Measurement method and test measurement system

Abstract

To improve measurement speed.SOLUTION: A test and measurement device 20 receives a signal from a device 10 under test through a probe 32. One or more processors 38 generate a waveform from the signal, apply an equalizer to the waveform, receive an input identifying one or more measurements to be made on the waveform from a user interface device 44, select the number of unit intervals (UIs) for a known data pattern, scan the waveform for the known data patterns having a length of the number of UIs, identify the known data patterns as short pattern waveforms, apply a machine learning system 46 to the short pattern waveforms to obtain a value for the one or more measurements, and supply the values of the one or more measurements for the waveform.SELECTED DRAWING: Figure 8

Description

[Technical Field] 【0001】 This disclosed technology relates to a test measurement system and method, and more specifically, to the application of machine learning to the measurement of signals from a device under test. [Background technology] 【0002】 Many electronic devices and systems use high-speed signals for communication and data transfer, specifically signals transmitted between transmitters and receivers according to high-speed serial data protocols such as PCIe (Peripheral Component Interconnect Express) and Ethernet®. Traditionally, test and measurement equipment such as oscilloscopes have been used to acquire these high-speed signals, generate eye diagrams, and measure the characteristics of the signals. 【0003】 As signal speeds increase, equalizers are widely used in transmitters and receivers to improve system performance. For example, PCIe Gen5 receivers have a 3-tap decision feedback equalizer (DFE) in addition to a continuous time linear equalizer (CTLE). See, for example, PCI-SIG, "PCI Express Base Specification 5.0, Version 10" 2019 (available at https: / / pcisig.com / specifications / ). The IEEE 100G / 400G Ethernet standard defines measurements using a 5-tap feed-forward equalizer (FFE). For example, see "IEEE P802.3bs-2017" (available at http: / / standards.ieee.org / findstds / standard / 802.3bs-2017.html) and "IEEE P802.3cd-2018" (available at http: / / standards.ieee.org / develop / project / 802.3cd.html). 【0004】 If the receiver has an equalizer, some of the measurements are performed on an equalized signal. For example, in PCIe 5th generation, eye height and eye width measurements are defined based on the eye diagram of the equalized waveform. [Prior art documents] [Patent Documents] 【0005】 [Patent Document 1] Japanese Patent Publication No. 2016-25662 [Patent Document 2] Japanese Patent Publication No. 2016-213834 [Patent Document 3] Japanese Patent Publication No. 2013-257329 [Patent Document 4] International Publication No. 2021 / 092156 [Non-patent literature] 【0006】 [Non-Patent Document 1] "PCI Express Base Specification 5.0, Version 10", PCI-SIG, published July 23, 2019, [online], [searched May 19, 2022], Internet<https: / / pcisig.com / specifications / > [Overview of the Initiative] [Problems that the invention aims to solve] 【0007】 In some methods, machine learning systems may use an eye diagram before equalization as input. The machine learning system can then provide the desired measurements. However, the eye diagram before equalization does not contain time-series information, and the post-equalizer waveform may differ significantly from the pre-equalizer waveform, which makes the process inaccurate. 【0008】 Embodiments of the present invention address these and other problems. [Means for solving the problem] 【0009】 Embodiments of the present invention include systems and methods for applying machine learning techniques to perform signal measurements on an input waveform. Embodiments of the present invention generally utilize a database of short patterns created from waveforms. This allows the process to perform analysis faster than if all or some pattern waveforms were used, by employing machine learning. Instead, embodiments use a machine learning system trained on short patterns of a large number of different symbols stored in one or more databases. The system can then scan to find these patterns and generate measurements associated with the short patterns. Embodiments of the present invention offer improved measurement accuracy compared to techniques using eye diagrams. [Brief explanation of the drawing] 【0010】 [Figure 1] Figure 1 shows one embodiment of an optical transceiver test system. [Figure 2] Figure 2 shows an explanatory diagram of TDECQ measurement. [Figure 3] Figure 3 shows two examples of eye diagrams before and after applying an equalizer. [Figure 4] Figure 4 shows a graphical representation of taps in a 5-tap feed-forward equalizer. [Figure 5] Figure 5 shows a partial display of a pattern waveform containing time-series information. [Figure 6] Figure 6 shows examples of waveform databases with various pattern lengths. [Figure 7] Figure 7 shows an example of a waveform database for various symbol sequences. [Figure 8] Figure 8 shows one embodiment of the test and measurement apparatus. [Figure 9] Figure 9 shows one embodiment of a machine learning system that includes a short pattern waveform database. [Figure 10] Figure 10 shows an example of a short pattern waveform tensor image. [Figure 11]FIG. 11 shows an example of a short-pattern waveform tensor image. [Figure 12] FIG. 12 shows an example of a short-pattern waveform database of a pattern with a length of 1 symbol. 