Signal defect root cause detection method, device and equipment

Abstract

The embodiment of the invention provides a method, a device and equipment for detecting a defect root cause of a signal, which are used for solving the problem that the defect root cause of the signal cannot be detected in the prior art. The method comprises the following steps: acquiring a plurality of measurement data of an eye pattern corresponding to a signal to be detected, and vectorizing the plurality of measurement data to obtain a plurality of measurement data vectors; performing feature screening on the plurality of measurement data vectors, and determining at least one target measurement data vector related to defects in the plurality of measurement data vectors; and performing feature extraction on the at least one target measurement data vector to obtain a defect feature matrix, and determining a detection result of the defect root cause of the to-be-detected signal according to the defect feature matrix.

Description

Technical Field

[0001] The present application relates to the field of signal detection, and in particular to a method, device and equipment for detecting root causes of signal defects. Background Art

[0002] High-speed signal transmission is one of the core technologies in modern electronic systems and is widely used in data centers, communication systems, computer networks and other fields. With the continuous increase in data transmission rates, the signal integrity issue of high-speed signal transmission has become particularly important. If the signal is interfered with and distorted during transmission, the signal quality will be reduced, which will affect the reliability and stability of data transmission. One of the existing signal detection methods is to superimpose the standard signal eye diagram template with the eye diagram of the signal to be detected, and then determine whether the signal waveform meets specific standards. Alternatively, the eye diagram is classified through deep learning to determine the quality of the signal to be detected. However, the above method can only perform compliance detection, that is, it can only determine the overall evaluation of the quality of the signal to be detected, but cannot identify the specific root cause of the signal defect problem. Summary of the invention

[0003] The embodiments of the present application provide a method, device and apparatus for detecting a root cause of a defect of a signal, so as to solve the problem that the root cause of a defect of a signal cannot be detected in the prior art.

[0004] In a first aspect, an embodiment of the present application provides a method for detecting a root cause of a defect of a signal, comprising:

[0005] Acquire multiple measurement data of the eye diagram corresponding to the signal to be detected, and perform vector processing on the multiple measurement data to obtain multiple measurement data vectors;

[0006] Performing feature screening on the multiple measurement data vectors to determine at least one target measurement data vector related to the defect among the multiple measurement data vectors;

[0007] Feature extraction is performed on the at least one target measurement data vector to obtain a defect feature matrix, and a detection result of the defect root cause of the signal to be detected is determined according to the defect feature matrix.

[0008] Through the above solution, by filtering the measurement data of the eye diagram, the interference of irrelevant attributes on defect identification can be avoided, and the accuracy and reliability of subsequent classification can be improved. Furthermore, by classifying the filtered features through a multi-label classification model, the specific root cause of the signal quality problem can be identified in a fine-grained manner, so that users can perform targeted optimization and troubleshooting.

[0009] In a possible implementation manner, the vectorizing the multiple measurement data to obtain multiple measurement data vectors includes:

[0010] For each measurement data, a measurement text is generated based on a comparison result between the measurement data and preset standard data; wherein the measurement text includes the measurement data, the preset standard data and the comparison result;

[0011] For each measurement text, vectorization is performed on the measurement text through a language model to obtain a measurement data vector.

[0012] Based on the above scheme, multiple measurement data in the eye diagram of the signal to be detected can be converted into descriptive text, and the text can be vectorized to facilitate subsequent processing.

[0013] In a possible implementation, the performing feature screening on the multiple measurement data vectors to determine at least one target measurement data vector related to the defect in the multiple measurement data vectors includes:

[0014] The multiple measurement data vectors are input into a defect feature screening model, and feature screening is performed on the multiple measurement data vectors through the defect feature screening model to determine at least one target measurement data vector related to the defect among the multiple measurement data vectors.

[0015] Based on the above scheme, by screening multiple measurement data vectors, features related to the defects can be selected, thereby avoiding the influence of irrelevant features on subsequent defect root cause identification.

[0016] In a possible implementation, the defect feature screening model is obtained by the following training method:

[0017] Acquire a first training sample set, wherein the first training sample set includes a plurality of first samples, wherein the first samples include a plurality of first sample measurement data of an eye diagram corresponding to a first sample signal, and a defect label set corresponding to each first sample measurement data, wherein the defect label set includes a correlation label between each first sample measurement data and a defect;

[0018] For a first sample signal in each first sample, vectorization processing is performed on a plurality of first sample measurement data corresponding to the first sample signal to obtain a plurality of first sample measurement data vectors;

[0019] Repeat the following operations: for each first sample, screen the plurality of first sample measurement data vectors through the first network model to obtain a prediction result, wherein the prediction result is used to characterize a set of first sample measurement data vectors related to the defect; and update the network parameters of the first network model based on the prediction result and the loss value determined by the defect label set;

[0020] When the first set condition is met, the updated first network model is used as a defect feature screening model;

[0021] Among them, the first setting condition is that the determined loss value is less than a first set loss threshold.

[0022] Based on the above scheme, the trained defect feature screening model can identify features related to defects, thereby improving the accuracy of defect root cause identification.

[0023] In a possible implementation manner, extracting features from the at least one target measurement data vector to obtain a defect feature matrix includes:

[0024] Inputting the at least one target measurement data vector into a multi-label classification model, and performing feature extraction on the at least one target measurement data vector through an encoder in the multi-label classification model to obtain a defect feature matrix;

[0025] The step of determining the detection result of the signal to be detected for each defect root cause label according to the defect feature matrix includes:

[0026] Performing multi-label classification on the defect feature matrix through a multi-label classification model to determine the probability value of the signal to be detected for each defect root cause;

[0027] According to the probability value of each defect root cause, a detection result of the defect root cause corresponding to the signal to be detected is determined; wherein the detection result includes at least one target probability value, and the target probability value is a probability value greater than a set probability threshold.

[0028] Based on the above scheme, the defect feature matrix is ​​analyzed through a multi-label classification model to determine the root cause of the defect that causes the defect in the signal to be detected, so as to facilitate subsequent processing.

