Multi-channel clock buffer test system and method

Abstract

The invention discloses a multi-channel clock buffer test system and method, and relates to the technical field of electronic testing, and the method comprises the steps: collecting an original waveform data set, synchronizing a unified reference baseline, and generating a clock output data stream; performing frequency domain transformation on the clock output data stream, extracting dominant frequency drift, harmonic intensity change and phase fluctuation, and generating a spectrum disturbance feature vector sequence; performing recursive estimation on the spectrum disturbance feature vector sequence by using a Kalman filter, correcting state distribution in combination with a Bayesian probability updating method, and generating a health score drift trend vector; and performing multi-scale comparison on the health score drift trend vector, and identifying a channel entering a sub-health area by using an abnormal drift acceleration factor to generate an abnormal channel feature data set. According to the method, complex association between historical failure modes and features is mastered through the training time sequence prediction network, potential health score changes and failure time in the future are effectively predicted, and early warning is achieved.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of electronic testing, in particular to a multi-channel clock buffer testing system and method. BACKGROUND

[0002] With the rapid development of integrated circuit technology and high-speed digital systems, as an important device to ensure the synchronization and stability of multi-channel clock signals, performance testing and health state monitoring have gradually become one of the key technologies in the field of electronic testing. The traditional multi-channel clock buffer testing method mainly relies on waveform observation and static parameter measurement based on time domain sampling, such as frequency stability, duty cycle and delay deviation detection. Clock signals are usually collected by an oscilloscope or a logic analyzer, and time interval measurement, edge detection and other technologies are used to evaluate the quality of the clock.

[0003] Although some research and practice use frequency domain analysis to assist clock performance evaluation, they generally stop at static feature extraction and lack dynamic health scoring methods combining state estimation and probability inference technology, which cannot effectively deal with complex interference of multi-channel clock signals and abnormal drift phenomenon of non-homogeneous channels. In addition, the multi-channel resource allocation and testing scheduling strategy cannot be dynamically adjusted according to the real-time health state, resulting in difficulty in balancing testing efficiency and fault coverage ability. SUMMARY

[0004] In view of the above existing problems, the present application is proposed.

[0005] Therefore, the present application provides a multi-channel clock buffer testing method to solve the problem that testing scheduling cannot be dynamically optimized according to real-time health state.

[0006] To solve the above technical problems, the present application provides the following technical solutions:

[0007] In a first aspect, the present application provides a multi-channel clock buffer testing method, which includes collecting an original waveform data set and synchronously unifying a reference baseline to generate a clock output data stream; performing frequency domain transformation on the clock output data stream to extract main frequency drift, harmonic intensity change and phase fluctuation, and generate a frequency spectrum disturbance feature vector sequence; using a Kalman filter to recursively estimate the frequency spectrum disturbance feature vector sequence, and combining a Bayesian probability updating method to correct the state distribution and generate a health score drift trend vector; performing multi-scale comparison on the health score drift trend vector, and using an abnormal drift acceleration factor to identify channels entering the sub-health area and generate an abnormal channel feature data set; inputting the abnormal channel feature data set into a time series prediction network to predict future health score trends and potential failure time, and generate a failure warning result; performing weight fusion on the failure warning result and the health score drift trend vector to construct a multi-channel differentiated testing scheduling matrix, and generate a testing task table and a channel usage level list.

[0008] As a preferred scheme of the multi-channel clock buffer test method, the clock output data stream is generated, and the specific steps are as follows,

[0009] The channel number, time stamp sequence and corresponding voltage value are collected to form an original clock signal waveform and are stored in a structured manner to generate an original waveform data set;

[0010] The original clock signal waveform of each channel in the original waveform data set is aligned to a unified time reference to generate a synchronized clock waveform matrix, and a period stable segment is extracted and signal abnormal disturbance is removed to generate effective waveform data;

[0011] The effective waveform data is rearranged according to the channel number and sampling time sequence to generate a clock output data stream.

[0012] As a preferred scheme of the multi-channel clock buffer test method, the clock output data stream is generated, and the specific steps are as follows,

[0013] The clock output data stream is divided into equal-width time windows, and fast Fourier transform is performed to identify frequency domain amplitude and phase information to generate window-level spectrum data;

[0014] The main frequency position is identified in the window-level spectrum data, and the relative offset is measured to generate a main frequency drift vector;

[0015] The energy change of the integer multiple frequency points on both sides of the main frequency is extracted from the window-level spectrum data to generate a harmonic intensity change vector;

[0016] The phase information is differentially analyzed to extract high-frequency fluctuation and mutation features to generate a phase fluctuation vector;

[0017] The main frequency drift vector, the harmonic intensity change vector and the phase fluctuation vector are combined in time sequence to generate a spectrum disturbance feature vector sequence.

[0018] As a preferred scheme of the multi-channel clock buffer test method, the clock output data stream is generated, and the specific steps are as follows,

[0019] The spectrum disturbance feature vector sequence is input into a Kalman filter for recursive state prediction to generate a preliminary state estimation result;

[0020] The spectrum disturbance feature vector sequence is combined with the preliminary state estimation result, and the state distribution is corrected using the Bayesian probability updating method to generate a state posterior probability;

[0021] The health score and drift trend are derived according to the state posterior probability to generate a health score drift trend vector.

[0022] As a preferred scheme of the multi-path clock buffer test method, wherein: the generation of the abnormal channel feature data set, the specific steps are as follows,

[0023] The multi-time scale sliding window comparison analysis is performed on the health score drift trend vector to generate a score change rate and a drift acceleration rate sequence.

[0024] An abnormal drift acceleration factor is calculated using the score change rate and the drift acceleration rate sequence to calibrate the deviation degree of each channel, and a channel health state identifier is generated.

[0025] According to the channel health state identifier, the channels with continuously decreasing scores and abnormal drift acceleration are screened out to generate an abnormal channel feature data set.

