conforms to the nature of BER data, and uses geometric mean and log standard deviation as core statistics to make the analysis results more accurate and more resistant to interference, and can truly reflect the "typical" performance and fluctuation characteristics of the link.

[0521] 3. Achieved objective optimization of configuration: Provided a unified and quantitative comparison standard (comprehensive health score) for multiple TX FIR configurations with similar performance, solved the problem of relying on experience or single data point for decision-making, and improved the scientific nature and efficiency of configuration selection.

[0522] 4. Enhanced system predictability and maintainability: By periodically scoring the long-term monitored BER data, scoring trends can be tracked. A decline in consistency or reliability scores can serve as an early warning of link performance degradation, thereby enabling predictive maintenance and improving system availability.

[0523] 5. Increased confidence in product deployment: Conducting multiple rounds of initial health tests and obtaining high scores before the product leaves the factory or the system goes live enhances confidence in the long-term stable operation of the link in the actual use environment.

[0524] Figure 9 This is a structural block diagram of a communication link quality assessment apparatus provided in an exemplary embodiment of this application. The communication link quality assessment apparatus includes: The acquisition module 910 is used to acquire at least two bit error rates of the communication link; Processing module 920 is used to determine a bit error rate statistical value based on the at least two bit error rates, wherein the bit error rate statistical value includes a logarithmic statistical value, and the logarithmic statistical value is a statistical value determined based on the logarithms corresponding to the at least two bit error rates respectively; The processing module 920 is used to perform a quality assessment on the communication link based on the bit error rate statistics, and obtain the quality assessment result of the communication link.

[0525] In some embodiments, the processing module 920 is configured to: The health score of the communication link is determined based on the bit error rate statistics. Based on the health score, the quality assessment result of the communication link is obtained.

[0526] In some embodiments, the logarithmic statistic includes the logarithmic standard deviation, which refers to the standard deviation of the logarithms corresponding to the at least two bit error rates; the processing module 920 is used for: Based on the logarithmic standard deviation, the consistency score of the communication link in the consistency dimension is obtained, and the consistency score is negatively correlated with the logarithmic standard deviation; Based on the consistency score, the health score of the communication link is determined.

[0527] In some embodiments, the logarithmic statistic includes the geometric mean bit error rate, which is the geometric average of the at least two bit error rates; the processing module 920 is configured to: Determine at least one bad bit error rate among the at least two bit error rates; A first penalty score is determined based on the degree of deviation between the at least one bad bit error rate and the geometric mean bit error rate; Based on the first penalty score, a reliability score for the communication link in the reliability dimension is determined, and the reliability score is negatively correlated with the first penalty score; Based on the reliability score, the health score of the communication link is determined.

[0528] In some embodiments, the processing module 920 is configured to: A proportional penalty score is determined based on the proportion of the at least one defective bit error rate to the at least two bit error rates, and the proportional penalty score is positively correlated with the proportion. A deviation penalty score is determined based on the degree of deviation between the at least one defective bit error rate and the geometric mean bit error rate, wherein the deviation penalty score is positively correlated with the degree of deviation. The first penalty score is determined based on the deviation penalty score and the proportion penalty score.

[0529] In some embodiments, the processing module 920 is configured to: Based on the quantiles of the at least two bit error rates, at least one candidate bad bit error rate is determined from the at least two bit error rates; The bit error rate that is greater than the geometric mean bit error rate among the at least one candidate bad bit error rates is determined as the bad bit error rate.

[0530] In some embodiments, the processing module 920 is configured to: Determine at least one quantile corresponding to the at least two bit error rates; Determine at least two quantile intervals obtained by dividing the at least one quantile, wherein the at least two quantile intervals include a target interval that meets the requirements for a poor bit error rate indication; The bit error rate that falls within the target range among the at least two bit error rates is determined as the candidate bad bit error rate.

[0531] In some embodiments, the logarithmic statistics include the geometric mean bit error rate and the logarithmic standard deviation; the processing module 920 is configured to: The anomaly threshold is determined based on the geometric mean bit error rate and the logarithmic standard deviation; Based on the abnormal threshold, at least one abnormal sequence among the at least two bit error rates is determined, the abnormal sequence comprising at least two consecutive abnormal bit error rates, and the abnormal bit error rate being the bit error rate among the at least two bit error rates that meets the abnormal threshold requirement; A second penalty score is determined based on the at least one abnormal sequence; Based on the second penalty score, the continuous anomaly score of the communication link in the continuous anomaly dimension is determined, and the continuous anomaly score is negatively correlated with the second penalty score; The health score of the communication link is determined based on the continuous anomaly score.

