Combined high-level statistics and artificial intelligence signal analysis

Abstract

A test and measurement instrument for analyzing signals using machine learning. The test and measurement instrument can determine a recovered clock signal based on a digital signal, set a window position for a fast Fourier transform of the digital signal, window the digital signal into a series of windowed waveform data based on the window position, transform each of the windowed waveform data into frequency-domain windowed waveform data using a fast Fourier transform, and determine high-order spectral data for each of the frequency-domain windowed waveform data. The test and measurement instrument includes a neural network configured to receive the high-order spectral data of the frequency-domain windowed transform data and classify each of the windowed waveform data based on the high-order spectral data.

Description

[0001] priority

[0002] This disclosure claims the benefit of U.S. Provisional Application No. 62 / 900,422, filed September 13, 2019, entitled “COMBINED BICOHERENCE AND ARTIFICIAL INTELLIGENCE SIGNAL ANALYSIS,” which is incorporated herein by reference in its entirety. Technical Field

[0003] The present disclosure relates to systems and methods related to test and measurement systems, and in particular, to test and measurement systems for signal analysis. Background Art

[0004] Jitter is a well-known term used to define the deviation of events in a signal from their ideal timing. Jitter causes significant edges in the data bit sequence to shift from their ideal positions. This jitter can result from errors in the position of recovered clock edges, or it can be caused by system reflections or system intersymbol interference derived from the system transfer function. Other effects, such as distortion, may also be present. Furthermore, there are other waveform measurements of interest that help characterize the accuracy and performance of channel data recovery. In modern serial data communication systems, the serial data clock is typically not transmitted along with the data, so jitter can lead to data errors at the receiver. Therefore, characterizing and classifying various measurements of waveform parameters and system characteristics is crucial.

[0005] Test and measurement instruments are continually being improved to more accurately and quickly identify and measure jitter from the signals being measured.

[0006] Embodiments of the present disclosure address these and other deficiencies of the prior art. BRIEF DESCRIPTION OF THE DRAWINGS

[0007] Aspects, features, and advantages of embodiments of the present disclosure will become apparent from the following description of embodiments with reference to the accompanying drawings, in which:

[0008] Figure 1 is an example diagram of the absolute phase reference position used for bispectral calculation;

[0009] Figure 2 is a bandwidth diagram illustrating a time versus frequency plot of an ideal pseudorandom binary sequence signal;

[0010] Figure 3 is a block diagram of an example test and measurement system according to some configurations of the present disclosure;

[0011] Figure 4is a block diagram of another example test and measurement system according to some configurations of the present disclosure;

[0012] Figure 5 is a block diagram of another example test and measurement system according to some configurations of the present disclosure;

[0013] Figure 6 is a block diagram of another example test and measurement system according to some configurations of the present disclosure;

[0014] Figure 7 is a timing diagram illustrating the recovered clock signal and the window function of the input signal. DETAILED DESCRIPTION

[0015] This document discloses an example of a test and measurement system that can analyze an input signal, such as a pseudo-random binary sequence (PRBS) signal, by applying clock recovery associated with synchronization to bi-coherence spectroscopy based on calculations such as short-time Fast Fourier Transform (FFT) calculations. The disclosed examples utilize bi-coherence spectroscopy to identify jitter and separate it from the signal, which includes other distortions such as inter-symbol interference (ISI), clock edge jitter, and other distortions of the signal. The vector space of the bi-coherence spectroscopy results can be fed into a neural network that can be trained to recognize and classify various characteristics of the signal.

[0016] To understand the bicoherence spectrum, the definitions of first-order and second-order cumulants are provided. The first-order cumulant of a stationary process is the average value. The first-order cumulant can be written using the notation in equation (1), where E is the expected value, which is the average value of the signal:

[0017] (1).

[0018] For higher-order cumulants, the mean is usually subtracted.

[0019] The second-order cumulant in the time domain is the autocorrelation function and is written as shown in equation (2), where the overline indicates the complex conjugate:

[0020] (2).

[0021] As shown in equation (2), autocorrelation is similar to filter convolution. Filter convolution reverses the order of the filter coefficients in time before multiplication, accumulation, and shifting. For autocorrelation, the order of x(n) and x(n+k) is not reversed as K is shifted. When the second-order cumulant is transformed into the frequency domain, it is called the autopower spectrum and is expressed by equation (3). Multiplying x with its complex conjugate gives zero phase. The square of the x(n) term represents the power. Therefore, the power spectrum represents the second-order cumulant in the frequency domain and can be written as shown in equation (3):

[0022] (3).

[0023] An alternative representation of equation (3) is shown in equation (4), where X is the spectrum of x:

[0024] (4).

[0025] The third-order cumulants transformed into three-dimensional space are called bispectra. The third-order cumulants in the time domain are shown in equation (5):

[0026] (5).

[0027] The bispectrum is defined as the Fourier transform of the third-order cumulant, as shown in equation (6):

[0028] (6).

[0029] The bispectrum with an alternative representation in the frequency domain is shown in Equation (7) as a triple product:

[0030] (7).