【DETAILED DESCRIPTION OF THE INVENTION】 【0011】 As described above, as the signal speed increases, in many systems, an equalizer is adopted to improve system performance. As described above, the sixth generation of PCIe (Peripheral Component Interface Express) uses a 3-tap decision feedback equalizer (DFE) in addition to a continuous-time linear equalizer (CTLE). In another example, the 100G / 400G Institute of Electrical and Electronics Engineers (IEEE) standard 802.3 specifies TDECQ (transmitter and dispersion eye closure quaternary) measurement as an important pass / fail criterion for 26 gigabaud (GBaud) and 53 GBaud PAM4 optical signaling. This gives an example showing the problem of using a waveform without time-series information as an input to a machine learning system. 【0012】 The TDECQ measurement includes a 5-tap FFE. FIG. 1 shows a test block diagram showing the acquisition of an optical signal from a transmitter (Tx) or a transceiver under test 10. The optical signal may interact with an optical system 12 such as a polarization rotator or a variable reflector. The signal passes through a test fiber 14 and reaches an optoelectronic (O / E) converter 16, which converts the optical signal into an electrical signal. An oscilloscope 20 may have a clock recovery unit (CRU) 18, and then samples the obtained electrical signal and digitizes the signal. The digitized samples are stored as waveforms. 【0013】 The reference equalizer and analysis module 22 in a conventional oscilloscope may then perform TDECQ measurement and analysis. Figure 2 shows an example of a diagram used when performing TDECQ measurement. In this example, the waveform is obtained from a 5-tap feed-forward equalizer (FFE) by optimizing the interval of 1 unit intervals (UI) to minimize the TDECQ value. The intervals between "0" and "1" represent the UI intervals. 【0014】 The TDECQ value is calculated using the following formula (Formula 1). 【number】 Here, OMA outer This relates to the power of the optical signal. Q r σ is a constant value related to the symbol error ratio. G 2 This is the standard deviation of weighted Gaussian noise that may be added to the eye diagram shown in Figure 2, and the symbol error ratio may be even larger in the two vertical slicers of 0.45 and 0.55, with a difference of 0.1 UI being 4.8e-4. S This section represents oscilloscope noise, which is recorded when no signal is supplied to the O / E module. 【0015】 A single TDECQ measurement for the compliance pattern SSPRQ (short stress pattern random quaternary) takes several seconds to complete using conventional methods. Patent Document 4 discloses a machine learning technique aimed at reducing the time required to acquire measurements of an optical transceiver, including TDECQ. One of the disclosed machine learning methods receives an eye diagram image representation of the waveform as input for training a neural network, and then for testing the optical transceiver. 【0016】 Figure 3 shows examples of eye diagrams before (left) and after (right) applying FFE to a waveform. The eye diagram after FFE application on the right has a larger eye opening. Using the eye diagram before FFE as input to a neural network for machine learning will not provide the information of the eye diagram after FFE. Five FFE taps are applied to five samples in the vicinity of the current sample (unit interval). The eye diagram before FFE does not contain time sequence information because all samples are folded into one or two UIs (wrapped). Figure 4 shows the FFE taps used in this example to create the eye diagram on the right in Figure 3. 【0017】 In contrast, as shown in Figure 5, in actual pattern waveforms, every sample has a time associated with it. This time sequence information can be used in FFE processing. Another machine learning technique using eye diagrams uses the equalized actual pattern waveform as input for training a neural network, and then as input for testing. 【0018】 However, the number of samples in a pattern waveform can be too large to be practical for training. For example, the SSPRQ pattern has 65,535 symbols. If there are multiple samples for each UI, the sampled waveform becomes very large. Using the actual sampled waveform would make machine learning training time-consuming. One option is to use a partial pattern waveform, but this approach may not include important information in the waveform, which can increase measurement errors. 【0019】 As described above, embodiments of the present invention perform signal measurements using a short pattern waveform database with a machine learning module (e.g., a neural network). Furthermore, for measurements requiring an equalizer, to obtain more accurate results, the input data to the neural network must include time-series information, as the equalizer operates on time-series samples. Conventional eye diagrams lost time-series information between symbols. Embodiments of the present invention use short pattern waveforms that include time-series information, providing a solution to the problems of waveform size, accuracy, and time series. 【0020】 This process constructs a short pattern waveform database based on short patterns appearing in the waveform. As used in this application, "short" refers to a portion of the waveform having a length equal to a predetermined number of UI. The system scans the data pattern, identifies and extracts short pattern waveforms, and places waveform samples of the extracted short pattern waveforms into the corresponding short pattern waveform database. This scanning process may be repeated or operated in parallel to construct multiple short pattern waveform databases for each target short pattern waveform. The database selection may be determined by the signaling format, such as pulse amplitude modulation 4-level (PAM4) signaling or non-return-to-zero (NRZ) signaling, and the signal levels of the pattern. For example, PAM4 has four levels corresponding to symbols 0, 1, 2, and 3, while NRZ has two levels, symbol 1 or 0. In the following description, the variable S indicates the number of signal levels derived from the signaling format. The data pattern is often known or detectable. 