[0532] In some embodiments, the processing module 920 is configured to: A first sub-score is determined based on the longest abnormal sequence among the at least one abnormal sequence; the first sub-score is positively correlated with the number of abnormal bit error rates in the longest abnormal sequence. A second sub-score is determined based on the number of sequences in the at least one abnormal sequence; the second sub-score is positively correlated with the number of sequences. The second penalty score is determined based on the first sub-score and the second sub-score.

[0533] In some embodiments, the bit error rate statistics further include linear statistics, which are statistical values ​​determined based on the at least two bit error rates; the processing module 920 is configured to: The health score of the communication link is determined based on the logarithmic and linear statistics.

[0534] In some embodiments, the linear statistics include the fluctuation range of the at least two bit error rates; the processing module 920 is configured to: Based on the fluctuation range of the at least two bit error rates, a stability score for the communication link in the stability dimension is obtained, and the stability score is negatively correlated with the fluctuation range. The health score of the communication link is determined based on the logarithmic statistics and the stability score.

[0535] In some embodiments, the processing module 920 is configured to: Based on the logarithmic statistics, at least one of the consistency score, reliability score, and continuous anomaly score is determined. The health score of the communication link is determined based on at least one of the consistency score, the reliability score, and the continuous anomaly score, as well as the stability score.

[0536] In some embodiments, the at least two bit error rates are obtained after performing at least two initialization operations on the communication link; the processing module 920 is configured to: After the initialization operation of the communication link is completed, the connectivity status of the communication link is determined; The connection success rate of the communication link is determined based on the link connectivity status of the communication link corresponding to the at least two initialization operations. The quality assessment result of the communication link is determined based on the health score and the connectivity success rate.

[0537] In some embodiments, the acquisition module 910 is configured to: Under the same link configuration parameters, the communication link is controlled to perform at least two initialization operations; after the i-th initialization operation is completed, the bit error rate of the communication link corresponding to the i-th initialization operation is obtained, where i is a positive integer; the bit error rates of the communication link corresponding to the at least two initialization operations are used as the at least two bit error rates. Under at least two link configuration parameters, the communication link is controlled to perform initialization operations respectively; after the initialization operation corresponding to the j-th link configuration parameter is completed, the bit error rate of the communication link corresponding to the j-th link configuration parameter is obtained, where j is a positive integer; the bit error rates of the communication links corresponding to the at least two link configuration parameters are used as the at least two bit error rates. The communication link is operated under fixed link configuration parameters; during the operation of the communication link, the bit error rate corresponding to at least two time windows is sampled as the at least two bit error rates.

[0538] In some embodiments, the communication link includes a high-speed link, which includes a serializer / deserializer SerDes channel.

[0539] In some embodiments, the acquisition module 910 is configured to: During the operation of the communication link, at least two bit error rate (BER) data points are collected; a structured data file is generated based on the at least two BER data points; the structured data file is parsed to obtain the at least two BER rates. Obtain the bit error rate array or structured data file corresponding to the communication link; parse the bit error rate array or the structured data file to obtain the at least two bit error rates.

[0540] In some embodiments, the acquisition module 910 is configured to: Based on the quality assessment scenario, a recommended sample size value is determined, which is used to indicate the recommended number of times the at least two bit error rates are to be obtained. The quality assessment scenarios include at least one of the following: rapid screening scenario, initial stability assessment scenario, production yield analysis scenario, and fault diagnosis scenario.

[0541] In some embodiments, the at least two bit error rates are bit error rate data acquired periodically; the processing module 920 is configured to: Determine the scoring trend corresponding to at least two periods, wherein the scoring trend includes at least one of the following: health score change trend, consistency score change trend, stability score change trend, reliability score change trend, and continuous anomaly score change trend corresponding to the communication link; Predictive maintenance is performed on the communication link based on the scoring trend.

[0542] In some embodiments, the acquisition module 910 is configured to: Obtain the configuration file corresponding to the communication link; The configuration file is used to configure the evaluation rule parameters in the quality evaluation process of the communication link.

[0543] In summary, the communication link quality assessment device provided in this application acquires at least two bit error rates (BERs), calculates BER statistics including logarithmic values, and performs quality assessment on the communication link based on the BER statistics to obtain the quality assessment result. Compared to the single-point testing method in related technologies that simply collects multiple BERs and calculates their arithmetic mean, the assessment process in this application utilizes the log-normal distribution characteristic of BERs and uses BER statistics including logarithmic values ​​for quality assessment. This can more accurately reproduce the actual distribution pattern of BERs, thereby more realistically characterizing the transmission quality of communication links (especially high-speed links) in actual production environments, improving the accuracy of quality assessment, and ensuring the operational stability and data processing efficiency of high-performance computing systems.