[0031] The variable B is the bispectrum at frequencies f1 and f2, and X(f) is the Fourier transform of the signal x(t). The overline indicates the complex conjugate. Using the complex multiplication rule, the amplitude of the bispectrum is equal to the product of the amplitudes of the function X at each frequency. The phase of the bispectrum is the sum of the phases of X at each frequency. The bispectrum depends on the amplitude of X at each frequency.

[0032] If X is a continuous, time-invariant signal, and if X(f1), X(f2), and X(f1+f2) are perfectly phase-locked, then if you calculate the short-time FFT at several random points in the time of x(t), the bispectrum will always have the same value. If added together, the bispectrum at each point will sum and not cancel. The phase from each short-time FFT is adjusted relative to an absolute reference time point on the waveform.

[0033] However, if X is a time-varying signal, that is, if the phase at each frequency is random in time and the amplitude is constant, then when each bispectrum is added together, there will be a tendency for them to cancel. The result will tend toward zero, depending on the relationship between the phase and amplitude of the time-varying signal. Therefore, for time-varying signals, such as pseudo-random binary sequence signals, clock recovery is performed so that the phase of the signal does not average to zero.

[0034] Bi-coherence spectroscopy normalizes the bi-spectrum to eliminate amplitude correlation, as shown in equation (8):

[0035] (8).

[0036] where n is the index of each FFT frame and ranges from 1 to the number of FFT frames in the waveform record. The numerator of equation (8) contains the amplitude of the spectrum. If the phase coupling is high, the amplitude value becomes larger, but if the coupling is low, it is close to zero. In the denominator of equation (8), the summation is only for the amplitude, essentially setting the phase to all zeros, thereby normalizing the bispectrum to obtain the bicoherent spectrum b, whose maximum value is 1 and the minimum value is 0.

[0037] When two sine waves of different frequencies are added together and passed through a nonlinear element, a multiplication process occurs. The resulting output consists of two sine waves whose frequencies are the sum and difference of the two frequencies, as shown in Equation (9):

[0038] (9).

[0039] Therefore, in the bispectral equation shown in Equation (7), X(f1) and X(f2) can be two in-phase sinusoidal waves, and X(f1+f2) can be the distortion component after passing through the nonlinear device. Bispectral analysis is often used to analyze nonlinear distortion components.

[0040] When X is a time-invariant signal composed of coherent frequencies, the bispectrum can be calculated by taking multiple short-time FFTs along the time record of x(t) and then summing the bispectra from each FFT. The zero-phase reference point of the FFT is typically defined as the time position of the first sample in the FFT time window. As the FFT window moves along the time record, the phases of X(f1) and X(f2) rotate as a function of their time position. Therefore, the bispectrum from F based on this phase reference will result in the sum of the bispectra from all FFT positions approaching zero.

[0041] Therefore, an absolute phase reference position on the x(t) waveform is required. The phase of each FFT result can then be adjusted relative to the absolute time position, such as in Figure 1 As shown. Figure 1 In Figure 1, the phase of the sine wave 100 is 90 degrees relative to the absolute zero phase reference point 102. However, the phase calculated from FFT 1 will be 0.0 degrees, and the phase calculated from FFT 2 will be 180 degrees. Therefore, the phase value from FFT 1 is corrected according to offset 1 so that FFT 1 has a zero phase reference 104. The phase value of FFT 2 is corrected according to offset 2 so that FFT 2 has a zero phase reference 106. This allows FFT 1 and FFT 2 to return a 90 degree phase of the signal relative to the absolute zero phase time location 102. This allows the bispectral values to be calculated more accurately by averaging over multiple time windows.

[0042] A pseudo-random binary sequence (PRBS) signal has a pseudo-random bit pattern of length determined by the order of the polynomial used to generate the PRBS signal. Therefore, the signal is not time-invariant and the bispectrum will tend towards zero.

[0043] Consider that an ideal PRBS signal consists of a series of ideal steps located at multiples of the unit interval (UI). Figure 2 A time versus frequency plot 200 of the derivative of x(t) with respect to time t is shown in . Since this is an idealized spectrogram, the frequency content of each pulse is the same. Figure 2 An idealized spectrogram is shown, which cannot be computed from a short-time FFT in reality due to the resolution tradeoff between time and frequency, but is shown for explanation purposes. Figure 2 For example, to capture only one pulse in the FFT window, a very short record length of less than one bit interval is required. This record length does not allow for accurate measurement of the low-frequency components of the signal. To achieve higher resolution for determining jitter and ISI effects, oversampling is desirable.

[0044] As the FFT window gets wider, it starts to include two or more bit intervals of the signal. For an ideal pulse at each time position, such as Figure 2 As shown in , where bit transitions occur, the spectrum has an amplitude of 1 at each frequency and the phase of each frequency is equal to zero relative to the temporal position of each individually transitioned pulse.