【0021】 The number N of UIs defines the length of the short pattern. The selection of this length may take into account the influence of preceding symbols on the current symbol. Figure 6 shows a waveform database of various numbers of preceding symbols leading to the current 3 symbols for a PAM4 signal. For example, consider 1 to 4 zeros as preceding symbols for 3 symbols representing one signaling level in the PAM4 signal. The more symbols considered, the cleaner the waveform database becomes for the current symbol (i.e., the less influence of blending from preceding symbols). As used in this application, the term “waveform database” means all sets of pattern waveform portions within the entire pattern waveform that have the same symbol pattern over a given short pattern length N. As will be described in more detail below, the system may use a subset of the database, the subset used being determined by measurements made on the pattern and the captured waveform. 【0022】 The waveform database in the upper left contains all short pattern waveforms, each spanning two UIs, with a pattern length of 2, and their short pattern being 03. The upper right shows a waveform database of short pattern waveforms spanning three UIs, with two 0 symbols before three symbols, and their short pattern being 003. The lower left shows a waveform database of four symbols with a short pattern of 0003, and the lower right shows a waveform database of five symbols with a short pattern of 00003. 【0023】 The receiver's equalizer configuration compensates for channel interference, such as channel loss. Channel loss is reflected in intersymbol interference. For measurements requiring equalization, the length of the short pattern can be selected to match the reach of the equalizer. This provides important information about the time series, which allows the data input to the machine learning block to form an accurate model, thus yielding accurate measurement results. 【0024】 For example, in TDECQ measurement, 5 tap FFEs are required, so the process sets the short pattern length to 5. As shown in Figure 7, the short pattern waveform database created from the SSPRQ data pattern has time-series information for each short pattern, and the machine learning system can capture important information from the data to obtain accurate measurement results. The top row of images shows the short pattern waveform database with symbol sequences 00030, 01030, and 02030 from left to right. The middle row shows the short pattern database with symbol sequences 03030, 00300, and 10300 from left to right. The bottom row shows symbol sequences 20300 and 30300 from left to right. 【0025】 To utilize these databases, machine learning systems first need to receive waveforms as input in a format that allows for fast and accurate training and runtime. Referring again to Figure 1, a test measurement device such as an oscilloscope 20 receives a signal from a transceiver and generates a waveform. Although the device under test (DUT) in this description consists of an optical transceiver, it should be noted that the systems and methods used in this application are applicable to any type of optical or electrical DUT. 【0026】 Figure 8 shows one embodiment of a test measurement device 20 that can be used with a machine learning system 46 to provide performance measurements on a DUT (e.g., 10). The test measurement device is generally connected to the DUT 10 via a probe 32. As described with respect to Figure 1, the input path may have a photoelectric converter that converts the input optical signal into an electrical signal. The acquisition circuit 36 ​​of the device 20 may have an analog-to-digital converter (ADC) that digitizes the input signal and clock recovery and trigger hardware that provides timing. A processor 38 may control the acquisition hardware and the rendering of the signal into a waveform. A display unit 42 displays the obtained waveform to the user. A user interface device 44 may include an optional touchscreen function on the display unit, allowing the user to interactively operate the device and select from a preset menu. This selection may include the type of measurement required for the waveform, the length of the short pattern waveform, etc. This length may be derived from a preset variable or a preset variable in the system, etc. 【0027】 The memory 40 allows the processor to store waveform data and process the waveform data, and the memory 40 may also store executable code (programs). The entire system, including the test measurement device, has one or more processors, and one or more processors are configured to execute code (programs) that cause one or more processors to perform the various tasks described herein. The one or more processors may include one or more processors on the test measurement device and one or more processors in the machine learning system 46. The machine learning system may have one or more independent computer devices that receive data from the test measurement device. The independent database structure 48 may store all waveform databases, or it may include the machine learning system and part of its computing devices. 【0028】 Upon receiving a signal from the DUT, the test measurement device generates a waveform of the signal and applies one or more equalizers to the waveform. This means that the equalizers act on the samples that make up the waveform. One or more processors in the system may perform these tasks. The system receives an input specifying the length N of a short pattern in units of UIs. As described above, the user may provide this input, or the system may determine it from predefined parameters, etc. Similarly, one or more required performance measurements, such as TDECQ or other measurements, will be identified. In some embodiments, the short pattern length may be automatically determined based on the selected measurement. For example, if the TDECQ measurement is selected, the system may automatically determine the short pattern length to be 5UI, corresponding to the reach of a 5-tap FFE equalizer specified for the TDECQ measurement. 