[0029] In one possible implementation, a multi-label classification model is obtained by the following training method:

[0030] Acquire a second training sample set, wherein the second training sample set includes a plurality of second samples, the second samples include a plurality of second measurement data of an eye diagram corresponding to a second sample signal, and a defect root cause label set corresponding to the second sample signal; the defect root cause label set is used to indicate at least one defect root cause of a defect in the second signal;

[0031] For the second sample signal in each second sample, vectorization processing is performed on a plurality of second sample measurement data corresponding to the second sample signal to obtain a plurality of second sample measurement data vectors;

[0032] Repeat the following operations: extract features from the plurality of second sample measurement data vectors based on an encoder in the second network model, and determine a predicted defect root cause set corresponding to the second sample signal; update network parameters of the second network model based on a loss value determined by the predicted defect root cause set and the defect root cause label set;

[0033] When the second set condition is met, the updated second network model is used as a multi-label classification model;

[0034] Among them, the second set condition is that the loss value determined in the current training round is less than the second set loss threshold.

[0035] In a possible implementation, the method further includes:

[0036] Determine a target solution corresponding to a detection result of a defect root cause of a signal to be detected based on a pre-configured association relationship between a defect root cause and a solution;

[0037] The detection result of the defect root cause of the signal to be detected and the corresponding target solution are displayed on the display interface.

[0038] Based on the above solution, according to the pre-configured association relationship between the defect root cause and the solution, after the defect root cause is determined, the defect root cause and the solution are displayed on the display interface to facilitate subsequent processing.

[0039] In a second aspect, an embodiment of the present application provides a signal defect root cause detection device, including:

[0040] An acquisition module, used for acquiring a plurality of measurement data of the eye diagram corresponding to the signal to be detected;

[0041] A processing module, used for performing vector processing on the multiple measurement data to obtain multiple measurement data vectors;

[0042] A screening module, configured to perform feature screening on the plurality of measurement data vectors, and determine at least one target measurement data vector related to a defect among the plurality of measurement data vectors;

[0043] The processing module is further used to extract features from the at least one target measurement data vector to obtain a defect feature matrix, and determine a detection result of a defect root cause of the signal to be detected based on the defect feature matrix.

[0044] In a possible implementation manner, the processing module, when performing vectorization processing on the multiple measurement data to obtain multiple measurement data vectors, is specifically configured to:

[0045] For each measurement data, a measurement text is generated based on a comparison result between the measurement data and preset standard data; wherein the measurement text includes the measurement data, the preset standard data and the comparison result;

[0046] For each measurement text, vectorization is performed on the measurement text through a language model to obtain a measurement data vector.

[0047] In a possible implementation, the screening module, when performing feature screening on the multiple measurement data vectors to determine at least one target measurement data vector related to a defect in the multiple measurement data vectors, is specifically configured to:

[0048] The multiple measurement data vectors are input into a defect feature screening model, and feature screening is performed on the multiple measurement data vectors through the defect feature screening model to determine at least one target measurement data vector related to the defect among the multiple measurement data vectors.

[0049] In a possible implementation manner, the processing module is specifically configured to:

[0050] The defect feature screening model is obtained through the following training method:

[0051] Acquire a first training sample set, wherein the first training sample set includes a plurality of first samples, wherein the first samples include a plurality of first sample measurement data of an eye diagram corresponding to a first sample signal, and a defect label set corresponding to each first sample measurement data, wherein the defect label set includes a correlation label between each first sample measurement data and a defect;

[0052] For a first sample signal in each first sample, vectorization processing is performed on a plurality of first sample measurement data corresponding to the first sample signal to obtain a plurality of first sample measurement data vectors;

[0053] Repeat the following operations: for each first sample, screen the plurality of first sample measurement data vectors through the first network model to obtain a prediction result, wherein the prediction result is used to characterize a set of first sample measurement data vectors related to the defect; and update the network parameters of the first network model based on the prediction result and the loss value determined by the defect label set;

[0054] When the first set condition is met, the updated first network model is used as a defect feature screening model;

[0055] Among them, the first setting condition is that the determined loss value is less than a first set loss threshold.

[0056] In a possible implementation manner, the processing module, when extracting features from the at least one target measurement data vector to obtain a defect feature matrix, is specifically configured to:

[0057] Inputting the at least one target measurement data vector into a multi-label classification model, and performing feature extraction on the at least one target measurement data vector through an encoder in the multi-label classification model to obtain a defect feature matrix;

[0058] The processing module, when determining the detection result of the signal to be detected for each defect root cause label according to the defect feature matrix, is specifically used to:

[0059] Performing multi-label classification on the defect feature matrix through a multi-label classification model to determine the probability value of the signal to be detected for each defect root cause;

[0060] According to the probability value of each defect root cause, a detection result of the defect root cause corresponding to the signal to be detected is determined; wherein the detection result includes at least one target probability value, and the target probability value is a probability value greater than a set probability threshold.

[0061] In a possible implementation manner, the processing module is specifically configured to:

[0062] The multi-label classification model is obtained through the following training method:

[0063] Acquire a second training sample set, wherein the second training sample set includes a plurality of second samples, the second samples include a plurality of second measurement data of an eye diagram corresponding to a second sample signal, and a defect root cause label set corresponding to the second sample signal; the defect root cause label set is used to indicate at least one defect root cause of a defect in the second signal;

[0064] For the second sample signal in each second sample, vectorization processing is performed on a plurality of second sample measurement data corresponding to the second sample signal to obtain a plurality of second sample measurement data vectors;

[0065] Repeat the following operations: extract features from the plurality of second sample measurement data vectors based on an encoder in the second network model, and determine a predicted defect root cause set corresponding to the second sample signal; update network parameters of the second network model based on a loss value determined by the predicted defect root cause set and the defect root cause label set;

[0066] When the second set condition is met, the updated second network model is used as a multi-label classification model;

[0067] Among them, the second set condition is that the loss value determined in the current training round is less than the second set loss threshold.

[0068] In a possible implementation, the processing module is further configured to:

[0069] Determine a target solution corresponding to a detection result of a defect root cause of a signal to be detected based on a pre-configured association relationship between a defect root cause and a solution;

[0070] The detection result of the defect root cause of the signal to be detected and the corresponding target solution are displayed on the display interface.

