Trend Plot on Fluke 884X meters

[robrenz]

Someone was surprised to see a graphic display on my bench meter of long term drift on this post https://www.eevblog.com/forum/product-reviews-photos-and-discussion/looking-for-goodinexpensive-bench-power-supply/msg114215/#msg114215 so I thought I would start this thread to show how trend plot works on the 8846A bench meter.   Trend plot sampling rate follows whatever reading rate the meter is currently set for.  Trend plot has two controls, start and stop.  These screen shots are from a PD 2005A power supply in CC mode at approx. 200mA directly connected to the meter 400ma input.  Meter is set at 6.5 digit mode at 100 NPLC (integrates over 100 power line cycles so in US 60Hz is about 1.7 seconds per reading)

Cold meter and power supply both turned on at the same time so we are going to plot the warm up of the meter and PS. Trend plot started and it has taken roughly 6 Min's to fill the horizontal time axis. The vertical axis continuously autoscales to accommodate the min and max reading span. In this case 11.99 mA.


When the plot reaches the right border of the display it is autoscaled horizontally to mid screen and continues. The plotting rate is now cut in half and it continues. This pic is when the plot is almost to the right border again after roughly 11.5 Min's and the plotting rate is now cut in half again and it continues and the vertical axis has autoscaled to accommodate the 13.45 mA span.


Again When the plot reaches the right border of the display it is autoscaled horizontally to mid screen and continues. The plotting rate is now cut in half again and it continues. This pic is when the plot is almost to the right border again after roughly 25 Min's and the vertical axis has autoscaled to accommodate the 14.93 mA vertical span.


This is right after the above plot reaches the right border and it autoscaled horizontally to mid screen and vertically to accommodate the additional 0.06 mA. The plotting rate is now cut in half again and it continues.


Roughly 25 Min's more and the vertical axis has continuously autoscaled to accommodate the 16.31 mA vertical span.


This is right after the above plot reaches the right border and it autoscaled horizontally to mid screen. The vertical did not need to change. The plotting rate is now cut in half again and it continues.


20 more minutes has elapsed and I can now graphically see that things have basically settled out, but now I don't have  the resolution to see how stable it is.  Next post will fix that problem.

[M. András]

is it realtime when you recording the trend with the meter?

[robrenz]

is it realtime when you recording the trend with the meter?

Yes.

[chrome]

The only thing that they should have added is the current value.

[alm]

I believe the Fluke 289 also implemented this feature. Didn't that predate the 8845A/6A? I'm not sure if the implementation is the same.

Does this work the same as the data logging feature: take one sample every x seconds/minutes, and take an extra sample each time the value changes by more than a few percent? In that case it should run out of memory at some point. Or does it employ some sort of consolidation function to consolidate old data? Can it export this data? Can you zoom in and go back in time? Or is it only for quick and dirty plotting? I can imagine wanting to do further analysis after seeing the results.

[robrenz]

Does this work the same as the data logging feature: take one sample every x seconds/minutes, and take an extra sample each time the value changes by more than a few percent? In that case it should run out of memory at some point. Or does it employ some sort of consolidation function to consolidate old data?

Consolidation, they combine adjacent recording intervals into a new min and max. You still get to see the worst-case measurements and the overall trend..

Can it export this data?

You can't export the data as captured by Trend plot. But if you want raw data you can just set the meter to record data at fixed intervals to the memory stick instead.

Can you zoom in and go back in time? Or is it only for quick and dirty plotting?

You cannot zoom pan etc. Quick and dirty only. But you do know the horizontal time scale and vertical scale at all times so you can interpolate fairly accurately what you are looking at.

I can imagine wanting to do further analysis after seeing the results.

You would need to record to the usb drive to do that.

Details of Trend plot here https://www.eevblog.com/forum/product-reviews-photos-and-discussion/looking-for-goodinexpensive-bench-power-supply/?action=dlattach;attach=24730
After having it I would not buy a bench meter without it.