【0029】 Next, the system scans the waveforms to find known patterns of their lengths and generates a set (group) of short pattern waveforms. In one embodiment, the system may convert the short pattern waveforms into tensors, but for the purposes of this description, these will still be considered as short pattern waveforms. Once the short patterns are identified, they are sent to a machine learning system, which then returns the desired measurements. This system operates much faster and provides measurement values ​​than conventional methods of calculating measurements. 【0030】 Figure 9 shows one embodiment of a machine learning structure using a short pattern waveform database. The short pattern waveform database may take the form of a machine learning-appropriate format, such as Tensor50, as input to the neural network for training and testing. Its output is the measurement result. The measurement result can be a scalar value or a vector and can be used as a label for machine learning. During training, both the short pattern waveform and the measurement result are used. During testing, only the short pattern waveform is used to obtain the measurement result. 【0031】 The training process includes a process of selecting the length of the short pattern. This can start from a small value (such as 3). The number L of possible short symbol sequences is determined by the signal level S and the short pattern length N in Equation 2. L = S N (2) 【0032】 For example, in the case of PAM4 signaling, for a short pattern of 3 symbols, 4 3 = 64 waveform databases may exist. In the case of NRZ signaling, for a short pattern of 3 symbols, 2 3 = 8 waveform databases may exist. As described above, this can lead to a very large database that covers all short patterns within a specific database. For example, in PAM4 signaling, when N = 5, 4 5 i.e., 1024 short patterns are possible. Training a machine learning system requires a large number of examples for each of the 1024 possible short patterns and the measured values associated with each pattern. This would take too much time and too many resources to train a machine learning system. 【0033】 In one embodiment, the system uses a subset of possible short patterns from a database and measurements associated with these short patterns. For example, suppose the desired result is to obtain a machine learning system in the form of a deep learning network that predicts tuning parameters affecting four levels of a PAM4 signal. Suppose the system uses four short patterns with consecutive UIs of the same level. In one embodiment, all four of these sequences (consecutive ones) are placed in a single tensor image, which serves as the input to the deep learning system for both runtime and training. Figure 10 shows an example of a tensor image of four sequences for four levels. 【0034】 For example, one tuning parameter in the system can adjust the signal gain, so that all four levels will be closer to each other at lower gain settings and further apart at higher gain settings. An offset control in the transmitter will cause all four symbols to move vertically or downwards in the image, while the distance between them remains the same. A third transmitter parameter may vary both the gain and the offset. By using this image, which represents a subset of a short pattern database containing all possible short pattern waveforms, a deep learning network can easily see these effects of all three parameters and make predictions about their values. 【0035】 In another example, a machine learning system can also predict FFE taps. Using a short pattern waveform showing pulses, as shown in Figure 11, machine learning works better because FFE taps influence pulse shape, allowing the deep learning network to associate pulse shape with a set of FFE taps. The image in Figure 11 shows a single tensor image containing three short patterns laid out horizontally from edge to edge. Each sequence, from left to right, shows different pulse heights: for the first pulse, level 0 to level 1 of the PAM4 signal; for the second pulse, level 0 to level 2; and for the third pulse, level 0 to level 3. A deep learning network can look at this image and predict what multiple FFE tap values ​​are. Of course, other applications exist, and these are simply examples of how machine learning systems can use images. 【0036】 If the desired results are not obtained from machine learning training with the current short pattern length setting, the process may increase the short pattern length and retry the training. The process may also select a different subset of the short training patterns and use that for training. Since the pattern used changes depending on the short pattern length, this also includes selecting a different subset of the short training patterns. The short pattern waveform used for training is sometimes called the short training pattern waveform or short training pattern. 