Test measurement system, method for tuning optical transmitter, and method for training machine learning system for tuning optical transmitter.

Abstract

To provide systems and methods for quickly tuning an optical transmitter.SOLUTION: In a test measurement device 10 connected to an optical transceiver 12 including a transmitter, a processor 18 is configured to: set operating parameters for the optical transceiver 12 to reference operating parameters; acquire a waveform from the optical transceiver 12; repeat setting of operating parameters and acquisition of a waveform, for each of a predetermined number of sets of reference operating parameters; build one or more tensors from the acquired waveforms; send the one or more tensors to a machine learning system or a neural network 22 to obtain a set of predicted operating parameters; set the operating parameters for the optical transceiver 12 to the predicted operating parameters; and test the optical transceiver 12 using the predicted operating parameters.SELECTED DRAWING: Figure 1

Description

Technical Field

[0001] The present disclosure relates to test and measurement systems, and more particularly, to systems and methods for tuning parameters of a device under test (for example, an optical transmitter).

Background Art

[0002] Manufacturers of optical transmitters tune and test their optical transmitters to confirm that the optical transmitters operate correctly.

Prior Art Documents

Patent Documents

[0003]

Patent Document 1

Summary of the Invention

Problems to be Solved by the Invention

[0004] In the worst case, each transmitter may take up to 2 hours. Tuning usually takes the form of sweeping tuning parameters, that is, the transmitter performs tuning at each level of each parameter. In the worst case, this may require 200 iterations, and in the best case, 3 to 5 iterations. Reducing the number of iterations for tuning the transceiver to speed up this process will reduce both manufacturing time and, consequently, manufacturing costs.

[0005] The disclosed embodiments of the apparatus and methods address the drawbacks in the prior art.

Means for Solving the Problems

[0006] Currently, no type of machine learning is used in optical transmitter tuning to reduce the time required for tuning and testing the transmitter. The time required for training machine learning increases the overall cost of the transmitter. Using a machine learning process can reduce the time required for tuning, testing, and determining whether a component is acceptable or unacceptable. Embodiments of the present invention use one or more neural networks to obtain tuning parameters for an optical transmitter, enabling a faster determination of whether the transmitter is functioning correctly than is currently possible.

[0007] Figure 1 shows an embodiment of a test measurement system or apparatus that uses a machine learning system to determine the operating parameters of a component. The output measurements may include TDECQ (transmitter dispersion eye closure penalty quaternary), extinction ratio, average optical power, optical modulation amplitude (OMA), and level separation mismatch ratio, which can be tuned by the user and used to measure operation. The tuning parameters referred to in this description include settings loaded into the transmitter to achieve the desired results. These parameters vary considerably depending on the characteristics of the optical transmitter and the way the optical transmitter operates. Parameters may include values ​​such as voltage, current, frequency, and modulation. [Brief explanation of the drawing]

[0008] [Figure 1] Figure 1 shows one embodiment of the test measurement system. [Figure 2] Figure 2 shows a part of one embodiment of the test measurement system. [Figure 3] Figure 3 shows an embodiment of the training process for an optical transceiver / transmitter tuning system. [Figure 4] Figure 4 shows an embodiment of the runtime process for an optical transceiver / transmitter tuning system. [Figure 5] Figure 5 shows a flowchart of one embodiment of a method for obtaining a set of tuning parameters. [Figure 6] Figure 6 shows a flowchart of one embodiment of a method for training a neural network to obtain optimized tuning parameters for an optical transceiver. [Figure 7] Figure 7 shows a flowchart of one embodiment of a method for obtaining optimized tuning parameters for an optical transceiver using a neural network. [Figure 8] Figure 8 shows examples of waveforms and tensor images. [Modes for carrying out the invention]