[0071] In a third aspect, an embodiment of the present application provides an execution device, including:

[0072] A memory for storing program instructions;

[0073] The processor is used to obtain the program instructions in the memory and execute the method described in the first aspect and different implementation methods of the first aspect according to the obtained program instructions.

[0074] In a fourth aspect, an embodiment of the present application provides a computer-readable storage medium, wherein the computer-readable storage medium includes computer instructions. When the computer instructions are executed by a computer, the method described in the first aspect and different implementation methods of the first aspect are implemented.

[0075] In addition, the technical effects brought about by any implementation method of the second to fourth aspects can refer to the technical effects brought about by the first aspect and different implementation methods of the first aspect, and will not be repeated here. BRIEF DESCRIPTION OF THE DRAWINGS

[0076] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative labor.

[0077] Figure 1 A schematic diagram of an application scenario provided for an embodiment of the present application;

[0078] Figure 2 A schematic diagram of a flow chart of a method for detecting root cause of a signal defect provided in an embodiment of the present application;

[0079] Figure 3 A schematic diagram of a network structure of a multi-label classification model provided in an embodiment of the present application;

[0080] Figure 4 A schematic diagram of a process for detecting the root cause of a defect of a signal provided in an embodiment of the present application;

[0081] Figure 5 A schematic diagram of a device for detecting a root cause of a signal defect provided by an embodiment of the present application;

[0082] Figure 6 A schematic diagram of an execution device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0083] The following will be combined with the drawings in the embodiments of the present application to clearly and completely describe the technical solutions in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. The components of the embodiments of the present application described and shown in the drawings here can be arranged and designed in various different configurations.

[0084] Therefore, the following detailed description of the embodiments of the present application provided in the accompanying drawings is not intended to limit the scope of the present application for which protection is sought, but merely represents selected embodiments of the present application. Based on the embodiments of the present application, all other embodiments obtained by those skilled in the art without creative work are within the scope of protection of the present application.

[0085] It should be noted that relational terms such as "first" and "second" are used only to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device. In the absence of further restrictions, the elements defined by the sentence "comprise a ..." do not exclude the existence of other identical elements in the process, method, article or device including the elements.

[0086] In order to facilitate understanding of the solution proposed in this application, the technical terms involved in this application are explained below.

[0087] The Eye Diagram is a visual analysis tool that can intuitively evaluate the quality of the signal by displaying the multi-cycle superimposed waveform of the signal at the receiving end. By analyzing the characteristic parameters of the eye diagram (such as eye width, eye height, crossing point percentage, jitter, rise time or fall time, quality factor and duty cycle distortion, etc.), the quality of signal transmission can be judged and whether there is a signal integrity problem can be identified. Among them, the eye width (Eye Width) represents the horizontal width of the eye opening in the signal eye diagram, reflecting the time domain stability of the signal. The eye height (Eye Height) represents the vertical height of the eye opening in the eye diagram, reflecting the voltage amplitude of the signal. The crossing point percentage (Crossing Point Percentage) represents the position percentage of the signal crossing point, reflecting the timing margin of the signal. Jitter (Jitter) represents the time jitter of the signal, reflecting the timing instability of the signal. The rise time (Rise Time) and the fall time (Fall Time) are used to represent the transition time of the signal from low level to high level and the transition time of the signal from high level to low level, reflecting the signal jump speed. The quality factor is used to measure the quality and bit error rate of the signal. The duty cycle distortion is used to indicate the distortion of the time ratio of the high level and low level of the signal.

[0088] Bidirectional Encoder Representations from Transformers (BERT) is a pre-trained language model that can learn the semantic and contextual information of a language from a large amount of text data through a bidirectional encoder. BERT performs well in processing text embedding and natural language processing tasks, and can be used to convert descriptive text into high-dimensional feature vectors.

[0089] Lasso regression is a linear regression model that introduces L1 regularization terms into the regression model for feature selection and parameter estimation. Lasso regression can effectively select important features from high-dimensional data, reduce the complexity of the model, and avoid overfitting.

[0090] Transformer is a neural network architecture based on the self-attention mechanism, which was first used in natural language processing tasks. It consists of an encoder and a decoder and can efficiently process sequence data. Transformer has advantages in processing long sequences and capturing global dependencies, and is widely used in tasks such as machine translation and text generation.

[0091] The multi-head attention mechanism is one of the core components of the Transformer architecture. It captures the dependencies between different positions in the input sequence through multiple attention heads in parallel, enhancing the model's ability to express complex features. Each attention head performs attention calculations independently, and then concatenates the results to obtain the final attention representation through linear transformation.

[0092] High-speed signal transmission is one of the core technologies in modern electronic systems and is widely used in data centers, communication systems, computer networks and other fields. As data transmission rates continue to increase, the signal integrity issue of high-speed signal transmission has become particularly important. If the signal is interfered and distorted during transmission, the signal quality will be reduced, which will in turn affect the reliability and stability of data transmission.

[0093] The eye diagram is a visual analysis tool that can intuitively evaluate the quality of the signal by displaying the multi-cycle superimposed waveform of the signal at the receiving end. By analyzing the characteristic parameters of the eye diagram (such as eye width, eye height, percentage of crossing points, jitter, rise time or fall time, etc.), the quality of signal transmission can be judged and possible signal integrity problems can be identified. One of the existing signal detection methods is to superimpose the standard signal eye diagram template with the eye diagram of the signal to be detected, and then determine whether the signal waveform meets the specific standard. Alternatively, the eye diagram is classified through deep learning to determine the quality of the signal to be detected. However, the above method can only perform compliance detection, that is, it can only determine the overall evaluation of the quality of the signal to be detected, but it cannot identify the specific root cause of the signal defect problem, that is, it is impossible to know the specific cause of the defect.

[0094] In response to the above problems, the embodiments of the present application provide a signal defect root cause detection method, device and equipment, which obtains multiple measurement data of the eye diagram corresponding to the signal to be detected, and then vectorizes the multiple measurement data to obtain multiple measurement data vectors, and then performs feature screening to screen out the target measurement data vector related to the defect, which can avoid the interference of irrelevant attributes on defect identification and improve the accuracy and reliability of subsequent classification. Furthermore, feature extraction and multi-label classification can be performed on the target measurement data vector to determine the detection result of the defect root cause of the signal to be detected, and then the specific root cause of the signal quality problem can be identified in a fine-grained manner, so that users can perform targeted optimization and troubleshooting.