【0037】 As mentioned above, the length of a short pattern is determined by the reach (range of influence) of the equalizer, but it should not "excessively" exceed the reach of the equalizer. For example, if the equalizer is a 5-tap FFE, the pattern length should not excessively exceed 5UI. For example, the pattern length can be selected as 5, 6, or 7. If the equalizer is a 3-tap DFE, since DFE refers only to the three preceding symbols, the pattern length should not excessively exceed 4. The determination of what constitutes "excessive" may depend on the nature of the input to the machine learning system. For example, in one embodiment, if the input to the neural network is image data as described above, the limitation of the image size may also be a factor in determining the number of UIs (i.e., length) of the pattern. In one embodiment, the image size is limited to 224 x 224 pixels, and a machine learning system designed to accept input images of this size is employed. 【0038】 Once the process finds an appropriate short pattern length that yields the desired machine learning results, the process may also check the weights (i.e., coefficients) within the input layer 52. This may identify connections with small associated weights, which may include comparisons with thresholds. The system then removes its corresponding short pattern waveform database from the input to reduce the input data size, and then checks whether the training results still meet the requirements. Some machine learning tools have a dimension reduction feature that automatically reduces the input data size. This process can be considered dimensionality reduction. 【0039】 The explanation so far has mainly focused on waveform databases associated with PAM4 signaling. By setting N to a different value, the short-pattern waveform database method can also cover other machine learning methods, such as the traditional eye diagram method and the full-data pattern waveform method. 【0040】 When N is set to 1, there are four short pattern waveform databases for the PAM4 signal, as shown in Figure 12: 0 in the upper left, 1 in the upper right, 2 in the lower left, and 3 in the lower right. Overlapping these four waveform databases yields a conventional eye diagram. When length N is set to 1, the short pattern waveform database method for machine learning is expected to produce results similar to those obtained when machine learning uses a conventional eye diagram. 【0041】 When N is set to the length of the full pattern waveform, there is only one short pattern waveform database containing the data. This short pattern waveform database has the same symbol sequence as the full pattern of the signal. The short pattern waveform database method allows for flexible setting of N, enabling adjustment of the accuracy and speed of measurements using machine learning. 【0042】 This technique can also be used to improve other measurements that require limited time-series information, such as intersymbol interference jitter measurements. 【0043】 Embodiments of the disclosed technology can operate on a specially programmed general-purpose computer, including specially created hardware, firmware, digital signal processors, or processors that operate according to programmed instructions. The terms “controller” or “processor” in this application mean microprocessors, microcomputers, ASICs, and dedicated hardware controllers, etc. Embodiments of the disclosed technology can be implemented by one or more computers (including monitoring modules) or other devices, using computer-readable data such as program modules and computer-executable instructions. Generally, program modules include routines, programs, objects, components, data structures, etc., which, when executed by a processor in a computer or other device, perform specific tasks or implement specific abstract data formats. Computer-executable instructions may be stored on computer-readable storage media such as hard disks, optical disks, removable storage media, solid-state memory, RAM, etc. As will be understood by those skilled in the art, the functions of the program modules may be combined or distributed as needed in various embodiments. Furthermore, these functions can be embodied in whole or in part in firmware or hardware equivalents such as integrated circuits or field-programmable gate arrays (FPGAs). One or more aspects of the disclosed technology can be more effectively implemented using specific data structures, such data structures are considered to be within the scope of computer-executable instructions and computer-usable data described herein. 【0044】 The disclosed embodiments may, in some cases, be implemented in hardware, firmware, software, or any combination thereof. The disclosed embodiments may also be implemented as instructions carried or stored in one or more computer-readable media that can be read and executed by one or more processors. Such instructions may be referred to as computer program products. The computer-readable media described herein means any medium accessible by a computing device. For example, but not limited to, computer-readable media may include computer storage media and communication media. 【0045】 Computer storage media means any medium that can be used to store computer-readable information. Examples of computer storage media include, but are not limited to, random-access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory and other memory technologies, compact disc read-only memory (CD-ROM), DVD (Digital Video Disc) and other optical disc storage devices, magnetic cassettes, magnetic tapes, magnetic disk storage devices and other magnetic storage devices, and any other volatile or non-volatile removable or non-removable media implemented by any technology. Computer storage media exclude signals themselves and temporary forms of signal transmission. 【0046】 A communication medium means any medium that can be used for the communication of computer-readable information. Examples of communication mediums, though not limited to them, include coaxial cables, fiber optic cables, air, or any other medium suitable for the communication of electrical, optical, radio frequency (RF), infrared, sound, or other forms of signals. 