[0009] The test measurement device 10 may have a wide variety of components. Figure 1 shows one possible embodiment, and it may have more or fewer components. The device 10 in Figure 1 is connected to the DUT (device under test), in this case to an optical transceiver 12 having a transmitter. The test measurement device 10, such as an oscilloscope, may also have a photoelectric (O / E) conversion module 14, which is connected to the test measurement device 10 via a connector (e.g., 16). The processor 18 represents one or more processors and will be configured to perform various tasks. Figure 1 shows a machine learning system 22 as part of the test measurement device 10. Alternatively, the machine learning system may reside on a separate, independent device such as a computer and communicate with the test measurement device 10. Similarly, the test automation process (program) 20 executed by the processor 18 may reside within the device being manufactured, and the processor 18 is connected to it.

[0010] In this description, the term “test measurement device” as used in this application includes embodiments having an external computing device as part of the test measurement device 10. The memory 24 may reside on the test measurement device 10 or be distributed between the test measurement device 10 and the computing device. As will be described in more detail later, the memory 24 may include training data and runtime data used by the machine learning system 22. The test measurement device 10 may also have a user interface 26, which may have one or more of the following: a touchscreen, a video screen, a knob, a button, a light (light source), a keyboard, a mouse, etc.

[0011] Figure 1 shows an embodiment of the system during training. During training, the optical transceiver 12 is connected to the test measurement device 10 via the O / E conversion unit 14 and the connection unit 16. The test automation process 20 transmits transceiver tuning parameters to the transceiver 12, which may be referred to as "loading" them into the transceiver 12 in a process described in more detail. The test automation process 20 operates to create training tuning parameters and transmits them to the machine learning system 22. The tuning parameters used for training may be based, for example, on average parameters generated by historical testing using conventional test methods. The test measurement device 10 acquires waveforms from the transceiver (DUT) operating with the loaded tuning parameters and returns the acquired waveforms to the processor 18. The system performs measurements on the waveforms and transmits the waveforms to the machine learning system 22.

[0012] Figure 2 shows a part of one embodiment of the system in operation. In the first part of the process, the test automation process 20 generates reference tuning parameters and loads them into the transceiver 12. The test measurement device 10 retrieves and returns a set of reference waveforms from the transceiver 12, which the machine learning system 22 uses to predict optimized tuning parameters. The machine learning system 22 then returns predicted tuning parameters, also called optimized parameters. The test automation process 20 then loads these predicted parameters into the transceiver 12. The test measurement device 10 retrieves a newly generated waveform (referred to as the predicted waveform in this application) from the transceiver 12 operating with the predicted parameters and returns this waveform to the test automation process 20 to determine whether the transceiver 12 has passed or failed. This waveform may also be returned to the machine learning system 22 as training feedback.

[0013] In this description, the term “reference” parameter or waveform refers to a set of waveforms (series of waveforms) generated by the test automation process and used in both the training and runtime operation of the machine learning system. The term “predicted” refers to the output of the machine learning system. The term “optimized” refers to tuning parameters resulting from an optimization process by the test automation or machine learning system.

[0014] In the following description, waveforms acquired from transceiver 12 are used in the automated test process. The following embodiments use three waveforms for a nonlinear transceiver. However, it is not intended or suggested that the process is limited to a specific number of waveforms.

[0015] Figure 3 shows a block diagram of one embodiment of a training system for an optical transceiver tuning system using machine learning. The manufacturing test automation system measures the quantity of optical transceivers Tx. The test automation system optimally tunes the optical transceivers using a “standard” or conventional manufacturer tuning process. For training, it is necessary to provide training data by measuring and optimally tuning a large number of sample TX units. The number of units should be sufficient to provide a good distribution in the optimal parameter histogram of the DUT. Each TX unit is tested one at a time in a serial manner, as indicated by the input serial data stream to Tx34 (which is supplied to each Tx).