[0045] However, a wider FFT window will include a set of frequencies from each pulse within the window, with the phase of the pulses varying with the delay between edges. These are summed together in the FFT calculation to provide a single combined phase at each frequency. If there are only two pulses within the FFT window, then by definition, one will be a rising edge, and the other must be a falling edge if it is a non-return-to-zero (NRZ) signal. A four-level pulse amplitude modulation (PAM4) signal can have two consecutive rising edges, two consecutive falling edges, or one falling edge and one rising edge. For ideal edges, the rising phase is 180 degrees different from the falling phase.

[0046] If multiple edges are in a window, the phase, jitter, ISI, etc. of each edge will be added together. If the pattern and position of the edges are different from one window to the next, the bispectral sum will be averaged over a larger range of differences, resulting in a smaller bicoherent spectrum value.

[0047] Current oscilloscopes have a sampling rate of approximately 200 gigabits per second (GS / s) to acquire a 53-gigabaud (GBd) PAM4 signal with a symbol width of 1 / 26.5 gigahertz (GHz) within a period of 37.7 picoseconds (ps). The sampling period is 5 ps, resulting in 7.547 samples per bit interval. The FFT length, encompassing four symbol periods, is approximately 32 points.

[0048] When the impulse response of the PRBS signal channel in a test and measurement instrument is convolved with the ideal bit pattern, the edge of the past bit will affect the following bits. This is called inter-symbol interference or ISI. ISI results in non-uniform rise and fall times that depend on the previous bit pattern. ISI will have an effect on the phase and amplitude of the spectrum according to the derivative of each edge. This causes edge jitter, which is usually classified as data-dependent jitter (DDJ). The transmitter clock jitter of the device under test may cause Figure 2 The vertical lines in the spectrum graph become wider.

[0049] Figure 3 An example test and measurement system 300 according to some examples of the present disclosure is illustrated. A signal x(n) is received at the input of a test and measurement instrument. An optional continuous time linear equalizer (CTLE) 302 may be provided to perform partial de-embedding of the test fixture and / or serial data link. Another optional filter (not shown) may also be provided, a de-embedding / embedding filter for de-embedding the test fixture and / or channel response. The de-embedding / embedding filter may be located near the CTLE 302, either before or after it.

[0050] Clock recovery 304 is performed by one or more processors. Clock recovery 304 receives an input signal directly or receives an equalized input waveform from CTLE 302 to generate a recovered clock waveform and / or a series of edge crossings. In some examples, an explicit clock obtained from an analog-to-digital converter can be used, or a burst explicit clock obtained in a double data rate (DDR) memory system can be used.

[0051] The input signal and the recovered clock are sent to the window sequencer 306. In some alternative embodiments, instead of sending the input signal to the window sequencer 306, the equalized signal from the CTLE 302 can be sent. The window sequencer 306, executed by one or more processors, keeps track of any short-time FFT window position relative to the recovered clock signal and the input waveform data. As will be discussed further below, the window sequencer 306 allows the use of bispectral or bi-coherent spectral functions to extract waveform characteristics in a meaningful way.

[0052] In some examples, an optional resampler 308 may be provided. The resampler 308 resamples the signal within the FFT window interval so that the starting point of the FFT is located at the center of the unit interval (UI). For cases where the input signal is a PRBS sequence, the resampler 308 eliminates the effects of asynchronous sampling relative to the window start time. Thus, the resampler 308 eliminates analog-to-digital sample jitter. In other words, after resampling, there is always a data sample exactly at the start of the window. For cases where the input signal is continuous, the resampler 308 can be used to ensure that the phase of the spectrum is relative to an absolute time reference point on the waveform.

[0053] Before performing an FFT 312 to convert the signal to the frequency domain, a window function 310, such as Figure 3 The Tukey window shown. Window function 310 can isolate a portion of the signal so that an FFT 312 can be performed on it. FFT 312 is performed by one or more processors and is a short-time FFT applied to the windowed waveform data at each time location determined by window sequence generator 306.

[0054] The phase reference point of FFT 312 is used when calculating the signal's higher-order spectral data 314. The recovered clock from clock recovery 304 is used to position the phase reference point for each FFT. Using a position based on the recovered clock allows the edges in the FFT window to be coherent with respect to the clock phase reference position.

[0055] The higher-order spectral data 314 is determined by one or more processors by acquiring a spectrum from the FFT 312 and transforming the data into a bispectrum. The higher-order spectral data 314 may include a bispectrum, a phase of the bispectrum, and / or a bicoherent spectrum as complex values. In order to determine the bicoherent spectrum, the bispectrum is normalized to determine the amplitude value. That is, before the higher-order spectral data 314 is fed to subsequent operations, the bispectral data 314 may be represented by at least one of the following: an amplitude value, a phase value, and / or a complex value. Throughout this disclosure, the higher-order spectral data 314 may include a bispectrum, a bicoherent spectrum, and / or a phase of the bispectrum. The higher-order spectral data 314 may also include higher-order statistics, such as fourth order or higher order, which may be determined based on the spectrum.