[0544] It should be noted that the specific limitations of the embodiments of the one or more communication link quality assessment devices provided above can be found in the specific limitations of the embodiments of the communication link quality assessment method above, and will not be repeated here. Each module of the above-mentioned device can be implemented entirely or partially by software, hardware, or a combination thereof. Each module can be embedded in the processor of the computer device in hardware form or independent of the processor, or it can be stored in the memory of the computer device in software form, so that the processor can call and execute the corresponding operations of each module.

[0545] Please refer to Figure 10 This diagram illustrates a structural block diagram of a computer device provided in one embodiment of this application. The computer device 1000 may be... Figure 1The terminal or server in the computer system shown is used to implement the communication link quality assessment method provided in the above embodiments. Specifically: Typically, computer device 1000 includes a processor 1001 and a memory 1002.

[0546] Processor 1001 may include one or more processing cores, such as a quad-core processor, an octa-core processor, etc. Processor 1001 may be implemented using at least one hardware form selected from DSP (Digital Signal Processing), FPGA (Field Programmable Gate Array), and PLA (Programmable Logic Array). Processor 1001 may also include a main processor and a coprocessor. The main processor, also known as a CPU (Central Processing Unit), is used to process data in the wake-up state; the coprocessor is a low-power processor used to process data in the standby state. In some embodiments, processor 1001 may integrate a GPU, which is responsible for rendering and drawing the content to be displayed on the screen. In some embodiments, processor 1001 may also include an AI (Artificial Intelligence) processor, which is used to handle computational operations related to machine learning.

[0547] The memory 1002 may include one or more computer-readable storage media, which may be non-transitory. The memory 1002 may also include high-speed random access memory and non-volatile memory, such as one or more disk storage devices or flash memory devices. In some embodiments, the non-transitory computer-readable storage media in the memory 1002 are used to store a computer program configured to be executed by one or more processors to implement the aforementioned communication link quality assessment method.

[0548] In some embodiments, the computer device 1000 may also optionally include other components 1003, such as a peripheral device interface and at least one peripheral device. The processor 1001, memory 1002, and peripheral device interface can be connected via a bus or signal lines. Each peripheral device can be connected to the peripheral device interface via a bus, signal lines, or a circuit board. Specifically, the peripheral device includes at least one of a radio frequency circuit, a display screen, an audio circuit, and a power supply.

[0549] Those skilled in the art will understand that Figure 10The structure shown does not constitute a limitation on the computer device 1000, and may include more or fewer components than shown, or combine certain components, or use different component arrangements.

[0550] In an exemplary embodiment, a computer-readable storage medium is also provided, wherein a computer program is stored in the storage medium, and the computer program, when executed by a processor, implements the aforementioned communication link quality assessment method. Optionally, the computer-readable storage medium may include: ROM (Read-Only Memory), RAM (Random Access Memory), SSD (Solid State Drives), or optical disc, etc. The random access memory may include ReRAM (Resistance Random Access Memory) and DRAM (Dynamic Random Access Memory).

[0551] In an exemplary embodiment, a computer program product is also provided, the computer program product including a computer program stored in a computer-readable storage medium. A processor of a computer device reads the computer program from the computer-readable storage medium, and the processor executes the computer program, causing the computer device to perform the above-described communication link quality assessment method.

[0552] It should be noted that the data collection and processing in this application should strictly comply with the requirements of relevant national laws and regulations, obtain the informed consent or separate consent of the personal information subject, and carry out subsequent data use and processing within the scope of laws and regulations and the authorization of the personal information subject.

[0553] It should be understood that "multiple" as used herein refers to two or more. "And / or" describes the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent: A alone, A and B simultaneously, or B alone. The character " / " generally indicates that the preceding and following related objects are in an "or" relationship. Furthermore, the step numbers described herein are merely illustrative of one possible execution order. In some other embodiments, the steps may not be executed in numerical order, such as two steps with different numbers being executed simultaneously, or two steps with different numbers being executed in the reverse order of the illustration. This application does not limit this.

[0554] The above description is merely an exemplary embodiment of this application and is not intended to limit this application. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the protection scope of this application.

Claims

1. A method for evaluating the quality of a communication link, characterized in that, The method includes: Obtain at least two bit error rates for the communication link; A bit error rate statistic is determined based on the at least two bit error rates, wherein the bit error rate statistic includes a logarithmic statistic, which is a statistic determined based on the logarithm corresponding to each of the at least two bit error rates; Based on the bit error rate statistics, a quality assessment is performed on the communication link to obtain the quality assessment result of the communication link.