[0016] For optimal tuning, a histogram of each tuning parameter for each device is created for each temperature. This is then used to create a mean parameter set. Note that while it is not mandatory to use a mean parameter set in this process, it will be used in this explanation for simplicity. The user or system can then generate two variations (modifications) of the mean parameter set (MeanPar). For example, by adding to the mean parameter, a Delta1Par set can be generated, and by subtracting from the mean parameter, a Delta2Par set can be generated.

[0017] Next, the MeanPar set and the Delta1Par and Delta2Par parameter sets are used as reference parameters for system training and runtime. The purpose is to provide three sample points on the nonlinear characteristic curve of the parameters, so that the neural network has sufficient data to accurately predict the optimal parameter set by observing the processed waveforms from these reference parameter settings. The number of reference parameter sets may be from 1 to N, where N is limited by computation time and resources. In one example, 200 transceivers undergo a test consisting of 5 acquisitions using these three parameter sets, resulting in 15 waveforms for each transceiver, for a total of 3000 waveforms obtained for training. In Figure 3, the switch 30 is not constituted by a hardware switch, but is a logic block that causes each parameter set to be loaded into each of the transceivers 32 to 34, where N is again limited only by time and resources.

[0018] Next, the obtained waveforms are processed sequentially, which is represented by the switch 36 followed by the digitizer 38. The waveforms for each set of reference parameters are stored in a memory as shown in Figure 1. This requires 3 acquisitions by the A / D, with a different reference parameter set used for each acquisition. The test automation application controls the system temperature associated with a particular tuning parameter set. In this example, the temperature (temp) associated with each set of parameters is provided at 44, and similarly, the device noise profile is provided at 46. This temperature is input to the hyperspectral tensor builder 42 and incorporated into the tensor images used during training and runtime. The temperature may be represented as a bar graph in the image. However, it may alternatively be represented as a pie chart or any other form as needed.

[0019] As described above, the test automation process 20 would use a "standard" tuning process, which is replaced by the faster machine learning process of the embodiments. The optimal tuning parameters used in training may, at 41, undergo a normalization process or a standardization process depending on the data. The normalization process or standardization process rescales the tuning parameter values so that they fall within a normalized range that the neural network 22 can process well. In one embodiment, the normalization process would rescale the tuning parameters to fall within values from 0 to 1, or to a range from -1 to 1 centered at zero. One embodiment of the formula for the normalization process is y=(x-min) / (max-min), where y is the normalized value and x is the original value. If the system uses a pre-trained network that does not have normalization processing built in, this normalization processing would be used.

[0020] The user algorithm 40 tunes each of the Tx device samples with respect to the optimal tuning parameters, and then uses these optimal tuning parameters to train the network to associate them with given waveform shapes and characteristics such as temperature and oscilloscope noise. This association is always performed using the same input reference tuning parameters that were stored in the Tx (DUT) during waveform acquisition.

[0021] The hyperspectral tensor builder 42 may use a short pattern tensor generation process for machine learning. The hyperspectral tensor builder 42 uses the decoded pattern and input waveform to search for its record length and find a pre-defined pattern, which may be defined from unit intervals within the machine learning system, although this may be hidden from the user depending on the application. Several different patterns may be used. In the case of optical tuning, the system may utilize a short pattern of each level without transitions between five unit intervals for the machine to see the PAM4 levels. This allows the machine to recognize the Tx level parameter settings more quickly.

[0022] To recognize the FFE taps in the TX parameter settings, an impulse response code sequence (i.e., a short pattern) is extracted and placed on the image input to the neural network 22. The term hyperspectral is used to mean that the image input to the deep learning neural network 22 has three color channels labeled R, G, and B (red, green, and blue). The waveforms that make up the spectral image are not actually these colors, but rather the system combines various images created by the tensor builder using three color channels. The system collects three waveforms, and each of the images derived from each of the three waveforms goes into one of these color channels. For example, but not limited to, wfmMean generated from the MeanPar dataset may be on the red channel, wfmDelta1 from the Delta1Par dataset may be on the blue channel, and wfmDelta2 from the Delta2Par dataset may be on the green channel. There are no restrictions on these channels. If necessary, additional color channels may be incorporated in addition to RGB. The pre-trained deep learning network used in this application has only three color channels, but a custom deep learning network or multiple pre-trained 3-color channel networks may result in many more channels.