[0056] In some examples, a threshold gate 316 can be provided to retain bispectral samples whose amplitude values are above a specified threshold level. The threshold level can be an amplitude threshold, a phase threshold, and / or a complex threshold. The threshold level can be set by a user or can be determined by one or more processors. Although not shown, the test and measurement instrument can include a user interface for receiving a user-specified threshold level.

[0057] If the threshold gate 316 is an amplitude threshold gate 316, such as Figure 3As shown, the amplitude threshold gate 316 passes the bispectrum if the amplitude is greater than or equal to the threshold, and sets the bispectrum to zero or some other low level when the amplitude is below the threshold. Alternatively, the bispectrum phase can be gated based on the bicoherent spectrum amplitude threshold or the bispectrum phase. In some examples, the bispectrum can be complex-valued, and a complex threshold can be used.

[0058] The output of the threshold gate 316 is sent to the neural network 318. The neural network 318 receives the bispectrum and / or bicoherence spectrum in the form of at least one of amplitude, phase, and / or complex numbers. Through the neural network 318, machine learning is applied to the bispectrum and / or bicoherence spectrum to classify and decode the signal. That is, the neural network 318 can process the threshold-gated bispectrum or bicoherence spectrum to obtain a desired learned output response. The output of the neural network 318 can be many different outputs, such as, but not limited to, bit error rate, distortion, pattern decoding, jitter measurement, ISI measurement, signal-to-noise ratio, etc. Although neural networks are shown in the various figures, the examples of the present disclosure are not limited to neural networks, and neural networks can mean any form of machine learning.

[0059] Figure 4 Another example of a test and measurement instrument 400 according to some examples of the present disclosure is illustrated. In this example, Figure 3 Similar components are given the same reference numerals and are not discussed further.

[0060] Figure 4 The test and measurement instrument 400 is similar to Figure 3 314. The test and measurement instrument 300 of FIG. 314 is similar to the test and measurement instrument 300 of FIG. 314, but provides a second threshold gate 402. The threshold of the second threshold gate 402 can be set by a user or one or more processors of the test and measurement instrument 400. The second threshold gate 402 receives the output from the FFT 312 and outputs only data that does not violate the threshold setting to the neural network 318. The neural network 318 can then use the output of the FFT and the output of the high-order spectral data 314 to classify the signal. The second threshold gate 402 can be one of an amplitude gate, a phase gate, or a complex threshold gate and can be the same type as the first threshold gate 316.

[0061] Figure 5 5 shows another example of a test and measurement system 500. Figure 3 and Figure 4 Similar components are given the same reference numerals and are not discussed further herein. The inclusion of the FFT 312 and the higher-order spectral data 314 incorporates second-order and third-order statistics, respectively. Some examples may also include the possibility of adding higher-order statistics, to be used alone or in combination with any lower-order statistics.

[0062] In some alternative examples, the high-order spectral data 314 window on the signal can be processed by separating the spectral sum from the sum of similar modes in each FFT 312 window of the reference recovered clock. The dual spectral windows of similar modes are then summed. For example, if there are 5 symbol intervals in one FFT window, this will allow for 32 modes and 32 different bi-coherent spectral result spectra. The test and measurement instrument 500 provides many different neural networks 318, as discussed in more detail below.

[0063] In the test and measurement system 500, an input signal x(n) is received at both the CTLE 302 and a derivative function 502, which can provide a derivative of the input signal x(n) and is executed by one or more processors of the test and measurement instrument 500. The output of the derivative 502 is also sent to the window sequencer 306.

[0064] As mentioned above, if we assume that there are five UIs in the FFT window, there are 32 possible bit patterns. The test and measurement instrument 500 includes a neural network pattern identifier 504. The neural network pattern identifier 504 receives the output of the FFT 312 and classifies the pattern of the FFT window. The neural network pattern identifier 504 can identify a pattern for several bits within the window, even in the presence of some noise and ISI.

[0065] The output pattern from the neural network pattern identifier 504 is sent to the multiplexer switch 506. The multiplexer switch 506 can select one of a different number of paths to send the output of the FFT 312. In the example above, there are 32 different patterns, so in such an example, 32 different paths will be provided. The number of paths provided or used can correspond to the number of potential patterns.

[0066] As discussed above, in each path, high-order spectral data 314 is determined. The output of the high-order spectral data 314 is sent to the neural network 318 for classification. In some examples, an optional summing or averaging block 508 can be provided in the path. The high-order spectral data 314 tends to reduce the noise to zero and maintain the coherent portion of the spectrum. However, jitter will cause these phases to vary from one pattern to the next. The summing block 508 averages the output of each high-order spectral data 314 that contains the same bit pattern as the previous one. This will tend to average out the jitter shown in the bi-coherent spectrum results. The neural network 318 can compare the averaged bi-coherent spectrum to the instantaneous bi-coherent spectrum output to provide a measure of jitter.

[0067] Although Figure 5Not shown, but in some examples, each individual path may also include a threshold gate 316. The threshold gate may include an average threshold and / or any one of an amplitude, phase, or complex threshold for each instantaneous bi-coherent spectrum output.