Details of stats mode here https://www.eevblog.com/forum/product-reviews-photos-and-discussion/looking-for-goodinexpensive-bench-power-supply/?action=dlattach;attach=24729

One of the amazing things about this meter is the extreme internal resolution when measuring very small DCV deviations at 100NPLC.  When doing long term stability checks with a shorting bar the resolution in stat mode is 0.2nV  BTW the DCV stability over 152 hours with a 2.6 degC ambient variation is a total min max span of 1.08uV.

[alm]

One of the amazing things about this meter is the extreme internal resolution when measuring very small DCV deviations at 100NPLC.  When doing long term stability checks with a shorting bar the resolution in stat mode is 0.2nV

Stats resolution does not equate ADC resolution. Stats resolution is cheap, especially with today's CPUs. It would not be very hard to incorporate arbitrary precision arithmetic. I think I've seen averages and variances with resolutions down to the (squared) nV level even on the old HP 3456A built with early eighties technology with a few kB of RAM. I also noted that some meters (eg. Keithley meters) return more digits of resolution than specced. When shorted, it might return +2.15488842E-04, which implies a resolution of 1 pV. I'm not sure whether these are actually extra bits from the ADC or just rounding artifacts. Effective resolution (actual observed difference between readings) appears to be at least < 100 pV.

BTW the DCV stability over 152 hours with a 2.6 degC ambient variation is a total min max span of 1.08uV.

That's quite impressive for a 6.5 digit meter. But this was probably at a fairly high NPLC? My usual strategy is to set NPLC to the lowest number giving me the max digits of resolution, and then take as many readings as possible and average them. Advantage is that you get more resolution this way, and can apply filters to correct for differences in transient response. Of course the noise as observed from the range function is much higher that way.

[robrenz]

Continuing from the first post.  An hour and ten minutes has passed and I can see that things have basically stabilized. But now I want to see the stability in more detail but there is not enough resolution to see any detail.


So I hit the stop soft key to stop trend plot and hit the start soft key (same button)to restart it. Now I have a vertical span ov 180uV and can see much more detail.


Horizontal compressed after hitting right border


Plot continues to full screen


Horizontal compressed after hitting right border


Plot continues to full screen

[alm]

This is where some zoom/pan features would be desirable in my opinion. It appears that it reached a plateau approximately in the middle between the second and third horizontal scale marker. Now I might want to know the average and standard deviation of the voltage from that point on, or view the fluctuations. The current method requires you to record another half hour or so of data, or ignore the trend plot feature all together and use the traditional logging mode.

[robrenz]

One of the amazing things about this meter is the extreme internal resolution when measuring very small DCV deviations at 100NPLC.  When doing long term stability checks with a shorting bar the resolution in stat mode is 0.2nV

Stats resolution does not equate ADC resolution. Stats resolution is cheap, especially with today's CPUs. It would not be very hard to incorporate arbitrary precision arithmetic. I think I've seen averages and variances with resolutions down to the (squared) nV level even on the old HP 3456A built with early eighties technology with a few kB of RAM. I also noted that some meters (eg. Keithley meters) return more digits of resolution than specced. When shorted, it might return +2.15488842E-04, which implies a resolution of 1 pV. I'm not sure whether these are actually extra bits from the ADC or just rounding artifacts. Effective resolution (actual observed difference between readings) appears to be at least < 100 pV.

I understand that computed values like SD and average can be computed to many more digits than the ADC's resolution. But this meter is displaying (in stat mode only) non computed (min max) values to .2nV resolution when using 100NPLC and digital filtering and very close to zero. I did call Fluke tech about the resolution and they said it is real and is always maximized to the highest resolution possible internally (dependent on your settings and input levels).

BTW the DCV stability over 152 hours with a 2.6 degC ambient variation is a total min max span of 1.08uV.

That's quite impressive for a 6.5 digit meter. But this was probably at a fairly high NPLC? My usual strategy is to set NPLC to the lowest number giving me the max digits of resolution, and then take as many readings as possible and average them. Advantage is that you get more resolution this way, and can apply filters to correct for differences in transient response. Of course the noise as observed from the range function is much higher that way.

Integration time of 100NPLC and digital filtering on.  Doesn't the longer ADC integration time give  a higher effective resolution anyway?  Would averaging 10 readings at 10NPLC be better or worse than just using A single 100NPLC reading?