[0023] As will be explained in more detail later, this description presents two instances of a tensor builder. One is used to create images used for training an optimal parameter network. The other can be used to tune a network that outputs TDECQ (transmitter dispersion eye closure quaternary) and optimized FFE (feed-forward equalization) filter taps. In summary, a machine learning system is trained to receive a waveform tensor image and output an optimized set of transmitter parameters, and may further output an optimized set of FFE taps and TDECQ.

[0024] Figure 4 shows a block diagram of one embodiment of a runtime system for an optical transceiver tuning system using machine learning. At runtime, the user's automated testing application switches (represented by switch 30) three different reference sets of tuning parameters to a single transceiver being tested (e.g., 32), and captures a waveform for each. The TDECQ analysis, shown as 90 in Figure 6, resamples the waveform, decodes the pattern, and aligns the pattern with the waveform. TDECQ is not calculated at runtime. The TDECQ and FFE taps are output from a second trained deep learning network, shown as 100 in Figure 6. The hyperspectral tensor builder 42 blocks create multiple tensor images of the same type, including bar graphs created during training. These tensor images are input to a trained deep learning neural network 22. The output of the trained deep learning neural network 22 is the optimized tuning parameters, TDECQ, and optimized FFE taps. Next, the system applies the inverse normalization or inverse standardization block 43 to obtain the desired value from the output of the neural network.

[0025] Figure 5 shows one embodiment of a method for obtaining a histogram of optimized tuning parameters used in the training and runtime processes. This process uses a conventional “standard” tuning method to generate a histogram and mean values ​​used to create the average parameter structure (APS). The manufacturer connects the transceiver to the test measurement system in 50, and then, in 52, sets the temperature and waits for it to stabilize. Subsequently, in 54, the waveform from the transceiver is captured. TDECQ analysis in 56 allows the system to calculate the mean in 58. Similarly, the standard tuning method in 60 generates tuning parameters for each transceiver, which are averaged in 62. This also allows for the formation of a histogram used for training in 64. All of this information generates the average parameter structure (APS) in 66 (described in more detail in relation to Figure 6), which is sent to the training and runtime processes in 68.

[0026] The process is then looped as many times as necessary. For each transceiver, at 70, the temperature is varied until the transceiver completes testing over the desired temperature range. Then, at 72, the process moves on to the next transceiver until a sufficient number have been completed to ensure a sufficient level of accuracy.

[0027] Figure 6 shows a flowchart of an embodiment for training a machine learning system in preparation for performing runtime prediction of optimized transceiver parameters. The user connects the transceiver to the test measurement system at 80 and waits for the temperature to stabilize at 82. At 84, one of several reference parameter sets in the Mean Parameter Structure (APS) 68 of Figure 5 is loaded into the transceiver. The transceiver is then operated and at 86, the test measurement system acquires the waveform. Next, the user performs a standard tuning method at 88, and the system runs the TDECQ analysis module at 90. This is repeated until all three reference parameter sets have been run, as shown at 92. The TDECQ analysis module then sends the waveforms and patterns of all reference sets to the tensor builder at 94. The TDECQ module resamples the waveform, decodes the pattern, aligns the pattern and waveform, and then sends them to the tensor builder. In this embodiment, there is only one pattern, as it is the same for all the waveforms processed for training and runtime.

[0028] Tensor Builder 42 then generates several outputs and sends them to a neural network, which may consist of multiple subnetworks within the machine learning system. The Tensor Builder sends a level tensor to the first neural network (identified as neural network A), an impulse tensor to the transceiver FFE tap neural network B, an impulse tensor along with the optimized FFE tap array from the TDECQ module to the optimized FFE tap neural network C, and a composite tensor regarding the TDECQ results to neural network D. This continues until the current transceiver has cycled through all of the multiple temperatures at 96, and then at 98, it is repeated for all transceivers until a sufficient number have been processed, at which point the training process ends. Determining whether a sufficient number of transceivers have been processed may include testing the machine learning system to determine its accuracy.