[0068] Additionally, the output of the neural network pattern identifier 504 can be sent to a transfer function block 510, which uses one or more processors to determine a transfer function. The output of the FFT 312 is also sent to the transfer function block 510. The transfer function block 510 can determine the transfer function of the transmitter and the serial data channel. One method of doing this can be to divide the spectrum of the ideal pattern waveform by the spectrum of the acquired waveform and then perform an inverse FFT (IFFT) to obtain the impulse response. Other known methods can also be used.

[0069] Figure 6 Another example of a test and measurement system 600 is shown. Figure 3-5 Similar components are given the same reference numerals and are not discussed further herein. The test and measurement system 600 may include a second multiplexer switch 602 having a plurality of output paths. The output paths correspond to the number of modes.

[0070] When a pattern is identified by the neural network pattern identifier 504, the multiplexer switch 602 determines the path that receives the output from the FFT 312. Each path may also include an optional summing block 508 to average the FFTs 312 for each window of similar patterns. The transfer function block 510 may include the output from each summing block 508, along with the recovered clock signal from the clock recovery 304 and the original input signal x(n), to calculate the transfer function.

[0071] We will now describe how to compute the short-time FFT 312 of each segment of the input signal x(n) at different window time positions along the time record. The first point of the FFT window is the time position representing zero phase. In order to maintain a constant phase reference for the bi-coherent spectrum calculation, the bit patterns in the gated region of each FFT window should be identified and categorized. Thus, only identical bit pattern sequences will be added together for averaging the bi-spectrum output. The averaged bi-spectrum is then normalized to obtain the bi-coherent spectrum, as described above in equation (8).

[0072] The width of the FFT 312 window is the integer number of bits in the input signal x(n) being analyzed. The zero-phase reference point of the FFT 312 is at the start time of the window. As mentioned above, Figure 2As shown, the spectrum of the derivative of an ideal step at each bit transition is constant amplitude at all frequencies from DC or zero to its Nyquist frequency. The phase of all ideal pulses in the FFT 312 window is zero at all frequencies, with the phase reference point being the time position of the pulse. However, the zero phase reference point for each FFT 312 will be at the beginning of the FFT 312, so that the frequency of the entire frequency band representing each edge will be different from the phase of other edges within the FFT 312 window.

[0073] The FFT result will have a set of frequencies and phases obtained from the sum of the frequencies from all edges. It can be expected that each bit pattern combination within the window interval has a unique set of phases and amplitudes in the spectrum, which can be used in the neural network 318 to decode the bit sequence within the window in order to classify each pattern into a different higher-order spectral data 314 path for summing them.

[0074] Figure 7 The relationship between the recovered clock 702, the input signal x(n) 704 and the Tukey window function 706 is shown. Figure 7 In the diagram of FIG, the input signal x(n) 704 is a PRBS data signal. Figure 7 As shown in the figure, the unit interval (UI) is the width of one pulse of the recovered clock.

[0075] like Figure 7 As shown, the Tukey window 706 is N samples long and starts and ends at the center of the clock cycle. The zero phase reference of each FFT is set at the beginning of each window. The FFT window is relatively short compared to the number of samples and UIs that are allowed to have a reasonable size for the bispectrum analyzed by the neural network 318.

[0076] The window is an integer number of UIs, where the starting point of each FFT is at the center of the UI, such as Figure 7 One way to address transients at the end of the windowing and cyclic windowing of the FFT 312 may be to calculate the derivative of the data, as illustrated in some of the figures discussed above, and then place the starting point of the FFT window at the center of the middle of the UI where the derivative is zero. Placing the starting point of the FFT window at the center of the middle of the UI where the derivative is zero may avoid transient leakage at the beginning and end of the recording.

[0077] Additionally, a Tukey window can be applied to taper the ends of the FFT record to zero. This is done to reduce the effect of ISI from the previous window, which is different for the same pattern in the window classified into a specific bispectral path. When using a five-bit window interval, each path has one of 32 possible patterns.

[0078] In some examples, taking derivatives may not be preferable because it can increase high-frequency noise. However, the bi-coherent path can take the sorted bit patterns and calculate a bi-coherent spectrum that tends to remove the noise. Furthermore, as shown in the various figures, the pattern paths can be averaged to remove the additional noise and jitter variations from one similar bit pattern to the next.

[0079] The resampler block 308 can remove the sampling clock jitter relative to the zero-phase reference point of the FFT window. This ensures that the time domain sampling of the waveform always occurs exactly simultaneously with the zero-phase reference.

[0080] As discussed above, examples of the present disclosure are not limited to a window interval of five bits, but can be smaller or larger depending on the sampling rate of the test and measurement instrument. The edges in the window may jitter due to signal clock jitter, or may be delayed by a random amount, so that their combination in this one window instance may be similar to the average value obtained from summation block 508. The best observation of jitter can be obtained by making the FFT window smaller and including fewer UIs and fewer edges. One edge per window is ideal, but the sampling rate may not be high enough to optimally support such a configuration.