[0026] As a preferred scheme of the multi-path clock buffer test method, wherein: the generation of the failure warning result, the specific steps are as follows,

[0027] The historical abnormal channel feature data set and the corresponding failure time label are used as training samples to optimize the time series prediction network to minimize the health score prediction error, and a trained time series prediction network is generated.

[0028] The abnormal channel feature data set is encoded into a prediction input sequence in chronological order to generate a time series modeling sample.

[0029] The time series modeling sample is input into the trained time series prediction network to predict the health score change in the future time period, and a health score prediction result is generated.

[0030] The score decline rate is calculated according to the health score prediction result, and the failure time point is calculated in combination with the drift trend to generate a failure warning result.

[0031] As a preferred scheme of the multi-path clock buffer test method, wherein: the generation of the test task table and the channel usage level list, the specific steps are as follows,

[0032] The failure warning result and the health score drift trend vector of the corresponding channel are weighted and fused to generate a channel comprehensive health evaluation index sequence.

[0033] According to the channel comprehensive health evaluation index sequence, a priority ranking matrix of each channel in different test periods is constructed to generate a multi-channel differentiated test scheduling matrix.

[0034] According to the multi-channel differentiated test scheduling matrix, test priorities and resource ratios are allocated to generate a test task table and a channel usage level list.

[0035] In a second aspect, the present application provides a multi-channel clock buffer test system, comprising a data acquisition module for acquiring an original waveform data set and synchronizing a unified reference baseline to generate a clock output data stream; a feature extraction module for performing frequency domain transformation on the clock output data stream to extract main frequency drift, harmonic intensity change and phase fluctuation and generate a spectrum disturbance feature vector sequence; a state update module for recursively estimating the spectrum disturbance feature vector sequence using a Kalman filter and correcting the state distribution using a Bayesian probability update method to generate a health score drift trend vector; an anomaly identification module for performing multi-scale comparison on the health score drift trend vector and identifying channels entering the sub-health zone using an anomaly drift acceleration factor to generate an abnormal channel feature data set; a failure prediction module for inputting the abnormal channel feature data set into a time series prediction network to predict future health score trends and potential failure times and generate a failure warning result; and a test scheduling module for performing weight fusion on the failure warning result and the health score drift trend vector to construct a multi-channel differentiated test scheduling matrix and generate a test task table and a channel usage level list.

[0036] In a third aspect, the present application provides a computer device comprising a memory and a processor, wherein the memory stores a computer program, and wherein the computer program, when executed by the processor, implements any step of the multi-channel clock buffer test method according to the first aspect of the present application.

[0037] In a fourth aspect, the present application provides a computer-readable storage medium having a computer program stored thereon, wherein the computer program, when executed by a processor, implements any step of the multi-channel clock buffer test method according to the first aspect of the present application.

[0038] The present application has the following advantages: by training the time series prediction network to master the complex correlation between historical failure patterns and features, the future potential health score change and failure time can be effectively predicted to realize early warning; not only can the imminent failure be actively identified to reduce the occurrence rate of sudden failure, but also the dynamic optimization basis can be provided for the test scheduling module to reasonably allocate test resources and avoid unnecessary repeated detection, thereby significantly improving the overall reliability and resource utilization. BRIEF DESCRIPTION OF DRAWINGS

[0039] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the drawings needed in the embodiment description will be briefly introduced. Obviously, the drawings in the following description are only some embodiments of the present application, and other drawings can be obtained by those skilled in the art without creative labor.

[0040] Fig. 1 The flowchart of the multi-channel clock buffer test method.

[0041] Fig. 2 A schematic diagram of a multi-channel clock buffer test system.

[0042] Fig. 3 A flowchart for generating a spectrum perturbation feature vector sequence.

[0043] Fig. 4 A flowchart for generating a test task table and a channel usage level list. DETAILED DESCRIPTION

[0044] In order to make the above objectives, features and advantages of the present application more apparent, the specific embodiments of the present application will be described in detail below with reference to the accompanying drawings.

[0045] In the following description, numerous specific details are set forth in order to provide a thorough understanding of the present application. However, it will be apparent to one skilled in the art that the present application can be practiced without the specific details given herein. In other instances, well-known methods have not been described in detail in order to avoid unnecessarily obscuring the present application. Therefore, the specific embodiments of the present application as disclosed below are not intended to limit the present application, but to illustrate it by reference to these specific embodiments.

[0046] Secondly, the term "one embodiment" or "an embodiment" as used herein means that a particular implementation can include a particular feature, structure, or characteristic, but every embodiment can not necessarily include the particular feature, structure, or characteristic. Furthermore, the following terms can be used throughout the specification and claims: includes, has, contains, etc. means that since a total of these terms represent various "inclusive" aspects, these terms are open-ended, and each of these terms does not mean "consisting of" or "consisting essentially of" when combined with other terms.

[0047] SUMMARY Figs. 1-4 For one embodiment of the present application, the embodiment provides a multi-channel clock buffer test method, comprising the following steps:

[0048] S1, collecting an original waveform data set, and synchronizing a unified reference baseline to generate a clock output data stream.

[0049] S1.1, collecting channel numbers, time stamp sequences, and corresponding voltage values, forming an original clock signal waveform, and structurally storing to generate an original waveform data set.

[0050] Specifically, by using a high-precision waveform acquisition instrument, each clock buffer output channel is sequentially sampled, and the channel number, corresponding time stamp information, and instantaneous voltage value of each sampling are recorded; the channel numbers are classified and sorted, and then arranged in ascending order of time stamp to form a three-tuple data structure of channel number, time stamp, and voltage value; the three-tuple data structure is arranged continuously to form a complete original clock signal waveform; finally, the original clock signal waveforms of all channels are respectively stored as structured data tables according to the channel numbers to generate an original waveform data set.

[0051] S1.2, align the original clock signal waveform of each channel in the original waveform data set to a unified time reference to generate a synchronized clock waveform matrix, and then extract the period stable segment and eliminate signal abnormal disturbance to generate effective waveform data.