2. The method according to claim 1, characterized in that, The step of performing a quality assessment on the communication link based on the bit error rate statistics to obtain the quality assessment result of the communication link includes: The health score of the communication link is determined based on the bit error rate statistics. Based on the health score, the quality assessment result of the communication link is obtained.

3. The method according to claim 2, characterized in that, The logarithmic statistics include the logarithmic standard deviation, which refers to the standard deviation of the logarithm corresponding to the at least two bit error rates; Determining the health score of the communication link based on the bit error rate statistics includes: Based on the logarithmic standard deviation, the consistency score of the communication link in the consistency dimension is obtained, and the consistency score is negatively correlated with the logarithmic standard deviation; Based on the consistency score, the health score of the communication link is determined.

4. The method according to claim 2, characterized in that, The logarithmic statistics include the geometric mean bit error rate, which is the geometric average of the at least two bit error rates. Determining the health score of the communication link based on the bit error rate statistics includes: Determine at least one bad bit error rate among the at least two bit error rates; A first penalty score is determined based on the degree of deviation between the at least one bad bit error rate and the geometric mean bit error rate; Based on the first penalty score, a reliability score for the communication link in the reliability dimension is determined, and the reliability score is negatively correlated with the first penalty score; Based on the reliability score, the health score of the communication link is determined.

5. The method according to claim 4, characterized in that, The method further includes: A proportional penalty score is determined based on the proportion of the at least one defective bit error rate to the at least two bit error rates, and the proportional penalty score is positively correlated with the proportion. The determination of the first penalty score based on the deviation between the at least one bad bit error rate and the geometric mean bit error rate includes: A deviation penalty score is determined based on the degree of deviation between the at least one defective bit error rate and the geometric mean bit error rate, wherein the deviation penalty score is positively correlated with the degree of deviation. The first penalty score is determined based on the deviation penalty score and the proportion penalty score.

6. The method according to claim 4, characterized in that, Determining at least one bad bit error rate among the at least two bit error rates includes: Based on the quantiles of the at least two bit error rates, at least one candidate bad bit error rate is determined from the at least two bit error rates; The bit error rate that is greater than the geometric mean bit error rate among the at least one candidate bad bit error rates is determined as the bad bit error rate.

7. The method according to claim 6, characterized in that, The step of determining at least one candidate bad bit error rate from the at least two bit error rates based on the quantiles of the at least two bit error rates includes: Determine at least one quantile corresponding to the at least two bit error rates; Determine at least two quantile intervals obtained by dividing the at least one quantile, wherein the at least two quantile intervals include a target interval that meets the requirements for a poor bit error rate indication; The bit error rate that falls within the target range among the at least two bit error rates is determined as the candidate bad bit error rate.

8. The method according to claim 2, characterized in that, The logarithmic statistics include the geometric mean bit error rate and the logarithmic standard deviation; Determining the health score of the communication link based on the bit error rate statistics includes: The anomaly threshold is determined based on the geometric mean bit error rate and the logarithmic standard deviation; Based on the abnormal threshold, at least one abnormal sequence among the at least two bit error rates is determined, the abnormal sequence comprising at least two consecutive abnormal bit error rates, and the abnormal bit error rate being the bit error rate among the at least two bit error rates that meets the abnormal threshold requirement; A second penalty score is determined based on the at least one abnormal sequence; Based on the second penalty score, the continuous anomaly score of the communication link in the continuous anomaly dimension is determined, and the continuous anomaly score is negatively correlated with the second penalty score; The health score of the communication link is determined based on the continuous anomaly score.

9. The method according to claim 8, characterized in that, Determining the second penalty score based on the at least one abnormal sequence includes: A first sub-score is determined based on the longest abnormal sequence among the at least one abnormal sequence; the first sub-score is positively correlated with the number of abnormal bit error rates in the longest abnormal sequence. A second sub-score is determined based on the number of sequences in the at least one abnormal sequence; the second sub-score is positively correlated with the number of sequences. The second penalty score is determined based on the first sub-score and the second sub-score.

10. The method according to claim 2, characterized in that, The bit error rate statistics also include linear statistics, which are statistical values ​​determined based on the at least two bit error rates; Determining the health score of the communication link based on the bit error rate statistics includes: The health score of the communication link is determined based on the logarithmic and linear statistics.

11. The method according to claim 10, characterized in that, The linear statistics include the fluctuation range of the at least two bit error rates; Determining the health score of the communication link based on the logarithmic and linear statistical values ​​includes: Based on the fluctuation range of the at least two bit error rates, a stability score for the communication link in the stability dimension is obtained, and the stability score is negatively correlated with the fluctuation range. The health score of the communication link is determined based on the logarithmic statistics and the stability score.

