New Keithley DMM6500 and now DAQ6510

[MegaVolt]

Tell me who knows:

1. Why the measurement period for 5 NPLC = 0.3 s, and not 0.1 s.
2. Which setting or script allows you to take measurements in DCV mode at regular intervals. How can this be done in digitize V mode using the Count setting?

[MegaVolt]

3. Measurement time for 10 NPLC have an exponential beginning?

[HighVoltage]

3. Measurement time for 10 NPLC have an exponential beginning?

Was the instrument warmed up?
Was the source stable?

[MegaVolt]

Was the instrument warmed up?
Was the source stable?

On the graph, time, not voltage.
These old ones seem to be from version 1.6.3 in the new one, I could not repeat it. Perhaps this has already been fixed. If I can guarantee it to repeat, I will definitely write it.

[HighVoltage]

My DMM6500 arrived a few days ago and it seems to be spot on in the 10V range calibration, with my 10.0000V source.

I did a warmup comparison between the DMM7510 and the new DMM6500.
They behave very differently, but after about 2 min, they are almost completely warmed up.

But interestingly, the DMM6500 is almost as fast in the warmup time as the DMM7510. Definitely much faster than all the Keysight DMMs.

So far, I am very happy with the DMM6500.
Especially, since it came with the free Kickstarter software license.

Edit: Mislabeled pictures

[MegaVolt]

They behave very differently, but after about 2 min, they are almost completely warmed up.

I watched a warming 7510 at a range of 0.1V with shorted inputs. Oscillations are established in 1.5 hours as indicated in the specification.

[HighVoltage]

Yes, I am not saying the instruments are warmed up to metrology specs after 2 min !
But it is impressive, how fast these two Keithley instruments warm up in comparison to other brands.

[MegaVolt]

Yes, I am not saying the instruments are warmed up to metrology specs after 2 min !
But it is impressive, how fast these two Keithley instruments warm up in comparison to other brands.

Yes, no doubt they are very quickly ready to work. And they also enter the specification very quickly.

I talked about the complete completion of transients for maximum accuracy.

[HighVoltage]

After a good warmup, I had both, the DMM7510 and the DMM6500 on the same 10V source in parallel.
Both instruments on 5 NPLC, 10 MOhm, AutoZero ON

Why would I get such a jump in the middle of the DMM6500 graph?

Also interesting (funny):
The DMM6500 counts the x-axis with [h:m:s] and the DMM7500 in [ks]

I will repeat this now with 10 NPLC and Average Filter ON

[MikeP]

HighVoltage
 Thanks for your experiment. Can you make this measurement with 1NPLC for both DMM's also? Can you show statistical data for all experiments? Thanks again.

[Kleinstein]

Both graphs for the 7510 and 6500 show a problem with the scaling. There is no way to tel the actual scale. I know this a rather extreme case, but the labeling could get some improvement. Maybe make sure the ticks are not fixed pixel width but something like a 1-2-5-10 sequence in real voltage / current and than somewhere also show the distance. So a little like with scopes.

The steps look a lot like popcorn noise. Likely from the 6500 internal reference, as from other points in the DMM such an error should be corrected by AZ mode.

[Brad O]

Time for another mass response!  I tried to address everything that other users haven't answered already.

First of all, there's been a new firmware release, 1.0.03.  It's not a major one though, it mostly includes emulation mode and some VISA communication fixes that were heavily requested.  One notable thing, the 2700 triggering on pin 6 of the DIO port that Jens01 asked about has been fixed so that the DAQ6510 triggers properly now.  The links have the full release notes. 
For the DAQ6510: https://www.tek.com/digital-multimeter/daq6510-software/daq6510-firmware-v1003-and-release-notes
For the DMM6500: https://www.tek.com/digital-multimeter/dmm6500-software/dmm6500-firmware-v1003-and-release-notes

Can you show the graph of the frequency response of the filter? So that I can make decisions about the need for additional filtering?

It became clear about the digitizing V mode.

