Separating noise to increase machine learning prediction accuracy in a test and measurement system

1 . A test and measurement instrument, comprising:

an input port to allow the instrument to receive one or more waveforms from a device under test (DUT);

two or more deep learning networks;

one or more low pass filters to remove a portion of the noise from the one or more waveforms to produce one or more filtered waveforms to increase accuracy of predictions from the two or more deep learning networks; and

one or more processors configured to execute code that causes the one or more processors to:

select one or more waveform patterns from the one or more waveforms;

measure noise in the one or more waveforms and generate a noise representation of the noise removed from the one or more waveforms;

create one or more images for the one or more waveforms using the one or more waveform patterns and the one or more filtered waveforms;

add the noise representation to the one or more images for the one or more waveforms to produce at least one combined image;

input the at least one combined image to the two or more deep learning networks comprised of at least one first deep learning network trained to produce predicted filter tap values, and at least one second deep learning network trained to produce one of either predicted performance measurement values or predicted operating parameters for the DUT; and

receive one or more of the predicted filter tap values, and the one of either predicted performance measurement values or predicted operating parameters for the DUT from the two or more deep learning networks.

2 . The test and measurement instrument as claimed in claim 1 , wherein the one or more low pass filters comprise a first low pass filter to produce a first filtered waveform and a second low pass filter to produce a second filtered waveform, and the code that causes the one or more processors to create at least one combined image comprises code that causes the one or more processors to:

create a levels image for the waveform using the waveform pattern and the first filtered waveform from the first low pass filter; and

create a pulse image for the waveform using the waveform pattern and the second filtered waveform from the second low pass filter.

3 . The test and measurement instrument as claimed in claim 2 , wherein the first low pass filter has a pass band lower than the second low pass filter.

4 . The test and measurement instrument as claimed in claim 1 , wherein the port also allows the instrument to receive temperature data for the DUT, and the combined image for each waveform includes a representation of the temperature data.

5 . The test and measurement instrument as claimed in claim 1 , wherein the code that causes the one or more processors to measure the noise causes the one or more processors to:

subtract the filtered waveform from the waveform to find a difference;

find a standard deviation of the difference; and

use the standard deviation to generate the noise representation.

6 . The test and measurement instrument as claimed in claim 1 , wherein the one of either predicted performance measurement values or predicted operating parameters, comprises the predicted performance measurement value comprising a predicted transmitter dispersion and eye closure quaternary value, and the predicted filter tap values comprises predicted filter tap values for feed-forward equalizer taps.

7 . The test and measurement instrument as claimed in claim 6 , wherein the code that causes the one or more processors to create one or more images comprises code that causes the one or more processors to create a levels and pulses image including noise.

8 . The test and measurement instrument as claimed in claim 1 , wherein the one of either predicted performance measurement values or predicted operating parameters comprise predicted operating parameters for the DUT.

9 . The test and measurement instrument as claimed in claim 8 , wherein the code that causes the processor to create one or more images for the one or more waveforms comprises code that causes the one or more processors to:

receive one or more waveforms from the DUT, each of the one or more waveforms resulting from the DUT being configured with a different set of reference operating parameters;

create the one or more images by creating a levels and pulse image, and a levels image for each of the one or more waveforms from the DUT; and

placing the levels and pulse image, the pulse image, and the noise representation for each of the one or more waveforms on a different color channel in the at least one combined image for the DUT.

10 . The test and measurement instrument as claimed in claim 8 , wherein the code that causes the one or more processors to input the at least one combined image to the one or more deep learning networks comprises code that causes the one or more processors to:

input a first combined image to the at least one first deep learning network, the first combined image comprising a levels and pulse image and a noise representation; and

input a second combined image to the at least one second deep learning networks, the second combined image comprising a levels image and a noise representation to a second deep learning network of the one or more deep learning networks.