[0029] Figure 8 shows an example of the original 4-level pulse amplitude modulation (PAM4) image derived from the waveform, and the resulting tensor. As mentioned above, this embodiment uses three channels. Figure 8 shows, on the left, multiple input waveforms from the three channels, which represent the input waveforms for red channel 140, blue channel 142, and green channel 144. Tensor 150 and the tensor at the same position spread across the figure correspond to the red channel, the position of tensor 152 corresponds to the blue channel, and the position of tensor 154 corresponds to the green channel.

[0030] Once training and validation are complete, the machine learning system will be operational at runtime. Figure 7 shows a flowchart of one embodiment of this process. At 110, the user connects the transceiver, and at 112, sets the temperature and waits for it to stabilize. At 114, the reference parameters contained in the mean parameter structure 68 are loaded into the transceiver as the appropriate parameters for this part of the process. As will be explained in more detail later, the appropriate parameters may vary depending on where the transceiver is in its cycle.

[0031] This process continues as in the training cycle, capturing or acquiring waveforms at 116 and performing TDECQ analysis at 118. At 120, the system verifies that the transceiver has been processed for all three reference parameter sets. Once all three reference sets are complete, the three waveforms and decoded patterns are sent to the tensor builder at 122. The tensor builder creates tensor images for all three waveforms and, as described above, combines all of these into a single RGB image. As in the training process, the tensor builder sends the level tensor, impulse tensor, and combined tensor image to the neural network 124. However, at runtime, the neural network / machine learning system generates optimized FFE taps and TDECQ information to supply the transceiver with a predictive optimization tuning parameter set.

[0032] After the prediction is made, at 126, the APS flag is set to false. The process returns to 114, and once the transceiver is set to the appropriate parameters, these parameters become the prediction tuning parameters. This also indicates that the system should capture only one waveform, generate only one tensor image, and that the neural network D generates a single TDECQ value. Since the APS flag remains false at 126, the value is checked at 130, and the transceiver either passes or fails. If it passes, at 128, the next temperature is selected if further testing is needed; otherwise, the process terminates.

[0033] In this way, multiple neural networks are trained within the overall neural network / machine learning system. Two networks predict the optimal transmitter tuning parameters, with the first network C being used to predict the FFE tap and network D being used to predict the TDECQ measurement.

[0034] This system uses TDECQ measurements as input to network D during training, and network D associates the input waveform with the TDECQ value. Similarly, the optimized FFE tap output from the TDECQ module is used to network C during training, and network C associates the waveform with the optimized tap value. Both networks A and B supplied tuning parameters that were associated with three reference waveforms. Network A supplied level and gain type parameters. Network B associated FFE tuning parameters. Since the transmitter has a built-in FFE, the taps of the FFE filter are the tuning parameters.

[0035] The optimized FFE taps output from the TDECQ module are varied. These are not TX tuning parameters, but rather represent what you would set in the FFE to open the eye as wide as possible for TDECQ measurements. These FFE taps are applied to the waveform before performing the TDECQ measurement.

[0036] Although the embodiments described above have been illustrated using an exemplary application, a tuning process for an optical transceiver, embodiments of the present invention can be used in any process involving the adjustment of operating parameters of any type of device under test, and would be particularly useful in improving processes involving iterative adjustment of operating parameters to achieve desired performance measurements of the DUT or to achieve pass in the test results of the DUT.