12. The method according to claim 11, characterized in that, Determining the health score of the communication link based on the logarithmic statistics and the stability score includes: Based on the logarithmic statistics, at least one of the consistency score, reliability score, and continuous anomaly score is determined. The health score of the communication link is determined based on at least one of the consistency score, the reliability score, and the continuous anomaly score, as well as the stability score.

13. The method according to claim 2, characterized in that, The at least two bit error rates are obtained after performing at least two initialization operations on the communication link; The method further includes: After the initialization operation of the communication link is completed, the connectivity status of the communication link is determined; The connection success rate of the communication link is determined based on the link connectivity status of the communication link corresponding to the at least two initialization operations. The process of obtaining the quality assessment result of the communication link based on the health score includes: The quality assessment result of the communication link is determined based on the health score and the connectivity success rate.

14. The method according to any one of claims 1 to 13, characterized in that, The acquisition of at least two bit error rates of the communication link includes at least one of the following: Under the same link configuration parameters, the communication link is controlled to perform at least two initialization operations; after the i-th initialization operation is completed, the bit error rate of the communication link corresponding to the i-th initialization operation is obtained, where i is a positive integer; The bit error rate of the communication link corresponding to the at least two initialization operations is taken as the at least two bit error rates; Under at least two link configuration parameters, the communication link is controlled to perform initialization operations respectively; after the initialization operation corresponding to the j-th link configuration parameter is completed, the bit error rate of the communication link corresponding to the j-th link configuration parameter is obtained, where j is a positive integer; the bit error rates of the communication links corresponding to the at least two link configuration parameters are used as the at least two bit error rates. The communication link is operated under fixed link configuration parameters; During the operation of the communication link, the bit error rate corresponding to at least two time windows is sampled as the at least two bit error rates.

15. The method according to any one of claims 1 to 13, characterized in that, The communication link includes a high-speed link, which includes a serializer / deserializer SerDes channel.

16. The method according to any one of claims 1 to 13, characterized in that, The acquisition of at least two bit error rates of the communication link includes at least one of the following: During the operation of the communication link, at least two bit error rate (BER) data points are collected; a structured data file is generated based on the at least two BER data points; the structured data file is parsed to obtain the at least two BER rates. Obtain the bit error rate array or structured data file corresponding to the communication link; parse the bit error rate array or the structured data file to obtain the at least two bit error rates.

17. The method according to any one of claims 1 to 13, characterized in that, Before obtaining at least two bit error rates of the communication link, the method further includes: Based on the quality assessment scenario, a recommended sample size value is determined, which is used to indicate the recommended number of times the at least two bit error rates are to be obtained. The quality assessment scenarios include at least one of the following: rapid screening scenario, initial stability assessment scenario, production yield analysis scenario, and fault diagnosis scenario.

18. The method according to any one of claims 1 to 13, characterized in that, The at least two bit error rates are bit error rate data acquired periodically; After performing a quality assessment on the communication link based on the bit error rate statistics to obtain the quality assessment result of the communication link, the method further includes: Determine the scoring trend corresponding to at least two periods, wherein the scoring trend includes at least one of the following: health score change trend, consistency score change trend, stability score change trend, reliability score change trend, and continuous anomaly score change trend corresponding to the communication link; Predictive maintenance is performed on the communication link based on the scoring trend.

19. The method according to any one of claims 1 to 13, characterized in that, The method further includes: Obtain the configuration file corresponding to the communication link; The configuration file is used to configure the evaluation rule parameters in the quality evaluation process of the communication link.

20. A communication link quality assessment device, characterized in that, The device includes: The acquisition module is used to acquire at least two bit error rates of the communication link; The processing module is configured to determine a bit error rate statistical value based on the at least two bit error rates, wherein the bit error rate statistical value includes a logarithmic statistical value, which is a statistical value determined based on the logarithms corresponding to the at least two bit error rates respectively; The processing module is used to perform a quality assessment on the communication link based on the bit error rate statistics, and obtain the quality assessment result of the communication link.

21. A computer device, characterized in that, The computer device includes a processor and a memory, the memory storing at least one program, which is loaded and executed by the processor to implement the communication link quality assessment method as described in any one of claims 1 to 19.

22. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores at least one program, which is loaded and executed by a processor to implement the communication link quality assessment method as described in any one of claims 1 to 19.

23. A computer program product, characterized in that, It includes a computer program that, when executed by a processor, implements the communication link quality assessment method as described in any one of claims 1 to 19.

Images