[0095] The following is a brief introduction to the application scenarios to which the technical solutions of the embodiments of the present application can be applied. It should be noted that the application scenarios introduced below are only used to illustrate the embodiments of the present application and are not limited. In specific implementation, the technical solutions provided by the embodiments of the present application can be flexibly applied according to actual needs.

[0096] like Figure 1 As shown, Figure 1 The application scenario of the signal defect root cause detection method provided by the present application is exemplarily shown, including an oscilloscope 100, a server 200 and a terminal device 300. The server 200 can be implemented by a physical server or a virtual server. The server can be implemented by a single server or a server cluster composed of multiple servers.

[0097] In some embodiments, the signal to be detected can be processed by the oscilloscope 100 to determine multiple measurement data of the eye diagram of the signal to be detected. In other embodiments, the server 200 can receive the detection task of the root cause of the defect of the signal sent by the terminal device 300, and perform the detection task of the root cause of the defect of the signal. The specific implementation is as follows: the signal to be detected can be processed by the server 200 to determine multiple measurement data of the eye diagram of the signal to be detected. Furthermore, the server 200 can perform vectorization processing, feature screening and feature extraction on multiple measurement data to determine the detection result of the root cause of the defect of the signal to be detected. Then, the detection result of the root cause of the defect of the signal to be detected is sent to the terminal device 300.

[0098] In some scenarios, the server 200 may also receive training tasks of a multi-label classification model and a defect feature screening model sent by the terminal device 300 to train the multi-label classification model and the defect feature screening model.

[0099] In some embodiments, the server 200 may send the trained multi-label classification model and defect feature screening model to the terminal device 300, so as to detect the defect root cause of the plurality of measurement data through the terminal device 300, and determine the detection result of the defect root cause of the signal to be detected. In a possible implementation, the terminal device 300 may be a terminal device having a display unit, and the detection result of the defect root cause of the signal to be detected may be displayed on the display interface of the terminal device 300.

[0100] See also Figure 2 FIG. 1 is a flow chart of a method for detecting a root cause of a signal defect provided by an embodiment of the present application. The method can be performed by Figure 1 The server 200 in the embodiment can also be executed by the terminal device 300. The specific process is as follows:

[0101] S201, obtaining a plurality of measurement data of an eye diagram corresponding to a signal to be detected, and performing vector processing on the plurality of measurement data to obtain a plurality of measurement data vectors.

[0102] In some embodiments, multiple measurement data of the eye diagram corresponding to the signal to be detected can be obtained from the oscilloscope, where the measurement data can also be called measurement item analysis data, including eye width, eye height, crossing point percentage, jitter, rise time and fall time, quality factor and duty cycle distortion.

[0103] In some embodiments, after obtaining multiple measurement data of the eye diagram from the oscilloscope, the multiple measurement data may be preprocessed to check whether there are missing values ​​in the multiple measurement data to ensure data integrity. In some scenarios, if one or more data are missing in the multiple measurement data of the eye diagram of the signal to be detected, the signal to be detected may be analyzed again by the oscilloscope to determine the multiple measurement data of the eye diagram to ensure the integrity of the multiple measurement data.

[0104] In a possible implementation, multiple measurement data are vectorized to obtain multiple measurement data vectors, which can be implemented as follows: for each measurement data, based on the comparison result between the measurement data and the preset standard data, a measurement text is generated; wherein the measurement text includes the measurement data, the preset standard data and the comparison result. Further, for each measurement text, the measurement text can be vectorized by a language model to obtain a measurement data vector.

[0105] Specifically, the standard value of each measurement data can be set according to the industry standard or product specification standard. The actual measurement data can be compared with the standard value to generate measurement text, i.e., descriptive text. As an example, the actual measurement data of the eye width is 8.512ps, and the standard value is 9.688ps. Then the descriptive text is "the actual value of the eye width is 8.512ps, which is lower than the standard value of 9.688ps, and the result is not good."

[0106] Furthermore, a language model can be used to vectorize each measurement text to obtain a measurement data vector. The language model can use a BERT model. Multiple measurement texts are respectively input into the BERT model for vectorization processing to obtain a measurement data vector corresponding to each measurement data. The measurement data vector is a high-dimensional vector that represents the semantic information of the measurement text.

[0107] S202: Perform feature screening on multiple measurement data vectors to determine at least one target measurement data vector related to the defect among the multiple measurement data vectors.

[0108] In some scenarios, since different defect root causes may be caused by different measurement data, multiple measurement data vectors can be screened through a defect feature screening model to filter out measurement data vectors that are not related to the defect and retain measurement data vectors that are related to the defect.

[0109] In some embodiments, multiple measurement data vectors may be input into a defect feature screening model, and feature screening may be performed on the multiple measurement data vectors through the defect feature screening model to determine at least one target measurement data vector related to the defect among the multiple measurement data vectors.

[0110] As an example, the Lasso regression model can be used to perform feature screening on multiple measurement data vectors, screen out target measurement data vectors related to defects, and filter out measurement feature vectors that are not related to defects. Specifically, based on the coefficients of the Lasso regression model, the measurement data vectors related to defects, that is, the measurement data vectors that have a significant contribution to the identification of the root cause of the defect, can be screened out, and the measurement data vectors with coefficients of zero or close to zero can be ignored.

[0111] S203, extracting features from at least one target measurement data vector to obtain a defect feature matrix, and determining a detection result of a defect root cause of the signal to be detected according to the defect feature matrix.

[0112] In some embodiments, when extracting features from at least one target measurement data vector to obtain a defect feature matrix, the at least one target measurement data vector may be input into a multi-label classification model, and features may be extracted from the at least one target measurement data vector by an encoder in the multi-label classification model to obtain a defect feature matrix. Furthermore, the defect feature matrix may be subjected to multi-label classification by a multi-label classification model to determine the probability value of each defect root cause of the signal to be detected. Then, for each defect root cause probability value, the detection result of the defect root cause corresponding to the signal to be detected is determined; wherein the detection result includes at least one target probability value, and the target probability value is a probability value greater than a set probability threshold.