[0081] exist Figure 3-7 In each of the test and measurement instruments illustrated in FIG, different components are shown in different arrangements. As will be appreciated by those skilled in the art, the components may be arranged in alternative ways, such as a threshold gate may or may not be provided for each bi-coherence spectrum output, a derivative may or may not be performed, a resampler may be present in some examples and not in other examples, a bispectrum may be used instead of a bi-coherence spectrum, higher order statistics may be incorporated, a machine learning algorithm may be used instead of a neural network, etc.

[0082] As will be readily appreciated by those skilled in the art, each of the test and measurement instruments discussed above may include additional components not shown in the embodiments. For example, each of the test and measurement instruments may include one or more processors, additional hardware or firmware, one or more memory components, a user interface, a display, and the like.

[0083] Aspects of the present disclosure may operate on specially created hardware, firmware, digital signal processors, or on a specially programmed computer including a processor operating according to programmed instructions. The terms controller or processor, as used herein, are intended to include microprocessors, microcomputers, application-specific integrated circuits, and dedicated hardware controllers. One or more aspects of the present disclosure may be embodied in computer-usable data and computer-executable instructions, such as in 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 types. Computer-executable instructions may be stored on computer-readable storage media such as hard disks, optical disks, removable storage media, solid-state memory, random access memory (RAM), etc. As will be appreciated by those skilled in the art, the functionality of the program modules may be combined or distributed as desired in various aspects. Furthermore, functionality may be embodied in whole or in part in firmware or hardware equivalents, such as integrated circuits and FPGAs. Specific data structures may be used to more efficiently implement one or more aspects of the present disclosure, and such data structures are contemplated within the scope of the computer-executable instructions and computer-usable data described herein.

[0084] In some cases, the disclosed aspects may be implemented in hardware, firmware, software, or any combination thereof. The disclosed aspects may also be implemented as instructions carried by or stored on one or more computer-readable storage media, which may be read and executed by one or more processors. Such instructions may be referred to as a computer program product. As discussed herein, computer-readable media refers to any medium that can be accessed by a computing device. By way of example and not limitation, computer-readable media may include computer storage media and communication media.

[0085] Computer storage media refers to any medium that can be used to store computer-readable information. By way of example, and not limitation, computer storage media may include RAM, ROM, electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technology, compact disc read-only memory (CD-ROM), digital video disc (DVD) or other optical disk storage, magnetic cassettes, magnetic tape, magnetic disk storage or other magnetic storage devices, and any other volatile or nonvolatile, removable or non-removable media implemented in any technology. Computer storage media does not include signals themselves or transient forms of signal transmission.

[0086] Communication media refers to any medium that can be used to transmit computer-readable information. By way of example, and not limitation, communication media may include coaxial cables, fiber optic cables, air, or any other medium suitable for communicating electrical, optical, radio frequency (RF), infrared, acoustic, or other types of signals.

[0087] Example

[0088] Illustrative examples of the technology disclosed herein are provided below. Implementations of the technology may include any one or more, and any combination, of the examples described below.

[0089] Example 1 is a test and measurement instrument for analyzing signals, comprising an input end configured to receive a signal; a digital-to-analog converter configured to convert the signal into a digital signal; and one or more processors configured to: determine a recovered clock signal based on the digital signal, set a window position for a fast Fourier transform of the digital signal, isolate the digital signal into a series of windowed waveform data based on the window position, transform each of the windowed waveform data into frequency domain windowed waveform data using a fast Fourier transform, and determine high-order spectral data of each frequency domain windowed waveform data; and a neural network configured to receive high-order spectral data of the frequency domain windowed transform data and classify each windowed waveform data based on the high-order spectral data.

[0090] Example 2 is the test and measurement instrument of Example 1, further comprising a threshold gate configured to receive high-order spectral data and pass high-order spectral data that does not violate the threshold, and set high-order spectral data that violates the threshold to a nominal value.

[0091] Example 3 is the test and measurement instrument of Example 2, where the nominal value is zero.

[0092] Example 4 is the test and measurement instrument of any of Examples 1-3, wherein the high-order spectral data includes at least one of amplitude values, phase values, or complex values.

[0093] An example is a test and measurement instrument of any one of Examples 1-4, further comprising a multiplexing switch configured to receive frequency domain windowed waveform data and route it to a specific path based on the received bit pattern, wherein the one or more processors include determining high-order spectral data for each of the frequency domain windowed waveform data of each path and providing a corresponding neural network for each path for classifying each windowed waveform data based on the high-order spectral data of the corresponding path; and a bit pattern recognition neural network configured to receive the frequency domain windowed waveform data and output the bit pattern.

[0094] Example 6 is a test and measurement instrument of Example 5, wherein the one or more processors are further configured to average the high-order spectral data of each of the frequency-domain windowed waveform data, and each corresponding neural network is configured to classify the windowed waveform data based on the average value of the high-order spectral data.

[0095] Example 7 is the test and measurement instrument of any of Examples 5 or 6, wherein the one or more processors are further configured to determine a transfer function of a channel of the test and measurement instrument based on an output bit pattern of the bit pattern recognition neural network.