11 . The test and measurement instrument as claimed in claim 1 , the one or more processors further configured to execute code that causes the one or more processors to train the at least one first deep learning network by:

acquiring values for a performance measurement using conventional testing for a plurality of waveforms from a plurality of devices under test;

creating at least one array of combined images from the at least one combined image for each of the plurality of waveforms;

creating at least one array of performance measurement values corresponding to the at least one array of combined images; and

using the at least one array of combined images and the array of performance measurement values to train the deep learning networks to produce predicted performance measurement values.

12 . The test and measurement instrument as claimed in claim 1 the one or more processors are further configured to execute code that causes the one or more processors to train at least one first deep learning network further comprises code that causes the one or more processors to create at least one array of filter tap values corresponding to the at least one array of combined images.

13 . The test and measurement instrument as claimed in claim 11 , wherein the code that causes the one or more processors to create at least one array of combined images from the at least one combined image comprises code that causes the one or more processors to:

create an array of levels and pulse images;

create an array of levels images;

create an array of levels and gains tuning parameters;

use the array of levels and pulse images, and the array of filter tap values to train the at least one first deep learning network to predict filter tap values for the DUTs; and

use the array of levels images and the levels and gains tuning parameters to the at least one second deep learning network to predict levels and gain operating parameters.

14 . A method of measuring performance of a device under test (DUT) comprising:

acquiring one or more waveforms from the DUT;

applying one or more low pass filters to the one or more waveforms to produce one or more filtered waveforms to increase accuracy of predictions from two or more deep learning networks;

generating a waveform pattern from the one or more waveforms, if not already generated;

measuring noise in at least one of the one or more waveforms and generating a noise representation;

creating one or more images using the waveform pattern and the one or more filtered waveforms;

adding the noise representation to the one or more images to produce at least one combined image;

sending the at least one combined image to two or more deep learning networks comprised of at least one first deep learning network trained to produce predicted filter tap values, and at least one second deep learning network trained to produce one of either predicted performance measurement values or predicted operating parameters for the DUT; and

receiving one or more of the predicted filter tap values, and the one of either predicted performance measurement values or predicted operating parameters for the DUT.

15 . The method as claimed in claim 14 , further comprising:

receiving temperature data for the DUT; and

including the temperature data in the combined image.

16 . The method as claimed in claim 14 , wherein measuring noise in the waveforms comprises:

subtracting one of the one or more filtered waveforms from the waveform to produce a difference;

finding a standard deviation of the difference; and

using the standard deviation to generate the noise representation.

17 . The method as claimed in claim 14 , wherein applying one or more low pass filters comprises applying a first low pass filter having a first cutoff frequency and a second low pass filter having a second cutoff frequency higher than the first cutoff frequency.

18 . The method as claimed in claim 14 , wherein creating one or more images for the waveform comprises:

creating a levels image for the waveform using one of the one or more filtered waveforms and the waveform pattern;

creating a pulse image for the waveform using another of the one or more filtered waveform and the waveform pattern; and

building one or more tensor images that include at least one of the levels image and the pulse image, and the noise representation as the combined image.

19 . The method as claimed in claim 18 , wherein building one or more tensor images comprises:

building a first tensor image from the levels image and the noise representation to be sent to the at least one first deep learning network; and

building a second tensor image from the levels image and the pulse image, and the noise representation to be sent to the at least one second deep learning network.

20 . The method as claimed in claim 14 , further comprising training the at least one second deep learning network, by:

acquiring values for the performance measurement using conventional testing for a plurality of waveforms from a plurality of devices under test;

creating at least one array of combined images from the at least one combined image for each of the plurality of waveforms;

creating an array of the performance measurement values corresponding to the at least one array of combined images; and

using the at least one array of combined images and the array of performance measurement values to train the at least one second deep learning networks to produce the predicted performance measurement values.

21 . The method as claimed in claim 20 , wherein creating at least one array of combined images comprises:

creating an array of levels and pulse images;

creating an array of levels images;

using the array of levels and pulse images to train the at least one first deep learning network; and

using the array of levels images to train the at least one second deep learning network.