[0052] Specifically, the original clock signal waveform of each channel is read from the original waveform data set, and based on the timestamp sequence corresponding to each channel, the original clock signal waveforms of all channels are resampled on a unified time scale by using linear interpolation method, and are aligned to a unified time reference. The resampled original clock signal waveform of each channel is arranged according to the channel number to form a two-dimensional matrix representing the channel and the time sequence, and a synchronized clock waveform matrix is generated. In the synchronized clock waveform matrix, a plurality of continuous periods are extracted by using a period detection method, the waveform stability index such as the maximum voltage difference or the zero-crossing point deviation in each period is calculated, and the time segment with the smallest change amplitude of the waveform stability index in the plurality of periods is selected as the period stable segment. In the period stable segment, abnormal mutation points are identified by statistics of the first derivative of the waveform voltage and the difference value of adjacent points, and the waveform data near the abnormal mutation points is eliminated, and the remaining continuous interval is reserved as effective waveform data.

[0053] S1.3, rearrange the effective waveform data according to the channel number and the sampling time sequence to generate a clock output data stream.

[0054] Specifically, the channel number, sampling time and voltage value information corresponding to each channel are extracted from the effective waveform data; the effective waveform data is classified and arranged according to the channel number, and arranged in ascending order of sampling time in each channel to form a time sequence structure of three fields of channel number, sampling time and voltage value; the rearranged effective waveform data of all channels is spliced according to the order of channel number to form a continuous time sequence of unified format; finally, the spliced effective waveform data is output in a structured format to generate a clock output data stream.

[0055] S2, perform frequency domain transformation on the clock output data stream to extract the main frequency drift, harmonic intensity change and phase fluctuation, and generate a frequency spectrum disturbance feature vector sequence.

[0056] S2.1, divide the clock output data stream into equal-width time windows, and perform fast Fourier transform to identify the frequency domain amplitude and phase information, and generate window-level frequency spectrum data.

[0057] Specifically, a continuous sample point sequence is read from the clock output data stream, and is divided into equal-width non-overlapping time windows according to a fixed time length, for example, each window contains 1024 sample points; the fast Fourier transform method is applied to the voltage value sequence in each time window to convert the time domain waveform data in the time window into a frequency domain complex spectrum; the amplitude value and the phase value corresponding to each frequency point are calculated from the frequency domain complex spectrum, the amplitude value is obtained by calculating the square root of the sum of the squares of the real part and the imaginary part, and the phase value is obtained by calculating the inverse tangent function; the frequency domain amplitude value and the phase value of all frequency points in each time window are stored in order of frequency, and output as window-level spectrum data.

[0058] S2.2, identify the main frequency position in the window-level spectrum data, measure the relative offset, and generate a main frequency drift vector.

[0059] Specifically, for the window-level spectrum data corresponding to each time window, the frequency domain amplitude values A(f i ) of all frequency points f i are counted to form a frequency-amplitude pair set, and the frequency point with the largest amplitude value is selected as the main frequency position of the current time window, and the main frequency position frequency value is represented as:

[0060]

[0061] Wherein, f max represents the frequency point with the largest frequency domain amplitude value in the current time window, f i represents the i-th frequency point in the window-level spectrum data, i represents the index number of the frequency point, A(f i ) represents the frequency domain amplitude value of the corresponding frequency point f i , and argmax represents the parameter value when a certain expression reaches the maximum value; that is, the frequency point f i with the largest A(f max ) is selected as the frequency component with the most concentrated energy in the current time window;

[0062] The main frequency position frequency value f max is subtracted from the theoretical reference frequency value f ref to obtain the frequency offset of the main frequency, and the expression is:

[0063] Δf=f max -f ref ;

[0064] Wherein, Δf represents the frequency offset of the main frequency in the current time window, and f ref represents the theoretical reference frequency value; the frequency offsets corresponding to all time windows are arranged in time sequence to form a time sequence of main frequency offsets; finally, the time sequence of main frequency offsets is structured and output to generate a main frequency drift vector.

[0065] It should also be noted that the theoretical reference frequency value is determined based on the equipment's factory setting or standard steady-state operating frequency, and serves as a reference benchmark for judging frequency deviation.

[0066] S2.3 Extract the energy changes of integer multiples of the main frequency on both sides of the window-level spectrum data to generate a harmonic intensity change vector.

[0067] Specifically, within each window-level spectrum data, based on the dominant frequency position in the dominant frequency drift vector, a sequence of integer multiples of the dominant frequency is determined, such as multiple harmonic frequency points like the second harmonic and third harmonic. Harmonic frequency points are located within the window-level spectrum data, and their corresponding frequency domain amplitude values ​​are read. A differential calculation method is used to calculate the difference between the amplitude value of each harmonic frequency point and the amplitude value at the current dominant frequency position, yielding the amplitude change of each harmonic relative to the dominant frequency. Then, the amplitude changes of each harmonic are arranged in order of harmonic order to form the harmonic intensity change vector corresponding to the current time window. Finally, all harmonic intensity change vectors corresponding to all time windows are combined in chronological order and output in a structured manner to generate the harmonic intensity change vector.

[0068] S2.4 Perform differential analysis on the phase information to extract high-frequency fluctuation and abrupt change features and generate a phase fluctuation vector.

[0069] Specifically, in the window-level spectral data of each time window, the phase information sequence is extracted, the difference between adjacent phase values ​​is calculated according to the sampling time order, and the phase change sequence is obtained using the first-order difference method; the sliding window variance calculation method is applied to the phase change sequence to statistically analyze the phase fluctuation intensity within the window and identify high-frequency fluctuation characteristics; at the same time, the amplitude threshold determination method of the phase change sequence is combined to screen out phase abrupt change points and extract abrupt change characteristics; the high-frequency fluctuation intensity and abrupt change characteristics are merged in chronological order to form a phase fluctuation vector; finally, the phase fluctuation vectors of all time windows are output in a structured manner to generate a complete phase fluctuation vector.