[0113] As an example, the network structure of a multi-label classification model is as follows Figure 3 As shown in the figure, it includes a Transformer layer, a fully connected layer and an output layer. Among them, the Transformer layer includes multiple encoders (Encoder), which can extract features from the input target measurement data vector to obtain a defect feature matrix. Among them, each encoder is composed of a multi-head attention mechanism and a feedforward neural network. Multi-head attention can calculate the correlation between different parts and enhance the model's ability to capture complex relationships between features, that is, effectively capture the dependencies between each position in the target measurement data vector to improve the classification accuracy of the model. However, since the nature of multi-head attention is linear, the feedforward neural network introduces nonlinear transformations to enable the model to learn and express more complex formulas and features.

[0114] The output of the Transformer layer is a defect feature matrix, in which the eigenvalues ​​of each row are used to determine the probability of each defect root cause. In order to use the defect feature matrix for multi-label classification tasks, a fully connected layer (i.e., Dense layer) can be added after the Transformer layer to convert the defect feature matrix into the fixed-dimensional output required by the classifier.

[0115] Output layer: The linear combination result is passed through the Sigmoid activation function to generate the probability values ​​of multiple defect root cause labels to determine the detection result of the defect root cause.

[0116] Assuming that the defect root cause categories include signal noise, intersymbol interference, clock jitter, bandwidth limitation, power supply noise, impedance mismatch, and transmission medium loss, the output results of the multi-label classification model include the following cases:

[0117] In a possible implementation, the output result of the multi-label classification model is the probability values ​​corresponding to the seven defect root causes, and each probability value corresponds to the possibility of the defect root cause. Further, the detection result of the defect root cause of the signal to be detected can be determined according to the probability values ​​corresponding to the seven defect root causes. Specifically, when the probability value of a certain category output is greater than the set threshold, it can be determined that the signal to be detected has the defect root cause corresponding to the probability value. As an example, the output result of the multi-label classification model is {0.7, 0.3, 0.4, 0.1, 0.3, 0.6, 0.8}, then the probability value of signal noise is 0.7, the probability value of inter-symbol interference is 0.3, the probability value of clock jitter is 0.4, the probability value of bandwidth limitation is 0.1, and the probability of power supply noise is 0.3. If the threshold is set to 0.5, the probability values ​​of signal noise, impedance mismatch and transmission medium loss are all greater than the set threshold 0.5, then it can be determined that the signal to be detected has the defect root causes of signal noise, impedance mismatch and transmission medium loss.

[0118] In another possible implementation, the output result of the multi-label classification model is the defect root cause category of the signal to be detected. After the multi-label classification model determines the probability values ​​corresponding to each defect root cause category, the defect root cause category of the signal to be detected can be further determined according to the size of the probability value. As an example, after the probability value of each defect root cause category is determined, it can be compared with the set threshold to determine the defect root cause category of the signal to be detected, and the existing defect root cause category is output. In some scenarios, the defect root cause categories correspond to values ​​0-6 respectively. After the defect root cause category of the signal to be detected is determined according to the probability value, the corresponding value can be output. As an example, signal noise, inter-symbol interference, clock jitter, bandwidth limitation, power supply noise, impedance mismatch and transmission medium loss correspond to 0-6 respectively. If the result output by the multi-label classification model is {0, 2, 5}, it is determined that the defect root cause of the signal to be detected includes signal noise, clock jitter and impedance mismatch.

[0119] In some embodiments, after determining the detection result of the defect root cause of the signal to be detected, the target solution corresponding to the detection result of the defect root cause of the signal to be detected can also be determined based on the pre-configured association relationship between the defect root cause and the solution. Further, the detection result of the defect root cause of the signal to be detected and the corresponding target solution can be displayed on the display interface.

[0120] In some embodiments, the defect feature screening model can be obtained by the following training method:

[0121] A first training sample set is obtained, the first training sample set includes a plurality of first samples, the first samples include a plurality of first sample measurement data of an eye diagram corresponding to a first sample signal, and a defect label set corresponding to each first sample measurement data, the defect label set includes a correlation label between each first sample measurement data and a defect. As an example, the correlation label may be a correlation that causes the existence of a defect, and the correlation may be annotated by a knowledge expert. In some scenarios, the correlation label may be a correlation value, a first measurement sample data includes eye width, eye height, crossover percentage, jitter, rise time and fall time, quality factor and duty cycle distortion, and its defect label set is {0.3, 0.7, 0.4, 0.6, 0.1, 0.3, 0.8}. When the correlation value in the defect label set is greater than a certain threshold, it is considered that the measurement sample data will cause a defect. Assuming that the threshold is 0.5, it is determined that the eye height, jitter and duty cycle distortion will cause a defect. In other scenarios, the correlation label may be a label category, including 0 and 1. Among them, 0 means that no defect will occur, and 1 means that an over-value defect will occur. A first measurement sample data includes eye width, eye height, crossing point percentage, jitter, rise time and fall time, quality factor and duty cycle distortion, and its defect label set is {0, 0, 1, 1, 1, 0, 1}. It is determined that the crossing point percentage, jitter, rise time and fall time and duty cycle distortion will cause defects.

[0122] Furthermore, for the first sample signal in each first sample, vectorization processing may be performed on a plurality of first sample measurement data corresponding to the first sample signal to obtain a plurality of first sample measurement data vectors;

[0123] Repeat the following operations: for each first sample, screen multiple first sample measurement data vectors through the first network model to obtain a prediction result, and update the network parameters of the first network model based on the prediction result and the loss value determined by the defect label set. The prediction result is used to characterize a set of first sample measurement data vectors related to the defect.

[0124] When the first set condition is met, the updated first network model is used as the defect feature screening model, wherein the first set condition is that the determined loss value is less than a first set loss threshold.