[0096] Example 8 is the test and measurement instrument of any of Examples 1-7, wherein the neural network is configured to output at least one of a bit error rate, distortion, pattern decoding, jitter measurement, signal-to-noise ratio, and intersymbol interference based on classification of the windowed waveform data.

[0097] Example 9 is the test and measurement instrument of any of Examples 1-8, wherein the one or more processors are configured to isolate the digital signal into a series of windowed waveform data based on a window position using a Tukey window function.

[0098] Example 10 is a method for analyzing an input signal in a test and measurement system, comprising: receiving an input signal; determining a recovered clock signal based on the input signal; setting a window position for a fast Fourier transform of a digital signal; isolating the input signal into a series of windowed waveform data based on the window position; transforming each windowed waveform data into frequency domain windowed waveform data using a fast Fourier transform; determining high-order spectral data for each of the frequency domain windowed waveform data; and classifying each windowed waveform data based on the high-order spectral data through a neural network.

[0099] Example 11 is the method of Example 10, further comprising passing the high-order spectral data that does not violate the threshold to the neural network, and setting the high-order spectral data that violates the threshold to a nominal value.

[0100] Example 12 is the method of any of Examples 10 or 11, wherein the higher-order spectral data includes at least one of amplitude values, phase values, or complex values.

[0101] Example 13 is the method of any one of Examples 10-12, further comprising: detecting a bit pattern based on the frequency domain windowed waveform data through a bit pattern recognition neural network; and routing the frequency domain windowed waveform data to a specific neural network based on the bit pattern to classify high-order spectral data of each of the frequency domain windowed waveform data based on the bit pattern.

[0102] Example 14 is the method of Example 13, further comprising averaging the high-order spectral data of each of the frequency-domain windowed waveform data for a specific bit pattern before classifying the high-order spectral data of each of the frequency-domain windowed waveform data based on the bit pattern.

[0103] Example 15 is the method of any of Examples 10-14, wherein classifying each windowed waveform data comprises outputting at least one of a bit error rate, distortion, mode decoding, jitter measurement, signal-to-noise ratio, and intersymbol interference based on the classification of the windowed waveform data.

[0104] Example 16 is the method of any of Examples 10-15, wherein the digital signal is isolated into a series of windowed waveform data based on a window position using a Tukey window function.

[0105] Example 17 is one or more non-transitory computer-readable storage media comprising instructions, which, when executed by one or more processors of a test and measurement instrument, cause the test and measurement instrument to: determine a recovered clock signal based on an input signal; set a window position for a fast Fourier transform of a digital signal; isolate the input signal into a series of windowed waveform data based on the window position; transform each of the windowed waveform data into frequency domain windowed waveform data using a fast Fourier transform; determine high-order spectral data for each of the frequency domain windowed waveform data; and classify each of the windowed waveform data based on the high-order spectral data through a neural network.

[0106] Example 18 is one or more non-transitory computer-readable storage media of Example 17, further comprising instructions for causing the test and measurement to perform the following operations: passing high-order spectral data that does not violate the threshold to the neural network and setting the high-order spectral data that violates the threshold to a nominal value.

[0107] Example 19 is one or more non-transitory computer-readable storage media of any of Examples 17 or 18, further comprising instructions for causing test and measurement to perform the following operations: detecting a bit pattern based on the frequency domain windowed waveform data via a bit pattern recognition neural network; and routing the frequency domain windowed waveform data to a specific neural network based on the bit pattern to classify high-order spectral data of each of the frequency domain windowed waveform data based on the bit pattern.

[0108] Example 20 is one or more non-transitory computer-readable storage media of Example 19, further including instructions for causing test and measurement to perform the following operations: averaging the high-order spectral data of each of the frequency-domain windowed waveform data for a specific bit pattern before classifying the high-order spectral data of each of the frequency-domain windowed waveform data based on the bit pattern.

[0109] The previously described versions of the disclosed subject matter have many advantages that have been described or are apparent to those of ordinary skill in the art. Even so, these advantages or features are not necessarily required in all versions of the disclosed apparatus, system, or method.

[0110] In addition, the written description mentions specific features. It should be understood that the disclosure in this specification includes all possible combinations of those specific features. Where a specific feature is disclosed in the context of a particular aspect or example, that feature may also be used in the context of other aspects and examples to the greatest extent possible.

[0111] Furthermore, when this application refers to a method having two or more defined steps or operations, the defined steps or operations may be performed in any order or concurrently, unless the context excludes those possibilities.

[0112] Although specific examples of the present invention have been shown and described for purposes of illustration, it will be appreciated that various modifications can be made without departing from the spirit and scope of the invention. Therefore, the present invention is not to be restricted except as in the appended claims.