[0070] It should also be explained that the specific steps for screening phase change points are as follows: the phase change amplitude is extracted for each moment in the phase fluctuation vector, the phase change amplitude is compared with the phase change amplitude threshold, and the time point where the phase change amplitude is greater than the phase change amplitude threshold is selected as the phase change point; then, with the phase change point as the center, the phase change rate and change direction information of multiple adjacent time points in the phase fluctuation vector are extracted to generate phase change features.

[0071] The threshold for the amplitude of phase abrupt change can be determined by statistical analysis combined with the phase fluctuation vector in the spectral perturbation feature vector sequence. The specific steps are as follows:

[0072] First, the phase change amplitude values of all time points in the historical phase fluctuation vector are extracted to form a complete phase fluctuation amplitude dataset; the average value and standard deviation of the phase fluctuation amplitude dataset are calculated by using a statistical method, for example, the sample mean is calculated as the central reference value, and the sample standard deviation represents the natural fluctuation range; the threshold is set in the form of average value plus weighted standard deviation, for example, the phase mutation amplitude threshold is set as the value of "average value plus three times standard deviation", which is used to define the statistically abnormal fluctuation behavior; finally, as the phase mutation amplitude threshold, it is applied to the mutation feature recognition process of the phase fluctuation vector.

[0073] S2.5, the main frequency drift vector, the harmonic intensity change vector and the phase fluctuation vector are merged in time sequence to generate a spectrum disturbance feature vector sequence.

[0074] Specifically, the main frequency drift vector, the harmonic intensity change vector and the phase fluctuation vector are arranged in time sequence according to the sampling time, and the corresponding time points of the main frequency drift vector, the harmonic intensity change vector and the phase fluctuation vector are matched by using the time stamp alignment method to generate a three-dimensional feature matrix; then the three-dimensional feature matrix is spliced to combine the values of the main frequency drift vector, the harmonic intensity change vector and the phase fluctuation vector at each time point, forming a unified spectrum disturbance feature vector sequence; finally, the combined spectrum disturbance feature vector sequence is output in a structured format to generate a spectrum disturbance feature vector sequence.

[0075] S3, using Kalman filter, recursively estimating the spectrum disturbance feature vector sequence, and combining the Bayesian probability updating method to correct the state distribution, generating a health score drift trend vector.

[0076] S3.1, input the spectrum disturbance feature vector sequence into the Kalman filter to perform recursive state prediction and generate a preliminary state estimation result.

[0077] Specifically, the spectrum disturbance feature vector sequence is taken as the observation vector z k Input Kalman filter; the state estimation at the current time is predicted according to the state transition equation, and the expression is:

[0078]

[0079] Wherein, is the predicted state at time k, F is the state transition matrix, is the state estimation at time k-1, B is the control matrix, u k is the control input (such as zero if there is no control input), k is the time step index;

[0080] The calculation expression of the prediction error covariance matrix is:

[0081] P k|k-1= FP k-1|k-1 F T + Q

[0082] where P k|k-1 is the prediction error covariance matrix, Q is the process noise covariance, P k-1|k-1 is the error covariance matrix at time k-1, and T is the sampling time interval.

[0083] The Kalman gain is calculated, and the expression is as follows:

[0084] K k = P k|k-1 H T (HP k|k-1 H T + R) -1 ;

[0085] where K k is the Kalman gain matrix at time k, H is the observation matrix, R is the observation noise covariance, H T is the transpose matrix of the observation matrix H.

[0086] The state estimation is updated using the Kalman gain, and the expression is as follows:

[0087]

[0088] where x is the state estimation vector after observation correction at time k, and z k is the observation vector at time k.

[0089] The error covariance matrix is updated, and the expression is as follows:

[0090] P k|k = (I-K k H)P k|k-1 ;

[0091] where P k|k is the error covariance matrix after updating at time k, and I is the identity matrix. Finally, the state estimation vector after observation correction x is output to generate the preliminary state estimation result at the current time.

[0092] S3.2, combine the spectral disturbance feature vector sequence with the preliminary state estimation result, and use the Bayesian probability updating method to correct the state distribution to generate the state posterior probability.

[0093] Specifically, taking the spectrum disturbance feature vector sequence as the current observation data, taking the preliminary state estimation result as the current prediction state, judging the consistency between the prediction state and the current observation data by searching the feature segment closest to the prediction state trend in the observation data, and marking the observation reliability of the prediction state according to the consistency; then combining the appearance frequency of the prediction state at the last moment, referring to the consistency of state change at adjacent moments, the reliability of the current prediction state is graded, forming a set of unnormalized posterior probability scores; finally, the maximum value of the posterior probability score set is normalized, for example, the sum is 1, to generate the posterior probability; all posterior probabilities are normalized, for example, the sum is 1, to form a complete state posterior probability distribution; finally, the state posterior probability distribution is structured and output to generate the state posterior probability.

[0094] S3.3, according to the state posterior probability, derive the health score and the drift trend, and generate a health score drift trend vector.

[0095] Specifically, according to the state posterior probability, the health score is calculated by using the weighted summation method, and the drift trend value is calculated by using the time sequence change of the state posterior probability. In the example, the drift speed is determined by the ratio of the difference value of the health score at adjacent moments to the time interval. The calculated health score and drift trend are combined in time sequence to form a health score drift trend vector and structured output.

[0096] S4, compare the health score drift trend vector in multiple scales, and use the abnormal drift acceleration factor to identify the channel entering the sub-health area, and generate an abnormal channel feature data set.

[0097] S4.1, perform multi-time scale sliding window comparison analysis on the health score drift trend vector to generate a score change rate and a drift acceleration sequence.

[0098] Specifically, for the health score drift trend vector, different length sliding windows are set according to a plurality of preset time scales, the health score in each sliding window is linearly fitted respectively, and the fitting slope is calculated as the score change rate. Similarly, the change rate of the drift trend data in the same sliding window is calculated to obtain the drift acceleration sequence. The sliding window under each time scale is moved in turn to generate a score change rate and a drift acceleration sequence.