【0047】 In addition, the description of this application refers to certain features. It should be understood that the disclosures herein include all possible combinations of these particular features. Where a particular feature is disclosed in relation to a particular aspect or example, that feature may, to the extent possible, also be used in relation to other aspects and examples. 【0048】 Furthermore, when this application refers to a method having two or more defined steps or processes, these defined steps or processes may be performed in any order or simultaneously, as long as the circumstances do not rule out such possibilities. Examples 【0049】 The following examples are provided that are useful for understanding the technology disclosed herein. These embodiments may include one or more of the examples described below, or any combination thereof. 【0050】 Example 1 is a method comprising: a process of receiving a signal from a device under test; a process of generating a waveform from the signal; a process of applying an equalizer to the waveform; a process of receiving an input that identifies one or more measurements to be performed on the waveform; a process of selecting a number of unit intervals (UIs); a process of scanning (investigating) the waveform to identify a short pattern waveform having a length equal to the number of UIs; a process of applying a machine learning system to the short pattern waveform to obtain the values ​​of one or more measurements; and a process of supplying the values ​​of one or more measurements of the waveform from the machine learning system. 【0051】 Example 2 is the method of Example 1, wherein the process of applying the machine learning system to the short pattern waveform includes the process of applying the machine learning system to a tensor as the short pattern waveform. 【0052】 Example 3 is a method of either Example 1 or 2, wherein the process of applying the machine learning system includes a process of analyzing the short pattern waveform using one or more short pattern databases. 【0053】 Example 4 is the method of Example 3, wherein the process of using one or more short pattern databases includes the process of using only a subset of the one or more short pattern databases. 【0054】 Example 5 is the method of Example 3, further comprising the process of using one or more short pattern databases, which in order to reduce the input data size, removes short pattern databases having coefficient values ​​below a certain threshold from the machine learning system. 【0055】 Example 6 is a method according to any of Examples 1 to 5, wherein the process of selecting the number of UIs includes the process of selecting the number of UIs based on the number of taps on the equalizer. 【0056】 Example 7 is a method of any of Examples 1 to 6, wherein the process of selecting the number of UIs includes the process of selecting the number of UIs based on one or more measurements performed on the waveform. 【0057】 Example 8 is a method of any of Examples 1 to 7, further comprising a process for training the machine learning system, wherein the training process includes: setting the length to be used for the short patterns; selecting a set of short training patterns from a waveform and selecting related measurements of the set of short training patterns as a dataset to be used by the machine learning system; testing the machine learning system to determine whether the results generated by the machine learning system satisfy the desired results; and, if the results do not satisfy the desired results, selecting a different set of short training patterns and repeating the testing process using the different set of short training patterns. 【0058】 Example 9 is the method of Example 8, wherein the process of selecting the different sets of the short training patterns includes the process of selecting different sets of short training patterns of the same length, or the process of selecting different sets of short training patterns of a longer length. 【0059】 Example 10 is the method of Example 9, wherein the short patterns are stored in a specific number of short pattern databases, and in this case, the number L in the short pattern sequence databases is L = S N According to the relationship, it is determined by the number of signal levels S and the pattern length N used in a particular form of signaling. 【0060】 Embodiment 11 is a test measurement system comprising a test measurement device configured to receive signals from a device under test, and one or more processors, the one or more processors being configured to execute a program (code) that causes the one or more processors to perform the following: generating a waveform from the signal, applying an equalizer to the waveform, receiving an input that identifies one or more measurements to be performed on the waveform, selecting a number of unit intervals (UIs) of a known data pattern, scanning (investigating) the waveform to find a known data pattern having a length of the number of UIs, identifying the known data pattern as a short pattern waveform, applying a machine learning system to the short pattern waveform to obtain the values ​​of the one or more measurements, and supplying the values ​​of the one or more measurements of the waveform. 【0061】 Example 12 is the test measurement system of Example 11, wherein the short pattern waveform is composed of tensors. 【0062】 Example 13 is a test measurement system according to either Example 11 or 12, wherein a program (code) that causes one or more processors to perform the process of applying the machine learning system to the short pattern waveform includes a program (code) that causes one or more processors to perform the process of using one or more short pattern waveform databases. 