[0125] Specifically, the present application can select the Lasso model to screen features. The measurement data vector is used as the feature input, and the related measurement item factors that cause the current defect are used as the target variable. The model is trained by setting the regularization parameter λ. For the Lasso model, the regularization parameter controls the strictness of feature selection. The larger the value, the fewer features are selected. According to the coefficients of the Lasso model, features that contribute significantly to defect identification are selected, and features with coefficients of zero or close to zero are ignored.

[0126] In some embodiments, a multi-label classification model is obtained by the following training method:

[0127] Acquire a second training sample set, the second training sample set includes a plurality of second samples, the second samples include a plurality of second measurement data of an eye diagram corresponding to a second sample signal, and a defect root cause label set corresponding to the second sample signal; the defect root cause label set is used to indicate at least one defect root cause of a defect in the second signal;

[0128] For the second sample signal in each second sample, vectorization processing is performed on a plurality of second sample measurement data corresponding to the second sample signal to obtain a plurality of second sample measurement data vectors;

[0129] Repeat the following operations: extract features from a plurality of second sample measurement data vectors based on an encoder in the second network model, and determine a predicted defect root cause set corresponding to the second sample signal; update network parameters of the second network model based on a loss value determined by the predicted defect root cause set and the defect root cause label set;

[0130] When the second set condition is met, the updated second network model is used as the multi-label classification model. The second set condition is that the loss value determined in the current training round is less than the second set loss threshold. In some other embodiments, the second set condition can also be that the number of training times meets the set iteration threshold.

[0131] Specifically, seven defect root cause categories are identified: signal noise, inter-symbol interference, clock jitter, bandwidth limitation, power supply noise, impedance mismatch, and transmission medium loss. Furthermore, the defect root cause category corresponding to each second sample is marked according to the abnormality of the measured data. The defect root cause label of each eye diagram sample is recorded. The data set is divided into a training set, a validation set, and a test set, with a ratio of 70%, 15%, and 15%. Ensure that each type of defect root cause is fully representative in the data set to avoid data bias in model training. Next, the learning rate can be set to 1e -5 The batch size is set to 32, the training rounds are set to 20, and the Adam optimizer is used to better handle large-scale text data. Binary Cross-Entropy with Logits Loss is used as the loss function to adapt to multi-label classification tasks.

[0132] During the training process, the training set is used for model training, and the prediction results are calculated through forward propagation. The loss is calculated based on the predicted defect root cause set and the defect root cause label set. The model weights are automatically adjusted using back propagation updates. According to the specific situation, after every 5-10 epochs of training, the hyperparameters are adjusted according to the validation set results to optimize the model performance.

[0133] See also Figure 4 As shown, an example of a method for detecting the root cause of a defect of a signal involved in this application is described, and the specific process is as follows:

[0134] Step S401: collect measurement item analysis data and pre-process.

[0135] Specifically, measurement item analysis data of the signal eye diagram can be obtained from the oscilloscope, that is, measurement data including eye width, eye height, crossing point percentage, jitter, rise time and fall time, quality factor and duty cycle distortion.

[0136] Furthermore, the collected measurement item analysis data may be preprocessed to check the actual measurement item analysis data, thereby ensuring the integrity of the data.

[0137] Step S402: converting and vectorizing the collected measurement item analysis data into text data.

[0138] Specifically, the specification value (SPEC) of each measurement item analysis data can be set according to product specifications or industry standards. The actual measurement value is compared with the specification value to generate descriptive text. For example, the descriptive text can be "The actual value of the eye width is 8.512ps, which is lower than the specification value of 9.688ps, and the result is not good."

[0139] Furthermore, the generated descriptive text can be input into the BERT model for vectorization processing to obtain the measurement item analysis data vector corresponding to the descriptive text.

[0140] Step S403: performing feature selection on the measurement item analysis data vector.

[0141] In some embodiments, since the root cause of a defect is not always related to all the measurement item analysis data, it is necessary to select features, and then select features that contribute significantly to defect identification, and ignore features that are irrelevant. Specifically, the features can be sparsely processed through the Lasso model to automatically filter out features related to the defect.

[0142] Step S404, identifying the defect root cause on the filtered measurement item analysis data vector, and determining the detection result of the defect root cause.

[0143] Specifically, the filtered measurement item analysis data vector may be input into a multi-label classification model to identify the defect root cause, so as to determine the defect root cause of the signal to be detected.

[0144] Step S405 , determining a target solution according to the detection result of the defect root cause, and displaying the detection result of the defect root cause and the target solution on a display interface.

[0145] Based on the above solution, through feature screening and defect root cause identification, the specific root cause of the eye diagram defect problem can be identified in fine granularity, providing a more targeted optimization solution. By using multi-dimensional attributes and feature screening technology, the interference of irrelevant data on the identification of defect root cause categories can be avoided, and the accuracy and reliability of the detection results can be improved. The use of a multi-label classification model for defect root cause detection can effectively identify complex eye diagram defect root cause problems. In addition, the detected defect root cause can be automatically provided to the corresponding solution, which can facilitate users to quickly troubleshoot and optimize the problem.

[0146] Based on the same technical concept, see Figure 5 As shown, a signal defect root cause detection device 500 provided in an embodiment of the present application can perform any step in the signal defect root cause detection method described above. To avoid repetition, it is not described here. The device 500 includes an acquisition module 501, a processing module 502 and a screening module 503.

[0147] An acquisition module 501 is used to acquire a plurality of measurement data of an eye diagram corresponding to a signal to be detected;

[0148] A processing module 502 is used to perform vectorization processing on the multiple measurement data to obtain multiple measurement data vectors;

[0149] A screening module 503 is used to perform feature screening on the multiple measurement data vectors to determine at least one target measurement data vector related to the defect among the multiple measurement data vectors;

[0150] The processing module 502 is further configured to extract features from the at least one target measurement data vector to obtain a defect feature matrix, and determine a detection result of a defect root cause of the signal to be detected based on the defect feature matrix.

[0151] In a possible implementation, the processing module 502, when performing vectorization processing on the multiple measurement data to obtain multiple measurement data vectors, is specifically configured to:

[0152] For each measurement data, a measurement text is generated based on a comparison result between the measurement data and preset standard data; wherein the measurement text includes the measurement data, the preset standard data and the comparison result;

[0153] For each measurement text, vectorization is performed on the measurement text through a language model to obtain a measurement data vector.