[0099] It should be noted that the preset multiple time scales set sliding windows of different lengths, and the specific steps are as follows: for the health score drift trend vector, a plurality of representative time scales are selected, for example, three scales of 5 sampling points, 10 sampling points and 20 sampling points are set as an example; taking each time scale as the length of the sliding window, the window data is extracted from the starting position of the health score drift trend vector according to the window length; for example, sliding by a step of 1 for each window position until the complete data sequence range is covered, and the setting of multiple length sliding windows is completed.

[0100] S4.2, calculate the abnormal drift acceleration factor by using the score change rate and the drift acceleration sequence, label the deviation degree of each channel, and generate the channel health state identification.

[0101] Specifically, for the score change rate and the drift acceleration sequence, the point-by-point multiplication method is used to calculate the product of the score change rate value and the drift acceleration value corresponding to each time point as the abnormal drift acceleration factor of the current time point; the sliding average method is used to calculate the average of the abnormal drift acceleration factors of each channel in the whole period to generate the average abnormal drift acceleration factor of the channel; then the minimum-maximum normalization method is used to normalize all the average abnormal drift acceleration factors of the channels to obtain the channel deviation degree score; finally, according to the numerical interval in which the channel deviation degree score is located, the grade is divided, for example, the channels with a score greater than 0.8 are highly correlated with the measured serious failure in the historical samples, and are therefore marked as serious deviation; the channels with a score between 0.5 and 0.8 show obvious abnormal behavior that does not immediately lead to failure, and are marked as moderate deviation; the channels with a score less than 0.5 have abnormal fluctuations but do not show degradation, and are marked as mild deviation, to generate the channel health state identification corresponding to each channel.

[0102] S4.3, according to the channel health state identification, the channels with continuously decreasing scores and abnormal drift acceleration are screened out to generate the abnormal channel feature data set.

[0103] Specifically, the channel health state identification is preliminarily screened, and the channels in the moderate deviation and serious deviation grades are selected as candidate channels; then the health score drift trend vector of each candidate channel is subjected to first-order difference processing, the sign sequence statistical method is used to identify the continuous negative growth section, and the time length and the decline amplitude of the continuous decline are calculated; then the drift behavior of the candidate channel is evaluated according to the change level of the drift acceleration sequence of each candidate channel in the specified time period, and the drift behavior that continuously rises and significantly deviates from the historical fluctuation range is evaluated; finally, the channel number and the corresponding feature data that meet the health score continuous decline feature and the drift acceleration abnormal feature are extracted and stored in a structured manner to generate the abnormal channel feature data set.

[0104] S5. Inputting the abnormal channel feature dataset into a time series prediction network to predict a future health score trend and a potential failure time, and generating a failure warning result.

[0105] S5.1. Optimizing the time series prediction network to minimize health score prediction error by taking the historical abnormal channel feature dataset and the corresponding failure time label as training samples, and generating a trained time series prediction network.

[0106] Specifically, the health score drift trend vector corresponding to each channel in the historical abnormal channel feature dataset is matched with the channel number, and the time series samples are reconstructed in chronological order. Each time series sample is aligned with the corresponding failure time label to form the training input and supervised output. The existing long short-term memory network method is used to construct the time series prediction network structure, and the loss function is set as the mean square error between the predicted health score and the actual score. Then, the back propagation algorithm is used to calculate the gradient of the loss function with respect to the network parameters, and the gradient descent method is used to iteratively update the parameters. The training is performed round by round until the error converges on the validation set or the preset number of rounds is met, and finally the trained time series prediction network is generated.

[0107] S5.2. Encoding the abnormal channel feature dataset in chronological order as a prediction input sequence to generate time series modeling samples.

[0108] Specifically, each channel number in the abnormal channel feature dataset is traversed. For each channel number, the corresponding health score drift trend vector, score change rate, drift speed sequence, and abnormal drift speed factor are extracted and arranged in chronological order. The multiple feature fields corresponding to each time point are combined into an input vector for one time step, and the consecutive time steps are combined into a fixed-length sequence segment. For each channel number, multiple non-overlapping or partially overlapping time series samples are generated and bound with the channel number, and finally the time series modeling samples encoded in chronological order are formed.

[0109] S5.3. Inputting the time series modeling samples into the trained time series prediction network to predict the future health score change, and generating a health score prediction result.

[0110] Specifically, the time series modeling samples corresponding to each channel number are read in sequence, and the consecutive time step input vectors in each sample are arranged in chronological order to form an input matrix. The input matrix is input into the trained time series prediction network to trigger the forward calculation process, and the calculation of the feature extraction layer, time dependence modeling layer, and output mapping layer is completed in sequence. The health score prediction values corresponding to multiple future time steps in the prediction output are obtained and arranged in chronological order to form a health score prediction sequence. The health score prediction sequence corresponding to each channel number is output in a structured format to generate a health score prediction result.

[0111] S5.4, calculate the score decline rate according to the health score prediction result, and combine the drift trend to calculate the failure time point, and generate the failure warning result.

[0112] Specifically, the continuous health score prediction value sequence corresponding to each channel number is extracted from the health score prediction result, the health score change amount at adjacent time points is calculated by using the difference method, and the score decline rate sequence is obtained by normalizing the unit time interval; the current score value and the historical score change direction are extracted from the health score drift trend vector to determine whether it is in a continuous decline trend; based on the predicted score decline rate sequence and the current score value, the linear extrapolation method is used to calculate the time point when the predicted score reaches failure as the failure time point of the channel; each channel number and the corresponding predicted failure time point are combined to generate the failure warning result.