[0037] Aspects of the disclosed technology include specially created hardware, firmware, and digital systems. It can operate on a specially programmed general-purpose computer, including a Gunal processor or a processor that operates according to programmed instructions. The terms “controller” or “processor” in this application mean microprocessors, microcomputers, ASICs, and dedicated hardware controllers, etc. Aspects of the disclosed technology can be implemented by computer-readable data and computer-executable instructions, such as one or more program modules, executed by one or more computers (including monitoring modules) or other devices. 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 in 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 program modules may be combined or distributed as needed in various embodiments. Furthermore, these functions may be embodied in whole or in part in firmware or hardware equivalents such as integrated circuits, field-programmable gate arrays (FPGAs), etc. Certain data structures can be used to more effectively implement one or more aspects of the disclosed technology, and such data structures are considered to be within the scope of computer executable instructions and computer usable data described in this application.

[0038] 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.

[0039] 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.

[0040] 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.

[0041] In addition, the description of this application refers to specific features. All features disclosed herein, including the claims, abstract and drawings, and all steps in all disclosed methods or processes, may be combined in any way, provided that at least a portion of them are not mutually exclusive. Each of the features disclosed herein, including the claims, abstract and drawings, may be replaced by an alternative feature that serves the same, equivalent or similar purpose, unless otherwise specified.

[0042] 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

[0043] 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.

[0044] Embodiment 1 is a test measurement system comprising a test measurement device, a connection unit that allows the test measurement device to be connected to an optical transmitter, and one or more processors, wherein the one or more processors execute a program that causes the one or more processors to perform the following: setting the operating parameters of the optical transmitter to reference operating parameters, acquiring waveforms from the optical transmitter, repeatedly executing a program that causes the one or more processors to perform the following for each of a predetermined number of sets of reference operating parameters: setting the operating parameters and acquiring the waveforms, generating one or more tensors from the acquired waveforms, sending the one or more tensors to a machine learning system to obtain a set of predicted operating parameters, setting the operating parameters of the optical transmitter to the predicted operating parameters, and testing the optical transmitter using the predicted operating parameters. The optical transmitter may be included in a transceiver.

[0045] Example 2 is the test measurement system of Example 1, further configured such that one or more processors perform a process of repeatedly executing a program for setting each temperature within a set temperature range.

[0046] Example 3 is a test measurement system according to Example 1 or 2, wherein one or more processors are distributed among the test measurement device, the test automation system, and the machine learning system independent of the test measurement device.

[0047] Example 4 is a test measurement system according to any of Examples 1 to 3, wherein the machine learning system has multiple neural networks, one or more of which generate predictive operating parameters, one of which generates predicted measurement values, and one of which generates a set of tap values ​​for an optimized FFE (feed-forward equalization) filter.

[0048] Example 5 is a test measurement system according to any of Examples 4, wherein the predetermined number of sets of reference parameters is 3, one set is the average value of the parameters, one set has a value higher than the average value, and one set has a value lower than the average value.

[0049] Example 6 is a test measurement system according to any of Examples 1 to 5, wherein one or more processors are further configured to execute a program that causes one or more processors to perform a process of applying an inverse normalization process to the predicted operating parameters.

[0050] Example 7 is a method for tuning an optical transmitter, comprising: connecting the optical transmitter to a test measurement device; setting the operating parameters for tuning the transmitter as reference operating parameters; acquiring a waveform from the optical transmitter; repeating the process of setting the operating parameters and acquiring the waveform for each of a predetermined number of sets of reference operating parameters; generating one or more tensors from the acquired waveform; sending the one or more tensors to a machine learning system to obtain a set of predicted operating parameters; and tuning the transmitter. The system includes a process for setting the above-mentioned operating parameters for the purpose of the operation to the above-mentioned predicted operating parameters, and a process for verifying the above-mentioned predicted operating parameters.

[0051] Example 8 is the method of Example 7, further comprising the process of repeating the above method for each temperature within the temperature range.

[0052] Example 9 is the method of Example 7 or 8, wherein the process of generating one or more tensors includes the process of generating a level tensor, an impulse tensor, and a composite tensor.