[0154] In a possible implementation, the screening module 503, when performing feature screening on the multiple measurement data vectors to determine at least one target measurement data vector related to a defect in the multiple measurement data vectors, is specifically configured to:

[0155] The multiple measurement data vectors are input into a defect feature screening model, and feature screening is performed on the multiple measurement data vectors through the defect feature screening model to determine at least one target measurement data vector related to the defect among the multiple measurement data vectors.

[0156] In a possible implementation, the processing module 502 is specifically configured to:

[0157] The defect feature screening model is obtained through the following training method:

[0158] Acquire a first training sample set, wherein the first training sample set includes a plurality of first samples, wherein the first samples include a plurality of first sample measurement data of an eye diagram corresponding to a first sample signal, and a defect label set corresponding to each first sample measurement data, wherein the defect label set includes a correlation label between each first sample measurement data and a defect;

[0159] For a first sample signal in each first sample, vectorization processing is performed on a plurality of first sample measurement data corresponding to the first sample signal to obtain a plurality of first sample measurement data vectors;

[0160] Repeat the following operations: for each first sample, screen the plurality of first sample measurement data vectors through the first network model to obtain a prediction result, wherein the prediction result is used to characterize a set of first sample measurement data vectors related to the defect; and update the network parameters of the first network model based on the prediction result and the loss value determined by the defect label set;

[0161] When the first set condition is met, the updated first network model is used as a defect feature screening model;

[0162] Among them, the first setting condition is that the determined loss value is less than a first set loss threshold.

[0163] In a possible implementation, the processing module 502, when extracting features from the at least one target measurement data vector to obtain a defect feature matrix, is specifically configured to:

[0164] Inputting the at least one target measurement data vector into a multi-label classification model, and performing feature extraction on the at least one target measurement data vector through an encoder in the multi-label classification model to obtain a defect feature matrix;

[0165] The processing module 502, when determining the detection result of the signal to be detected for each defect root cause label according to the defect feature matrix, is specifically used to:

[0166] Performing multi-label classification on the defect feature matrix through a multi-label classification model to determine the probability value of the signal to be detected for each defect root cause;

[0167] According to the probability value of each defect root cause, a detection result of the defect root cause corresponding to the signal to be detected is determined; wherein the detection result includes at least one target probability value, and the target probability value is a probability value greater than a set probability threshold.

[0168] In a possible implementation, the processing module 502 is specifically configured to:

[0169] The multi-label classification model is obtained through the following training method:

[0170] Acquire a second training sample set, wherein the second training sample set includes a plurality of second samples, the second samples include a plurality of second measurement data of an eye diagram corresponding to a second sample signal, and a defect root cause label set corresponding to the second sample signal; the defect root cause label set is used to indicate at least one defect root cause of a defect in the second signal;

[0171] For the second sample signal in each second sample, vectorization processing is performed on a plurality of second sample measurement data corresponding to the second sample signal to obtain a plurality of second sample measurement data vectors;

[0172] Repeat the following operations: extract features from the plurality of second sample measurement data vectors based on an encoder in the second network model, and determine a predicted defect root cause set corresponding to the second sample signal; update network parameters of the second network model based on a loss value determined by the predicted defect root cause set and the defect root cause label set;

[0173] When the second set condition is met, the updated second network model is used as a multi-label classification model;

[0174] Among them, the second set condition is that the loss value determined in the current training round is less than the second set loss threshold.

[0175] In a possible implementation, the processing module 502 is further configured to:

[0176] Determine a target solution corresponding to a detection result of a defect root cause of a signal to be detected based on a pre-configured association relationship between a defect root cause and a solution;

[0177] The detection result of the defect root cause of the signal to be detected and the corresponding target solution are displayed on the display interface.

[0178] Based on the same technical concept, the embodiment of the present application provides an execution device 600, which can implement any step of the signal defect root cause detection method discussed above, please refer to Figure 6 The device includes a memory 601 and a processor 602 .

[0179] The memory 601 is used to store program instructions;

[0180] The processor 602 is used to call the program instructions stored in the memory and execute the method for detecting the defect root cause of the signal according to the obtained program.

[0181] In the embodiment of the present application, the processor 602 may be a general-purpose processor, a digital signal processor, an application-specific integrated circuit, a field programmable gate array or other programmable logic device, a discrete gate or transistor logic device, or a discrete hardware component, and may implement or execute the methods, steps, and logic block diagrams disclosed in the embodiment of the present application. The general-purpose processor may be a microprocessor or any conventional processor, etc. The steps of the method disclosed in the embodiment of the present application may be directly embodied as being executed by a hardware processor, or may be executed by a combination of hardware and software modules in the processor.

[0182] Memory 601, as a non-volatile computer-readable storage medium, can be used to store non-volatile software programs, non-volatile computer executable programs and modules. Memory 601 may include at least one type of storage medium, such as flash memory, hard disk, multimedia card, card-type memory, random access memory (Random Access Memory, RAM), static random access memory (Static Random Access Memory, SRAM), programmable read-only memory (Programmable Read Only Memory, PROM), read-only memory (Read Only Memory, ROM), electrically erasable programmable read-only memory (Electrically Erasable Programmable Read-Only Memory, EEPROM), magnetic storage, disk, optical disk, etc. Memory 601 is any other medium that can be used to carry or store a desired program code in the form of an instruction or data structure and can be accessed by a computer, but is not limited thereto. The memory 601 in the embodiment of the present application can also be a circuit or any other device that can realize a storage function, for storing program instructions and / or data.

[0183] Based on the same technical concept, the embodiment of the present application provides a computer-readable storage medium, including: computer instructions, when the computer instructions are executed on a computer, the computer executes the method for detecting the root cause of a defect of a signal as discussed above. Since the principle of solving the problem by the above-mentioned computer-readable storage medium is similar to the method for detecting the root cause of a defect of a signal, the implementation of the above-mentioned computer-readable storage medium can refer to the implementation of the method, and the repeated parts will not be repeated.