[0113] Further, the current score value and the historical score change direction are extracted from the health score drift trend vector to determine whether it is in a continuous decline trend, and the specific steps are as follows:

[0114] The time index position corresponding to the current time is determined, and the score value at the index position in the health score drift trend vector is read as the current score value; a set number of time points are backtracked, for example, three consecutive time points are backtracked, and the score values at the corresponding index positions are read to form a continuous historical score sequence;

[0115] Then, the historical score sequence is calculated by using the first-order difference method, specifically: the current score value is subtracted from the score value at the last time point, and the score change amount between a plurality of adjacent time points is calculated in sequence; if a plurality of consecutive score change amounts are negative, it indicates that the score value is continuously decreasing, and it is determined that it is in a continuous decline trend. Otherwise, it is determined to be not in a continuous decline. The judgment result is used for subsequent determination of whether to need to calculate the failure time point.

[0116] S6, weight fusion is performed on the failure warning result and the health score drift trend vector to construct a multi-channel differentiated test scheduling matrix, and a test task table and a channel usage level list are generated.

[0117] S6.1, the failure warning result and the health score drift trend vector of the corresponding channel are weighted and fused to generate a channel comprehensive health assessment index sequence.

[0118] Specifically, the failure confidence value in the failure early warning result and each time point score value in the health score drift trend vector are read for each channel respectively, and a one-to-one correspondence is established in time sequence; the early warning weight factor and the score weight factor are set, for example, 0.6 and 0.4 are set as examples, which respectively represent the fusion proportion of the failure confidence and the score value; then for each time point, the channel comprehensive health evaluation index value is calculated, and all time points are sequentially weighted and calculated, and the obtained result is arranged in time sequence, and output as a channel comprehensive health evaluation index sequence.

[0119] It should also be noted that the setting step of the early warning weight factor and the score weight factor includes: statistical analysis of the accuracy of the channel health state prediction according to the failure early warning result and the health score drift trend in the historical data, calculation of the contribution rate of the early warning weight factor and the score weight factor to the actual fault prediction; combining the actual application requirements, determining the reasonable range of the early warning weight factor and the score weight factor, for example, the early warning weight factor is usually set between 0.5 and 0.7, and the score weight factor is adjusted to between 0.5 and 0.3, so that the early warning weight factor and the score weight factor add up to 1; finally, through cross-validation or offline simulation, the weight factor parameters are repeatedly adjusted to make the prediction effect of the comprehensive health evaluation index optimal, and the final values of the early warning weight factor and the score weight factor are determined.

[0120] S6.2, according to the channel comprehensive health evaluation index sequence, constructing a priority ranking matrix of each channel under different test periods to generate a multi-channel differentiated test scheduling matrix.

[0121] Specifically, according to the channel comprehensive health evaluation index sequence, the health evaluation value corresponding to each channel in different test periods is determined, and the priority of each channel in each test period is obtained according to the high and low of the health evaluation value; the priority arrangements of all test periods are summarized to form a multi-period channel priority ranking matrix, wherein the rows of the matrix represent the test periods, the columns represent the channel numbers, and the matrix elements represent the priority of the channel in the corresponding period; the multi-channel differentiated test scheduling matrix is generated according to the priority ranking matrix.

[0122] S6.3, according to the multi-channel differentiated test scheduling matrix, allocating test priority and resource ratio, generating a test task table and a channel usage level list.

[0123] Specifically, according to the multi-channel differentiated test scheduling matrix, the channel priority information corresponding to each test period is read in turn, and the test resource allocation ratio is allocated according to the priority in combination with the total amount of available test resources, so as to determine the resource proportion and test order of each channel in the corresponding test period; the test priority and resource allocation ratio data of each channel in each test period are summarized to generate a test task table containing test period, channel number, priority level and resource allocation ratio; at the same time, according to the resource allocation ratio and priority, a channel usage level list is formulated, the usage frequency and resource consumption level of each channel are classified and labeled to form a unified format channel usage level list.

[0124] The embodiment also provides a multi-channel clock buffer test system, comprising: a data acquisition module, configured to acquire an original waveform data set and synchronize a unified reference baseline to generate a clock output data stream; a feature extraction module, configured to perform frequency domain transformation on the clock output data stream, extract a main frequency drift, a harmonic intensity change and a phase fluctuation, and generate a spectrum disturbance feature vector sequence; a state updating module, configured to use a Kalman filter to recursively estimate the spectrum disturbance feature vector sequence, correct a state distribution by using a Bayesian probability updating method, and generate a health score drift trend vector; an anomaly identification module, configured to perform multi-scale comparison on the health score drift trend vector, and identify a channel entering a sub-health area by using an abnormal drift acceleration factor to generate an abnormal channel feature data set; a failure prediction module, configured to input the abnormal channel feature data set into a time series prediction network to predict a future health score trend and a potential failure time, and generate a failure warning result; and a test scheduling module, configured to perform weight fusion on the failure warning result and the health score drift trend vector, construct a multi-channel differentiated test scheduling matrix, and generate a test task table and a channel usage level list.

[0125] The embodiment also provides a computer device suitable for the case of the multi-channel clock buffer test method, comprising: a memory and a processor; the memory is configured to store computer executable instructions, and the processor is configured to execute the computer executable instructions to implement the multi-channel clock buffer test method proposed in the above embodiment.

[0126] The computer device can be a terminal, and the computer device includes a processor, a memory, a communication interface, a display screen and an input device connected by a system bus. The processor of the computer device is configured to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and a computer program. The internal memory provides an environment for running the operating system and the computer program in the non-volatile storage medium. The communication interface of the computer device is configured to perform wired or wireless communication with an external terminal. The wireless communication can be achieved by WIFI, an operator network, NFC (Near Field Communication) or other technologies. The display screen of the computer device can be a liquid crystal display screen or an electronic ink display screen. The input device of the computer device can be a touch layer overlaid on the display screen, or a key, trackball or touchpad arranged on the shell of the computer device, or an external keyboard, touchpad or mouse, etc.

[0127] The embodiment also provides a storage medium having a computer program stored thereon, the program being executed by a processor to implement the method for testing a multi-path clock buffer according to the above embodiment; and the storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as a static random access memory (SRAM), an electrically erasable programmable read-only memory (EEPROM), an erasable programmable read-only memory (EPROM), a programmable read-only memory (PROM), a read-only memory (ROM), a magnetic storage, a flash memory, a magnetic disk or an optical disk.