I can try making a graph next if this doesn't answer your questions but I think I'd have to order some equipment to make a good one.  This answer comes from the hardware team:
The filter can be modelled as a simple low pass filter with a time constant of ~ 150ns putting the transition at around 1MHz. Investigate what that means to your signal input being digitized. There are higher poles also between 5MHz and 50Mhz but the upper limit response is dominated by the ~ 1MHz.

Suggestions for best added anti-alias additional filtering to really kill the artifacts. These match up well to the internal ADC. The more poles the better until ~ 8:
8 pole BESSEL type low pass filter set at ~ 400kHz – use this if you want to preserve signal waveform fidelity/accuracy (constant group delay, won’t show dispersive effects)
8 Pole MFB or BUTTERWORTH low pass type set at ~ 450kHz – use this for best attenuation, has pass band ripple (can be designed for < 1%), has some dispersion.

NOTE: The suggested pole locations are chosen to support the full input bandwidth of the digitizer for all signals. If that’s actually not needed and you wish to tailor it to your signal, the it is better to place these poles approximately 1 decade higher  than your highest signal frequency of interest. This will optimize it for your specific signal achieving the highest input noise attenuation. When implementing your filter, keep resistances < 20k if possible to avoid making additional high frequency noise sources.

Example1: Square wave(PWM etc..) just want to see the signal,  – need full BW set to just below 500kHz (400, 450 is as in above suggestions)  with appropriate # of poles can use any type. Consider no filter at all if you aren’t looking at low frequencies that could be aliased and you really want highest bandwidth.
Example2: Noisy sinewaves, ramps etc.. and care about amplitude accuracy < 10kHz. Set at ~ 150kHz and use Bessel up to 8 poles.

NOTE: If you design an input filter, keep in mind the DC specifications. The digitizer function is specified for DC performance and if you want to preserve that, you will need to select the right components for your filter.
The hardware was designed to not force an artificial bandwidth limitation so that it can be the most flexible for all applications. Programmable HW filtering options were considered but the team decided to leave them out to make the unit as versatile as possible.

How should I set up the device so that after filling in one buffer, it will without a break begin to fill the next buffer?

I don't think this is possible and I'm not sure why you would want to do this, wouldn't it be better to set one buffer to the maximum size than have 2 half size buffers?  Do you want all the data you're taking or are you aggregating it in some way?  If you're processing, it might be possible to use a TSP script.  If you want all the readings directly, you're limited by the bus speeds.

Brad O.
Thanks for manual, interesting read! We volt-nuts always like to know what more equipment we need to buy for calibrations 
I missed it before, because Tek site ain't too much friendly for old-school folk. Clicked DMM6500 -> Manuals -> Service = nothing, so I assumed it's not published yet.

Ah it looks like the web team assigned it as a User manual, I think that's a mistake.  I filed a bug report to have them update the classification, I fully understand your confusion.  I'm also working with the web team to improve the way tek.com handles documents like these, you're not the only one to have problems finding things.

After a good warmup, I had both, the DMM7510 and the DMM6500 on the same 10V source in parallel.
Both instruments on 5 NPLC, 10 MOhm, AutoZero ON

Why would I get such a jump in the middle of the DMM6500 graph?

That does look strange and I don't like it.  I'll try to reproduce it after the Holidays.  Was your source a custom 10V reference?  I would expect popcorn noise to have a shorter duration but it's a possibility. 

Also interesting (funny):
The DMM6500 counts the x-axis with [h:m:s] and the DMM7500 in [ks]

This was specifically done for the DAQ6510, but I think the change to [h:m:s] is slated to go in the next 7510 firmware too.  I think it makes a lot more sense than kiloseconds. 

Both graphs for the 7510 and 6500 show a problem with the scaling. There is no way to tel the actual scale. I know this a rather extreme case, but the labeling could get some improvement. Maybe make sure the ticks are not fixed pixel width but something like a 1-2-5-10 sequence in real voltage / current and than somewhere also show the distance. So a little like with scopes.

The firmware team is looking into ways to improve labeling when you have a super stable signal like that.  I don't know where they are in that process, but that's on the docket to be addressed.