[0053] Example 10 is a method of any of Examples 7 to 9, wherein the process of generating a composite tensor includes the process of generating an image using three or more channels.

[0054] Example 11 is the method of Example 10, wherein the image includes a bar graph image representing at least one of temperature values ​​and noise values.

[0055] Example 12 is a method of any of Examples 7 to 11, further comprising a process for generating reference operating parameters, the process of tuning a plurality of optical transmitters at a plurality of temperatures, the process of generating a histogram of each of the tuning parameters for each of the tuned plurality of optical transmitters at each of the temperatures, the process of calculating an average reference parameter set by averaging the values ​​of each parameter in the histogram, the process of creating a first delta reference parameter set by increasing the values ​​in the average parameter set, and the process of creating a second delta reference parameter set by decreasing the values ​​in the average parameter set.

[0056] Example 13 is a method according to any of Examples 7 to 13, wherein the process of generating one or more tensors includes the process of generating a tensor having three channels: one channel for the mean reference tensor, one channel for the first delta tensor, and one channel for the second delta tensor.

[0057] Example 14 is a method of any of Examples 7 to 13, wherein the process of sending one or more tensors to a machine learning system includes the process of sending one or more tensors to two neural networks in the machine learning system to obtain optimized operating parameters, the process of sending an impulse tensor to a third neural network to obtain taps for an optimized FFE (feed-forward equalization) filter, and the process of sending a composite tensor to a fourth neural network to obtain predicted TDECQ (transmitter and dispersion eye closure penalty quaternary) measurements.

[0058] Example 15 is a method according to any of Examples 7 to 14, wherein the process for verifying the predicted operating parameters includes a process for measuring the output of the transmitter and a process for determining whether the output is within a predetermined range.

[0059] Example 16 is a method of any of Examples 7 to 15, further comprising the process of applying an inverse normalization process to the above-mentioned predictive operation parameters.

[0060] Example 17 is a method for training a machine learning system for tuning an optical transmitter, comprising the process of creating a predetermined number of sets of reference operating parameters, and for each of the sets of reference operating parameters, the process of setting the operating parameters of the optical transmitter to the set of reference operating parameters, the process of acquiring a waveform from the optical transmitter, the process of tuning the optical transmitter to achieve a desired output, and the process of performing TDECQ analysis on the waveform, repeated, and the results of the analysis and The system comprises the processes of supplying a waveform to a tensor builder, transmitting one or more tensors, along with associated waveforms and reference operating parameters, to one or more neural networks, and repeating the above method until a sufficient number of transmitters are tuned.

[0061] Example 18 is the method of Example 17, wherein the process of creating a predetermined number of sets of the reference operating parameters includes: a process of tuning multiple optical transmitters at multiple temperatures; a process of generating histograms of each of the parameters for each of the multiple tuned optical transmitters at each of the temperatures; a process of calculating an average reference parameter set by averaging the values ​​of each of the parameters in the histograms; a process of creating a first delta reference parameter set by increasing the values ​​in the average parameter set; and a process of creating a second delta reference parameter set by decreasing the values ​​in the average parameter set.

[0062] Example 19 is the method of Example 17 or 18, further comprising a process for normalizing the above-mentioned reference parameters.

[0063] Example 20 is a method of any of Examples 17 to 19, wherein the one or more neural networks include one or more neural networks for levels, one network for supplying values ​​for an FFE (feed-forward equalization) filter, and one network for TDECQ values.

[0064] All functions disclosed in the specification, abstract, claims and drawings, and all steps in any method or process disclosed, may be combined in any combination, except where at least some of such functions or steps are mutually exclusive. Each of the functions disclosed in the specification, abstract, claims and drawings may be replaced by an alternative function that serves the same, equivalent, or similar purpose, unless otherwise specified.

