search

jetperch / pyjoulescope_ui

Joulescope graphical user interface
https://www.joulescope.com
Apache License 2.0
75 stars 23 forks source link

Dual marker shows wrong values #213

Closed atsju closed 11 months ago

atsju commented 11 months ago

Joulescope model

JS110

UI version

other

What OS are you seeing the problem on?

Windows 10

What happened?

UI v1.0.25

I observed a discrepancy between dual marker values and window value. They should be almost the same because signal is 0 outside markers. image

I exported both "visible data" and "all data" and when I opened the files, the value for markers has changed. For me this means there is a bug in dual marker integral. exportedJLSfiles.zip

What was expected?

Correct computed values in dual marker

How to reproduce?

capture some data and put dual marker. I did not do anything special and I'm not sure this is reproductible. I will update the issue if I find something.

Extra information

No response

mliberty1 commented 11 months ago

Hi @atsju - I don't know about this particular waveform. It looks like there could be a small negative current event at about 925, but you can double-check. On the marker with text, right-click, then select Show statistics → Left. Drag the right dual marker all the way to the right side and the left dual marker all the way to the left side. Do the values now match? If so, then this is not an issue with the UI. Instead, it is a feature of the waveform.

If you do not want to adjust your measurement dual markers, you can place additional dual markers to the left and right that go to the display edge. You can sum up the integrals.

What do you find?

atsju commented 11 months ago

Please open the zip file. I shared the exact same file. And you should see that value is different from the print screen.

In the mean time I will do what you request. Lucky you, I didn't close the capture

atsju commented 11 months ago

UI bug confirmed :
image

Please note that right click on a thin maker line is not easy and not natural. I did not expect this feature there.

mliberty1 commented 11 months ago

Did you recreate this on the buffered data or the one of the JLS files that you provided?

atsju commented 11 months ago

this capture is from the same instance as first capture with same data and window not moved. Opening an export of same data does not show the bug (as you can see in the zip)

mliberty1 commented 11 months ago

So far, I have not been able to recreate what you are seeing. However, I am also using a JS220, not a JS110. I do expect small differences between the dual markers and the right-hand side statistics, like this: image

This does not explain the 10% difference in the second image above.

It sounds like this is easy data to capture. Since I only have access to the JLS file (which does not show this issue), can you check if this is a consistent issue? Capture data, then place dual markers over one event, as you did above. Repeat for a couple other events. Is the value dual marker value and the right-hand side statitics values always the same as you captured above (dual marker integral approximately +10% greater than right-hand side statistics)?

Note the these statistics values are computed differently. The dual markers are computed over the exact marker positions, which is the pixel center when repositioning the markers. In contrast, the right-hand side statistics are computed over the visible plot data, which is left-most sample in the first pixel to the right-most sample in the last pixel. To make things more complicated, the buffered sample data and JLS data compute their statistics using completely different code.

For performance reasons, this code is highly optimized which makes it complicated. For example, here is the code for the buffered sample data. The code for the JS110 and JS220 is the same, but how it is used could be slightly different.

I am traveling now, and I will try some more testing with a JS110 when I return to the office next week.

atsju commented 11 months ago

For information, I did some more tests right now with same device and it's definitely reproductible. image

mliberty1 commented 11 months ago

I successfully duplicated this issue:

bad_summary

I was able to track it down to the summary computation in joulescope_driver. These summaries are used to greatly speed the computation of statistics and waveforms. Unfortunately, when more than one summary value was computed over a single input block, the statistics accumulator was not reset correctly. See joulescope_driver 8c6714d. This error also was likely slightly distorting zoomed out waveforms for live streaming and buffered samples. It would not have affected JLS files.

This will be fixed in joulescope_driver 1.3.18 and Joulescope UI 1.0.26.

mliberty1 commented 11 months ago

You should be able to download and install 1.0.26, which fixes this issue and more. The new relative x-axis markers do not quite work correctly, so we will make a new release 1.0.27 that will hopefully make it to our next stable release.

mliberty1 commented 11 months ago

Fixed in 1.0.27