Claims

1. A test and measurement instrument for analyzing signals, comprising: an input terminal configured to receive a signal; a digital-to-analog converter configured to convert the signal into a digital signal; and One or more processors configured to: determining a recovered clock signal based on the digital signal, Set the window position for the FFT of a digital signal, Isolate the digital signal into a series of windowed waveform data based on the window position, transforming each of the windowed waveform data into frequency domain windowed waveform data using a fast Fourier transform, and determining high-order spectral data for each of the frequency-domain windowed waveform data; a neural network configured to receive high-order spectral data of the frequency-domain windowed transformed data and classify each windowed waveform data based on the high-order spectral data; a bit pattern recognition neural network configured to receive frequency-domain windowed waveform data and output a bit pattern; as well as A multiplexing switch is configured to receive frequency domain windowed waveform data and route it to a specific path based on the bit pattern, wherein the one or more processors include determining high-order spectral data of each of the frequency domain windowed waveform data of each path and providing a corresponding neural network for each path for classifying each windowed waveform data based on the high-order spectral data of the corresponding path.

2. The test and measurement instrument according to claim 1 further includes a threshold gate, which is configured to receive the high-order spectral data and pass the high-order spectral data that does not violate the threshold of the threshold gate, and set the high-order spectral data that violates the threshold to a nominal value. The test and measurement instrument of claim 2 , wherein the nominal value is zero. 4 . The test and measurement instrument of claim 1 , wherein the high-order spectral data comprises at least one of amplitude values, phase values, or complex values.

5. The test and measurement instrument of claim 1 , wherein the one or more processors are further configured to average the high-order spectral data of each of the frequency-domain windowed waveform data, and each corresponding neural network is configured to classify each windowed waveform data based on the average value of the high-order spectral data.

6. The test and measurement instrument of claim 1 , wherein the one or more processors are further configured to determine a transfer function of a channel of the test and measurement instrument based on an output bit pattern of the bit pattern recognition neural network.

7. The test and measurement instrument of claim 1 , wherein the neural network is configured to output at least one of bit error rate, distortion, pattern decoding, jitter measurement, signal-to-noise ratio, and intersymbol interference based on classification of the windowed waveform data. 8 . The test and measurement instrument of claim 1 , wherein the one or more processors are configured to isolate the digital signal into a series of windowed waveform data based on the window position using a Turkey window function.

9. A method for analyzing an input signal in a test and measurement system, comprising: receiving an input signal; determining a recovered clock signal based on the input signal; Set the window position for the Fast Fourier Transform of a digital signal; isolating the input signal into a series of windowed waveform data based on the window position; transforming each of the windowed waveform data into frequency-domain windowed waveform data using a fast Fourier transform; determining higher-order spectral data for each of the frequency-domain windowed waveform data; classifying each windowed waveform data based on the high-order spectrum data by a neural network; Detecting bit patterns based on frequency-domain windowed waveform data using a bit pattern recognition neural network; and The frequency-domain windowed waveform data are routed to a specific neural network based on the bit pattern to classify high-order spectral data of each of the frequency-domain windowed waveform data based on the bit pattern.

10. The method according to claim 9, further comprising passing the high-order spectrum data that does not violate the threshold to the neural network, and setting the high-order spectrum data that violates the threshold to a nominal value. The method of claim 9 , wherein the high-order spectral data comprises at least one of an amplitude value, a phase value, or a complex value.

12. The method according to claim 9 further includes averaging the high-order spectral data of each of the frequency-domain windowed waveform data of a specific bit pattern before classifying the high-order spectral data of each of the frequency-domain windowed waveform data based on the bit pattern.

13. The method of claim 9, wherein classifying each windowed waveform data comprises outputting at least one of a bit error rate, distortion, mode decoding, jitter measurement, signal-to-noise ratio, and intersymbol interference based on the classification of each windowed waveform data.

14. The method of claim 9, wherein isolating the digital signal into a series of windowed waveform data based on the window position comprises using a Tukey window function.

15. One or more non-transitory computer-readable storage media comprising instructions that, when executed by one or more processors of a test and measurement instrument, cause the test and measurement instrument to: determining a recovered clock signal based on the input signal; Set the window position for the Fast Fourier Transform of a digital signal; Separating the input signal into a series of windowed waveform data based on the window position; transforming each of the windowed waveform data into frequency-domain windowed waveform data using a fast Fourier transform; determining higher-order spectral data for each of the frequency-domain windowed waveform data; Classifying each windowed waveform data based on high-order spectral data using a neural network; Detecting bit patterns based on frequency-domain windowed waveform data using a bit pattern recognition neural network; and The frequency-domain windowed waveform data are routed to a specific neural network based on the bit pattern to classify high-order spectral data of each of the frequency-domain windowed waveform data based on the bit pattern.

16. The one or more non-transitory computer-readable storage media of claim 15, further comprising instructions for causing the test and measurement to perform the following operations: passing high-order spectral data that does not violate a threshold to the neural network and setting high-order spectral data that violates a threshold to a nominal value.

17. The one or more non-transitory computer-readable storage media according to claim 15 further include instructions for causing the test and measurement to perform the following operations: averaging the high-order spectral data of each of the frequency-domain windowed waveform data of a specific bit pattern before classifying the high-order spectral data of each of the frequency-domain windowed waveform data based on the bit pattern.

Images