[0065] For the sake of explanation, specific embodiments of the present invention 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 present invention should not be limited to anything other than the appended claims. [Explanation of Symbols]

[0066] 10 Test and measurement device 12. Optical transceiver (DUT: Device under test) 14. Optical / Electric (O / E) Conversion Module 16 Connection part of the test and measurement device 18 processors 20. Test automation process (program) 22 Machine learning systems or neural networks 24 memory 26 User Interface 30 switches 32 transceivers 34 transceivers 36 switches 38 Digitizer 40 User Algorithms 42 Hyperspectral Tensor Builder 43 Inverse Normalization or Standardization Block 44 Temperature Parameters 46 Noise Profile Parameters 140 Red Channel 142 Blue Channel 144 Green Channel 150 Red channel tensor 152 Blue channel tensor 154 Green Channel Tensor

Claims

1. A test and measurement system, Test and measurement device, A connection section that allows the test measurement device to be connected to an optical transmitter, One or more processors and Equipped with, One or more processors execute a program and perform the following on one or more processors The process of setting the operating parameters of the above optical transmitter to reference operating parameters, The process of acquiring the waveform from the above optical transmitter, A process of repeatedly executing a program that causes one or more processors to perform the process of setting the above operating parameters and the process of acquiring the above waveform for each of a predetermined number of sets of the above reference operating parameters, A process to generate one or more tensors from the captured waveform, The process involves sending one or more of the above tensors to a machine learning system to obtain a set of predictive operating parameters, The process of setting the above operating parameters of the above optical transmitter to the above predicted operating parameters, A process to test the optical transmitter using the above-mentioned predicted operating parameters and A test and measurement system configured to perform the following.

2. The test measurement system according to claim 1, further configured such that one or more of the above-mentioned processors are made to perform a process of repeatedly executing a program for setting each temperature within a temperature setting range.

3. The test measurement system according to claim 1, wherein the machine learning system has multiple neural networks, one or more of which generate the predicted operating parameters, one of which generates the predicted measured values, and one of which generates a set of tap values ​​for the optimized FFE filter.

4. A method for tuning an optical transmitter, The process of connecting the above optical transmitter to the test measurement device, The process of setting the operating parameters for tuning the above optical transmitter to reference operating parameters, The process of acquiring the waveform from the above optical transmitter, A process that repeats the process of setting the above operating parameters and the process of acquiring the above waveform for each of a predetermined number of sets of the above reference operating parameters, A process to generate one or more tensors from the captured waveform, The process involves sending one or more of the above tensors to a machine learning system to obtain a set of predictive operating parameters, A process to set the above operating parameters for tuning the above optical transmitter to the above predicted operating parameters, The process for verifying the above predicted operating parameters and A tuning method for an optical transmitter equipped with [a specific feature / ability].

5. The process of sending one or more of the above tensors to a machine learning system is, The process involves sending one or more of the above tensors to two neural networks within the above machine learning system to obtain optimization operating parameters, The process involves sending an impulse tensor to a third neural network to obtain the taps of the optimized FFE filter, The process involves sending a synthetic tensor to a fourth neural network to obtain a predicted TDECQ measurement. A method for tuning an optical transmitter according to claim 4.

6. A method for training a machine learning system for performing the tuning method of an optical transmitter according to Claim 4, A process to create a predetermined number of sets of reference operating parameters, For each of the above sets of reference operating parameters, The process of setting the operating parameters of the above optical transmitter to a set of reference operating parameters, The process of acquiring the waveform from the above optical transmitter, A process to tune the optical transmitter to achieve the desired output, The process involves performing a TDECQ analysis on the above waveform. A process that repeats, The process of supplying the results of the above analysis and the above waveform to the tensor builder, The process involves transmitting one or more tensors, along with associated waveforms and reference operating parameters, to one or more neural networks. The above method is repeated until a sufficient number of the above optical transmitters are tuned. A method for training a machine learning system for performing the tuning method of the optical transmitter of claim 4, comprising the features of claim 4.

Images