[0128] To sum up, the application can master the complex correlation between historical failure patterns and features by training a time series prediction network, effectively predict potential health score changes and failure times in the future, and realize early warning; not only can the application actively identify impending failures and reduce the occurrence rate of sudden failures, but also can provide dynamic optimization basis for a test scheduling module, reasonably allocate test resources, avoid unnecessary repeated detection, and significantly improve overall reliability and resource utilization.

[0129] It should be noted that the above examples are only used to illustrate the technical solutions of the present application but not limit the present application. Although the present application is described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present application can be modified or equivalently replaced, without departing from the spirit and scope of the technical solutions of the present application, which should be covered in the scope of the claims of the present application.

Claims

1. A method of testing a multipath clock buffer, the method comprising: The application relates to a health score drift trend vector generation method based on Kalman filter and Bayesian probability updating method. The application relates to a health score drift trend vector generation method based on Kalman filter and Bayesian probability updating method. The application relates to a health score drift trend vector generation method based on Kalman filter and Bayesian probability updating method. The application relates to a health score drift trend vector generation method based on Kalman filter and Bayesian probability updating method. The application relates to a health score drift trend vector generation method based on Kalman filter and Bayesian probability updating method. The application relates to a health score drift trend vector generation method based on Kalman filter and Bayesian probability updating method. The application relates to a health score drift trend vector generation method based on Kalman filter and Bayesian probability updating method.

2. The method of claim 1, wherein: The application relates to a health score drift trend vector generation method based on Kalman filter and Bayesian probability updating method. The application relates to a health score drift trend vector generation method based on Kalman filter and Bayesian probability updating method. The application relates to a health score drift trend vector generation method based on Kalman filter and Bayesian probability updating method. The application relates to a health score drift trend vector generation method based on Kalman filter and Bayesian probability updating method.

3. The method of claim 1, wherein: The application relates to a health score drift trend vector generation method based on Kalman filter and Bayesian probability updating method. The application relates to a health score drift trend vector generation method based on Kalman filter and Bayesian probability updating method. The application relates to a health score drift trend vector generation method based on Kalman filter and Bayesian probability updating method. The application relates to a health score drift trend vector generation method based on Kalman filter and Bayesian probability updating method. The application relates to a health score drift trend vector generation method based on Kalman filter and Bayesian probability updating method. The application relates to a health score drift trend vector generation method based on Kalman filter and Bayesian probability updating method.

4. The method of claim 1, wherein: The application relates to a health score drift trend vector generation method based on Kalman filter and Bayesian probability updating method. The application relates to a health score drift trend vector generation method based on Kalman filter and Bayesian probability updating method. The application relates to a health score drift trend vector generation method based on Kalman filter and Bayesian probability updating method. The application relates to a health score drift trend vector generation method based on Kalman filter and Bayesian probability updating method.