【0063】 Example 14 is the test measurement system of Example 13, wherein a program (code) that causes one or more processors to perform a process using one or more short pattern waveform databases further includes a program (code) that causes one or more processors to perform a process of removing short pattern waveform databases having coefficient values ​​below a certain threshold from the machine learning system in order to reduce the input data size. 【0064】 Example 15 is the test measurement system of Example 13, wherein a program (code) that causes one or more processors to perform a process using one or more short pattern waveform databases further includes a program (code) that causes one or more processors to perform a process using only a subset of the one or more short pattern waveform databases. 【0065】 Example 16 is a test measurement system according to any of Examples 11 to 15, wherein a program (code) that causes one or more processors to perform the process of selecting the number of UIs further includes a program (code) for selecting the number of UIs based on the number of taps of the equalizer applied to the waveform. 【0066】 Example 17 is a test measurement system according to any of Examples 11 to 16, wherein a program (code) that causes one or more processors to scan (investigate) the waveforms and identify the known data patterns as short pattern waveforms includes a program (code) for selecting short pattern waveforms and including time sequence information. 【0067】 Example 18 is a test measurement system of any one of Examples 11 to 17, wherein one or more processors are further configured to execute a program (code) for training the machine learning system, which includes: setting the length to be used for the short training pattern and setting a subset of the short training pattern having the set length; selecting an available subset of the short training pattern and associated measurements of the short training pattern, which are provided to the machine learning system as a dataset; testing the machine learning system to determine whether the results produced by the machine learning system satisfy the desired results; and, if the results do not satisfy the desired results, selecting another subset of the short training pattern and repeating the testing process. 【0068】 Example 19 is the test measurement system of Example 18, wherein a program (code) that causes one or more processors to perform the process of selecting different subsets of the short training pattern includes the process of selecting either different subsets of the short training pattern having the same length or different subsets of the short training pattern having a longer length. 【0069】 Example 20 is a test measurement system of any of Examples 11 to 19, wherein short patterns are stored in a specific number of short pattern databases, and the above number L in the short pattern sequence database is L = S N According to the relationship, it is determined by the number of signal levels S and the pattern length N used in a particular form of signaling. 【0070】 For the sake of explanation, specific embodiments of the disclosed technology have been illustrated and described, but it should be understood that various modifications are possible without deviating from the gist and scope of the present invention. Therefore, the disclosed technology should not be limited to anything other than the appended claims. [Explanation of symbols] 【0071】 10. Transmitter (Tx) or Transceiver under Test (DUT) 12 Optical system 14 Test Fibers 16. Photoelectric (O / E) Converter 18. Clock Recovery Unit (CRU) 20. Test and measurement equipment (oscilloscope) 22 Reference Equalizer and Analysis Modules 32 probes 36. Acquisition Circuits 38 processors 40 memory 42 Display section 44 User Interface Devices 46 Machine Learning Systems 48 Database Structure 50 Input Tensors 52 Input Layers 54 Hidden Layer 1 56 Hidden Layer 2 58 Output Layer

Claims

[Claim 1] The process of receiving signals from the device under test, The process of generating a waveform from the above signal, The process of applying an equalizer to the above waveform, A process that receives an input specifying one or more measurements to be performed on the above waveform, The process of selecting the number of unit intervals (UI), The process involves investigating the above waveforms and identifying short pattern waveforms with a length equal to the number of UIs, The process involves applying a machine learning system to the above short pattern waveform to obtain one or more of the above measurement values, A process of supplying one or more measurement values ​​of the above waveform from the above machine learning system, and A measurement method that includes [the following]. [Claim 2] The measurement method according to claim 1, wherein the process of applying the above-mentioned machine learning system to the above-mentioned short pattern waveform includes the process of applying the above-mentioned machine learning system to the tensor as the above-mentioned short pattern waveform. [Claim 3] The measurement method according to claim 1, wherein the process of applying the above-mentioned machine learning system includes a process of analyzing the short pattern waveform using one or more short pattern databases. [Claim 4] The measurement method according to claim 3, further comprising a process of removing from the machine learning system short pattern databases having coefficient values ​​below a certain threshold in order to reduce the input data size, in order to use one or more short pattern databases. [Claim 5] The measurement method according to claim 1, wherein the process of selecting the number of UIs includes a process of selecting the number of UIs based on the