[0184] Those skilled in the art will appreciate that the embodiments of the present application may be provided as methods, systems, or computer program products. Therefore, the present application may adopt the form of a complete hardware embodiment, a complete software embodiment, or an embodiment in combination with software and hardware. Moreover, the present application may adopt the form of a computer program product implemented in one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) that include computer-usable program code.

[0185] The present application is described with reference to the flowcharts and / or block diagrams of the methods, devices (systems), and computer program products according to the present application. It should be understood that each process and / or box in the flowchart and / or block diagram, as well as the combination of the processes and / or boxes in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0186] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0187] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process. Figure 1 A process or multiple processes and / or boxes Figure 1 A step that specifies a function in one or more boxes.

[0188] Obviously, those skilled in the art can make various changes and modifications to the present application without departing from the spirit and scope of the present application. Thus, if these modifications and variations of the present application fall within the scope of the claims of the present application and their equivalents, the present application is also intended to include these modifications and variations.

Claims

1. A method for detecting root causes of signal defects, characterized in that: include: Acquire multiple measurement data of the eye diagram corresponding to the signal to be detected, and perform vector processing on the multiple measurement data to obtain multiple measurement data vectors; Performing feature screening on the multiple measurement data vectors to determine at least one target measurement data vector related to the defect among the multiple measurement data vectors; Feature extraction is performed on the at least one target measurement data vector to obtain a defect feature matrix, and a detection result of the defect root cause of the signal to be detected is determined according to the defect feature matrix.

2. The method according to claim 1, characterized in that The vectorization processing of the plurality of measurement data to obtain a plurality of measurement data vectors includes: For each measurement data, a measurement text is generated based on a comparison result between the measurement data and preset standard data; wherein the measurement text includes the measurement data, the preset standard data and the comparison result; For each measurement text, vectorization is performed on the measurement text through a language model to obtain a measurement data vector.

3. The method according to claim 1, characterized in that The performing feature screening on the multiple measurement data vectors to determine at least one target measurement data vector related to the defect in the multiple measurement data vectors includes: The multiple measurement data vectors are input into a defect feature screening model, and feature screening is performed on the multiple measurement data vectors through the defect feature screening model to determine at least one target measurement data vector related to the defect among the multiple measurement data vectors.

4. The method according to claim 3, characterized in that The defect feature screening model is obtained through the following training method: Acquire a first training sample set, wherein the first training sample set includes a plurality of first samples, wherein the first samples include a plurality of first sample measurement data of an eye diagram corresponding to a first sample signal, and a defect label set corresponding to each first sample measurement data, wherein the defect label set includes a correlation label between each first sample measurement data and a defect; For a first sample signal in each first sample, vectorization processing is performed on a plurality of first sample measurement data corresponding to the first sample signal to obtain a plurality of first sample measurement data vectors; Repeat the following operations: for each first sample, screen the plurality of first sample measurement data vectors through the first network model to obtain a prediction result, wherein the prediction result is used to characterize a set of first sample measurement data vectors related to the defect; and update the network parameters of the first network model based on the prediction result and the loss value determined by the defect label set; When the first set condition is met, the updated first network model is used as a defect feature screening model; Among them, the first setting condition is that the determined loss value is less than a first set loss threshold.

5. The method according to claim 1, characterized in that The step of extracting features from the at least one target measurement data vector to obtain a defect feature matrix comprises: Inputting the at least one target measurement data vector into a multi-label classification model, and performing feature extraction on the at least one target measurement data vector through an encoder in the multi-label classification model to obtain a defect feature matrix; The step of determining the detection result of the signal to be detected for each defect root cause label according to the defect feature matrix includes: Performing multi-label classification on the defect feature matrix through a multi-label classification model to determine the probability value of the signal to be detected for each defect root cause; According to the probability value of each defect root cause, a detection result of the defect root cause corresponding to the signal to be detected is determined; wherein the detection result includes at least one target probability value, and the target probability value is a probability value greater than a set probability threshold.

6. The method according to claim 5, characterized in that The multi-label classification model is obtained through the following training method: Acquire a second training sample set, wherein the second training sample set includes a plurality of second samples, the second samples include a plurality of second measurement data of an eye diagram corresponding to a second sample signal, and a defect root cause label set corresponding to the second sample signal; the defect root cause label set is used to indicate at least one defect root cause of a defect in the second signal; For the second sample signal in each second sample, vectorization processing is performed on a plurality of second sample measurement data corresponding to the second sample signal to obtain a plurality of second sample measurement data vectors; Repeat the following operations: extract features from the plurality of second sample measurement data vectors based on an encoder in the second network model, and determine a predicted defect root cause set corresponding to the second sample signal; update network parameters of the second network model based on a loss value determined by the predicted defect root cause set and the defect root cause label set; When the second set condition is met, the updated second network model is used as a multi-label classification model; Among them, the second set condition is that the loss value determined in the current training round is less than the second set loss threshold.

7. The method according to any one of claims 1 to 6, characterized in that: The method further comprises: Determine a target solution corresponding to a detection result of a defect root cause of a signal to be detected based on a pre-configured association relationship between a defect root cause and a solution; The detection result of the defect root cause of the signal to be detected and the corresponding target solution are displayed on the display interface.

8. A signal defect root cause detection device, characterized in that: include: An acquisition module, used for acquiring a plurality of measurement data of the eye diagram corresponding to the signal to be detected; A processing module, used for performing vector processing on the multiple measurement data to obtain multiple measurement data vectors; A screening module, configured to perform feature screening on the plurality of measurement data vectors, and determine at least one target measurement data vector related to a defect among the plurality of measurement data vectors; The processing module is further used to extract features from the at least one target measurement data vector to obtain a defect feature matrix, and determine a detection result of a defect root cause of the signal to be detected based on the defect feature matrix.

9. An execution device, characterized in that: include: A memory for storing program instructions; A processor, configured to obtain program instructions in the memory and execute the method according to any one of claims 1 to 7 according to the obtained program instructions.

10. A computer-readable storage medium, characterized in that: The computer-readable storage medium includes computer instructions, and when the computer instructions are executed by a computer, the method according to any one of claims 1 to 7 is implemented.

Images