[robrenz]

This is where some zoom/pan features would be desirable in my opinion. It appears that it reached a plateau approximately in the middle between the second and third horizontal scale marker. Now I might want to know the average and standard deviation of the voltage from that point on, or view the fluctuations. The current method requires you to record another half hour or so of data, or ignore the trend plot feature all together and use the traditional logging mode.

Here is an example of exactly what you are talking about. Trend plot to see when stabilizaton occurs (8 hours) Then just switch to stat mode to record 8 hours of stability.  https://www.eevblog.com/forum/product-reviews-photos-and-discussion/looking-for-goodinexpensive-bench-power-supply/msg114215/#msg114215

[alm]

Integration time of 100NPLC and digital filtering on.  Doesn't the longer ADC integration time give  a higher effective resolution anyway?  Would averaging 10 readings at 10NPLC be better or worse than just using A single 100NPLC reading?

Digital filtering is another form of smoothing that will reduce the apparent noise. In theory the average of 10 readings at 10 NPLC should be the same as one reading at 100 NPLC, except that it would take slightly longer. I didn't observe any significant differences when I compared them either. My reasons for doing the averaging during data analysis was that many meters only report 6.5 digits of resolution with both 10 NPLC and 100 NPLC. Noise can be way down in the resolution limit with a stable source (eg. shorting bar), so the extra resolution gained is quite welcome.

The other major reason in my case was transient response, as I already mentioned. I was dealing with multiple instruments with different integration times and transient responses, and some not so stable sources. Taking as many samples as possible made it easier to match the transient responses.

[M. András]

I believe the Fluke 289 also implemented this feature. Didn't that predate the 8845A/6A? I'm not sure if the implementation is the same.

Does this work the same as the data logging feature: take one sample every x seconds/minutes, and take an extra sample each time the value changes by more than a few percent? In that case it should run out of memory at some point. Or does it employ some sort of consolidation function to consolidate old data? Can it export this data? Can you zoom in and go back in time? Or is it only for quick and dirty plotting? I can imagine wanting to do further analysis after seeing the results.

the 289 dont show you the trend in real time, current value, start time, duration, interval samples, events, you can see the trend after stopping it, and its damn slow to make the trend when you recorded the maximum points. while the 8846 can record to flashdrive, i dont recall from the datasheet is it infinite relative to the flashdrive capacity or limited.

[robrenz]

the 289 dont show you the trend in real time, current value, start time, duration, interval samples, events, you can see the trend after stopping it, and its damn slow to make the trend when you recorded the maximum points. while the 8846 can record to flashdrive, i dont recall from the datasheet is it infinite relative to the flashdrive capacity or limited.

Trend plot is real time on the 884X.
You can't access stored Trend plot data. You can only see it as displayed.
8846A Internal memory is 9999 readings in one file maximum.  When you add a 2GB (implied maximum) memory stick it can store an additional 999 files containing 10000 readings each.

[M. András]

the 289 dont show you the trend in real time, current value, start time, duration, interval samples, events, you can see the trend after stopping it, and its damn slow to make the trend when you recorded the maximum points. while the 8846 can record to flashdrive, i dont recall from the datasheet is it infinite relative to the flashdrive capacity or limited.

Trend plot is real time on the 884X.
You can't access stored Trend plot data. You can only see it as displayed.
8846A Internal memory is 9999 readings in one file maximum.  When you add a 2GB (implied maximum) memory stick it can store an additional 999 files containing 10000 readings each.

hmm the engineers at fluke simply love that 10k number i guess.
is there any limit for direct pc logging?

[robrenz]

is there any limit for direct pc logging?

Not that I am aware of.  Quote from programmers manual:

Several factors affect the remote system speed of the 8845A/46A. To make the fastest
measurements, it is best to store the measurements in internal memory before transferring
across the I/O bus. This is particularly true if you want to collect a large number of
samples per measurement. Set up all of the parameters (i.e. function, range, filters, etc...)
prior to taking the measurements. Use the internal trigger. Set the trigger delay to zero.
Turn off the display.

You then use the fetch command to get that reading from the meters memory.