number of taps on the equalizer. [Claim 6] The above machine learning system is further comprising a process for training the machine learning system, wherein the training process is The process of setting the length to be used for the short pattern, The machine learning system uses the above-mentioned machine learning system as a dataset, which involves selecting a set of short training patterns from waveforms and the associated measurements of the set of short training patterns. A process to test the machine learning system in order to determine whether the results generated by the machine learning system satisfy the desired results, The measurement method according to claim 1, further comprising: selecting a different set of the short training patterns if the above results do not satisfy the desired results; and repeating the process of testing using the different set of the short training patterns. [Claim 7] The measurement method according to claim 6, wherein the process of selecting the different sets of the above short training patterns includes the process of selecting different sets of short training patterns of the same length, or the process of selecting different sets of short training patterns having a longer length. [Claim 8] The above short patterns are stored in a specific number of short pattern databases, and the above number L in the short pattern sequence database is L = S N The measurement method according to claim 1, which is determined by the number S and pattern length N of signal levels used in a particular form of signaling, according to the relationship. [Claim 9] A test measurement device configured to receive signals from the device under test, One or more processors and The system comprises, and one or more processors The process of generating a waveform from the above signal, The process of applying an equalizer to the above waveform, A process that receives an input specifying one or more measurements to be performed on the above waveform, The process of selecting the number of unit intervals (UIs) for known data patterns, The process involves investigating the waveform in order to find the known data pattern having the length of the above UI, The process involves identifying the above known data patterns as short pattern waveforms, The process involves applying a machine learning system to the above short pattern waveform to obtain one or more of the above measurement values, A process to supply one or more measured values ​​of the above waveform. A test and measurement system configured to execute a program that causes one or more of the above-mentioned processors to perform the above task. [Claim 10] The test measurement system according to claim 9, wherein a program that causes one or more processors to perform the process of applying the machine learning system to the short pattern waveform includes a program that causes one or more processors to perform the process of using one or more short pattern waveform databases. [Claim 11] The test measurement system according to claim 10, wherein a program that causes one or more processors to perform processing using one or more short pattern waveform databases further includes a program that causes one or more processors to perform processing to remove short pattern waveform databases having coefficient values ​​below a certain threshold from the machine learning system in order to reduce the input data size. [Claim 12] The test measurement system according to claim 9, wherein a program that causes one or more processors to perform the process of selecting the number of UIs further includes a program for selecting the number of UIs based on the number of taps of the equalizer applied to the waveform. [Claim 13] The test measurement system according to claim 9, wherein a program that causes one or more processors to investigate the above waveforms and identify the above known data patterns as short pattern waveforms includes a program for selecting short pattern waveforms and including time-series information. [Claim 14] The one or more processors described above are further configured to run a program for training the machine learning system described above, and the program is configured to run on the one or more processors described above. The process involves setting the length to be used for a short training pattern, and setting a subset of the short training pattern having the set length. A process of selecting a subset of the above short training patterns of waveforms provided as a dataset to a machine learning system, and the associated measurements of the above short training patterns. A process to test the machine learning system in order to determine whether the results generated by the machine learning system satisfy the desired results, If the above results do not satisfy the desired results, select another subset of the above short training pattern and repeat the process to be tested. A test and measurement system according to claim 9, which causes the following to be performed. [Claim 15] The test measurement system according to claim 14, wherein the program that causes one or more processors to perform the process of selecting different subsets of the short training patterns includes the process of selecting different subsets of the short training patterns having the same length, or different subsets of the short training patterns having a longer length. [Claim 16] The above short patterns are stored in a specific number of short pattern databases, and the number of short pattern sequence databases L is given by L = S N The test measurement system according to claim 9, which is determined by the number S and pattern length N of signal levels used in a particular form of signaling, according to the relationship.

Images