5. The method of claim 1, wherein: The application relates to a health score drift trend vector generation method based on Kalman filter and Bayesian probability updating method. The application relates to a health score drift trend vector generation method based on Kalman filter and Bayesian probability updating method. The application relates to a health score drift trend vector generation method based on Kalman filter and Bayesian probability updating method. The application relates to a health score drift trend vector generation method based on Kalman filter and Bayesian probability updating method. The application relates to a health score drift trend vector generation method based on Kalman filter and Bayesian probability updating method. The application relates to a health score drift trend vector generation method based on Kalman filter and Bayesian probability updating method. The application relates to a health score drift trend vector generation method based on Kalman filter and Bayesian probability updating method. The application relates to a health score drift trend vector generation method based on Kalman filter and Bayesian probability updating method. The application relates to a health score drift trend vector generation method based on Kalman filter and Bayesian probability updating method. The application relates to a health score drift trend vector generation method based on Kalman filter and Bayesian probability updating method. The application relates to a health score drift trend vector generation method based on Kalman filter and Bayesian probability updating method. The application relates to a health score drift trend vector generation method based on Kalman filter and Bayesian probability updating method. The application relates to a health score drift trend vector generation method based on Kalman filter and Bayesian probability updating method. The application relates to a health score drift trend vector generation method based on Kalman filter and Bayesian probability updating method. The application relates to a health score drift trend vector generation method based on Kalman filter and Bayesian probability updating method. The application relates to a health score drift trend vector generation method based on Kalman filter and Bayesian probability updating method. The application relates to a health score drift trend vector generation method based on Kalman filter and Bayesian probability updating method. The application relates to a health score drift trend vector generation method based on Kalman filter and Bayesian probability updating method. The application relates to a health score drift trend vector generation method based on Kalman filter and Bayesian probability updating method. The application relates to a health score drift trend vector generation method based on Kalman filter and Bayesian probability updating method. The application relates to a health score drift trend vector generation method based on Kalman filter and Bayesian probability updating method. The application relates to a health score drift trend vector generation method based on Kalman filter and Bayesian probability updating method. The application relates to a health score drift trend vector generation method based on Kalman filter and Bayesian probability updating method. The application relates to a health score drift trend vector generation method based on Kalman filter and Bayesian probability updating method. The application relates to a health score drift trend vector generation method based on Kalman filter and Bayesian probability updating method. The application relates to a health score drift trend vector generation method based on Kalman filter and Bayesian probability updating method. The application relates to a health score drift trend vector generation method based on Kalman filter and Bayesian probability updating method. The application relates to a health score drift trend vector generation method based on Kalman filter and Bayesian probability updating method. The application relates to a health score drift trend vector generation method based on Kalman filter and Bayesian probability updating method. The application relates to a health score drift trend vector generation method based on Kalman filter and Bayesian probability updating method. The application relates to a health score drift trend vector generation method based on Kalman filter and Bayesian probability updating method. The application relates to a health score drift trend vector generation method based on Kalman filter and Bayesian probability updating method. The application relates to a health score drift trend vector generation method based on Kalman filter and Bayesian probability updating method. The application relates to a health score drift trend vector generation method based on Kalman filter and Bayesian probability updating method. The application relates to a health score drift trend vector generation method based on Kalman filter and Bayesian probability updating method. The application relates to a health score drift trend vector generation method based on Kalman filter and Bayesian probability updating method. The application relates to a health score drift trend vector generation method based on Kalman filter and Bayesian probability updating method. The application relates to a health score drift trend vector generation method based on Kalman filter and Bayesian probability updating method. The application relates to a health score drift trend vector generation method based on Kalman filter and Bayesian probability updating method. The application relates to a health score drift trend vector generation method based on Kalman filter and Bayesian probability updating method. The application relates to a health score drift trend vector generation method based on Kalman filter and Bayesian probability updating method. The application relates to a health score drift trend vector generation method based on Kalman filter and Bayesian probability updating method. The application relates to a health score drift trend vector generation method based on Kalman filter and Bayesian probability updating method. The application relates to a health score drift trend vector generation method based on Kalman filter and Bayesian probability updating method. The application relates to a health score drift trend vector generation method based on Kalman filter and Bayesian probability updating method. The application relates to a health score drift trend vector generation method based on Kalman filter and Bayesian probability updating method. The application relates to a health score drift trend vector generation method based on Kalman filter and Bayesian probability updating method. The application relates to a health score drift trend vector generation method based on Kalman filter and Bayesian probability updating method. The application relates to a health score drift trend vector generation method based on Kalman filter and Bayesian probability updating method. The application relates to a health score drift trend vector generation method based on Kalman filter and Bayesian probability updating method. The application relates to a health score drift trend vector generation method based on Kalman filter and Bayesian probability updating method. The application relates to a health score drift trend vector generation method based on Kalman filter and Bayesian probability updating method. The application relates to a health score drift trend vector generation method based on Kalman filter and Bayesian probability updating method. The application relates to a health score drift trend vector generation method based on Kalman filter and Bayesian probability updating method. The application relates to a health score drift trend vector generation method based on Kalman filter and Bayesian probability updating method. The application relates to a health score drift trend vector generation method based on Kalman filter and Bayesian probability updating method. The application relates to a health score drift trend vector generation method based on Kalman filter and Bayesian probability updating method. The application relates to a health score drift trend vector generation method based on Kalman filter and Bayesian probability updating method. The application relates to a health score drift trend vector generation method based on Kalman filter and Bayesian probability updating method. The application relates to a health score drift trend vector generation method based on Kalman filter and Bayesian probability updating method. The application relates to a health score drift trend vector generation method based on Kalman filter and Bayesian probability updating method. The application relates to a health score drift trend vector generation method based on Kalman filter and Bayesian probability updating method. The application relates to a health score drift trend vector generation method based on Kalman filter and Bayesian probability updating method. The application relates to a health score drift trend vector generation method based on Kalman filter and Bayesian probability updating method. The application relates to a health score drift trend vector generation method based on Kalman filter and Bayesian probability updating method. The application relates to a health score drift trend vector generation method based on Kalman filter and Bayesian probability updating method. The application relates to a health score drift trend vector generation method based on Kalman filter and Bayesian probability updating method. The application relates to a health score drift trend vector generation method based on Kalman filter and Bayesian probability updating method. The 6. The method of claim 1, wherein: The failure warning result is generated, and the specific steps are as follows, The historical abnormal channel feature data set and the corresponding failure time label are taken as training samples, and the time series prediction network is optimized to minimize the health score prediction error, thereby generating a trained time series prediction network; The abnormal channel feature data set is encoded in time sequence as a prediction input sequence to generate a time series modeling sample; The time series modeling sample is input into the trained time series prediction network to predict the health score change in the future time period, thereby generating a health score prediction result; According to the health score prediction result, the score decline rate is calculated, and the failure time point is calculated combined with the drift trend, thereby generating a failure warning result.

7. The method of claim 1, wherein: The test task table and the channel usage level list are generated, and the specific steps are as follows, The failure warning result and the health score drift trend vector of the corresponding channel are weighted and fused to generate a channel comprehensive health evaluation index sequence; According to the channel comprehensive health evaluation index sequence, a priority ranking matrix of each channel at different test periods is constructed to generate a multi-channel differentiated test scheduling matrix; According to the multi-channel differentiated test scheduling matrix, test priorities and resource ratios are allocated to generate a test task table and a channel usage level list.

8. A multi-path clock buffer test system based on the multi-path clock buffer test method according to any one of claims 1 to 7, characterized in that: It includes, The data acquisition module is used for acquiring the original waveform data set and synchronously unifying the reference baseline to generate a clock output data stream; The feature extraction module is used for performing frequency domain transformation on the clock output data stream to extract the main frequency drift, harmonic intensity change and phase fluctuation, and generate a spectrum disturbance feature vector sequence; The state updating module is used for using a Kalman filter to recursively estimate the spectrum disturbance feature vector sequence, and combining a Bayesian probability updating method to correct the state distribution, thereby generating a health score drift trend vector; The abnormality identification module is used for comparing the health score drift trend vector in multiple scales, and using an abnormal drift acceleration factor to identify the channels entering the sub-health area, thereby generating an abnormal channel feature data set; The failure prediction module is used for inputting the abnormal channel feature data set into the time series prediction network to predict the future health score trend and the potential failure time, thereby generating a failure warning result; The test scheduling module is used for weighting and fusing the failure warning result and the health score drift trend vector to construct a multi-channel differentiated test scheduling matrix, thereby generating a test task table and a channel usage level list. 9.A computer device, comprising a memory and a processor, wherein the memory stores a computer program, and the computer device is characterized in that: The processor executes the computer program to realize the steps of the multi-channel clock buffer test method of any one of claims 1-7.

10. A computer readable storage medium having stored thereon a computer program, characterized in that: The computer program is executed by the processor to realize the steps of the multi-channel clock buffer test method of any one of claims 1-7.

Images