[Brad O]

It's because it cannot run the trigger script a million times per second. To get a guaranteed timing there's no way around setting the size of your segments.

For the digitizing V mode, is it possible to somehow disable the trigger and possibly the screen.

Are there any other ways to get continuous data in the buffer? Particularly interested in DCV mode.

Is there somewhere a tutorial on setting up triggers and writing scripts?

If you want the minimum spacing between each reading: disable autoranging and autozero, and set a count or use a trigger model (Setting a count is the easiest). 
Here's a small walk through on how to use the trigger model to have equal spacing between your measurements, where the spacing you want isn't necessarily the minimum.  There's an image of the final model attached.

• Go to MENU > Configure under Trigger (The model should be empty, otherwise clear it or do a system reset)
• Insert a Notify block, and notify a timer, it doesn't matter which timer, but remember your choice
• "Insert After" the notify block, an Action > Measure/Digitize
• "Insert After" the Measure Block, a Wait
• Select Event 1 and change it to Timer, select the timer you chose in step 2
• Select the gear next to Timer 1 and select Delay
• Set the delay to what you want your measurement spacing to be, say 1s
• "Insert After" the Wait block, a Branch, and select Loop Counter
• Set the Target Count of the branch to be however many reading you want to take, leave it branching to Block 1.  If you want it to be infinite, choose an Always branch block instead of Loop Counter
• Hold down the TRIGGER key for 5 sec, then select "Initiate Trigger Model" to start the model.

We just created a trigger model, so what does it do?  First we notify a timer, all that really does is start the timer.  Then we measure with what ever measurement settings are currently configured.  Note that the model doesn't specify what the measurement is or even what settings to use for it, only to take one.  This means you need to set up your measurement settings before you start this trigger model.
Then we wait for the timer we started in block 1 to expire, finally we branch back to block one where we start the timer again. 

What if I set a short timer so that it expires before the measurement is done? Block 3 will know that the timer has already expired so the model will leave the block immediately. 
What if there's a problem with my trigger model? The model editor will show you where it is in the trigger model while it's running.  If you abort the model, it'll show a little red error symbol next to the block it was in when you aborted, probably meaning there's a problem with that block.

Hopefully this is helpful for you all.  All the trigger model blocks and their options are explained in detail in the Reference Manual starting on page 317/9-28.  It is very powerful, but definitely different from what you might be used to.  Once you have a trigger model you like, you can save it as an instrument setup and recall it any time (an instrument setup WILL know your measurement settings, unlike the trigger model itself).  You can then export that setup as a TSP script to use on a different instrument or edit it directly in a text editor.

[HighVoltage]

After a good warmup, I had both, the DMM7510 and the DMM6500 on the same 10V source in parallel.
Both instruments on 5 NPLC, 10 MOhm, AutoZero ON

Why would I get such a jump in the middle of the DMM6500 graph?

That does look strange and I don't like it.  I'll try to reproduce it after the Holidays.  Was your source a custom 10V reference?  I would expect popcorn noise to have a shorter duration but it's a possibility. 

Thanks for looking in to that.
The source is a very old, but stable Fluke 731B
And as you can see in the comparison to the DMM7510, the signal is stable.
This problem is repeatable, at least on my 6500 !

[Kleinstein]

...
This problem is repeatable, at least on my 6500 !

How repeatable ?  I would be very surprised if the jumps would occur at the same time again - that would be an unlikely software problem.
For popcorn noise one can expect the same size jumps (for a given meter), but at different times. Other meters could have smaller (or larger) jumps of less frequent ones.

[HighVoltage]

...
This problem is repeatable, at least on my 6500 !

How repeatable ?  I would be very surprised if the jumps would occur at the same time again - that would be an unlikely software problem.
For popcorn noise one can expect the same size jumps (for a given meter), but at different times. Other meters could have smaller (or larger) jumps of less frequent ones.

Just the jump is repeatable it happens randomly just once in a while.
Its about the same height but sometimes shorter, sometimes longer.
Sometimes the instrument can run for a long time and there are no jumps at all.

[HighVoltage]

HighVoltage
 Thanks for your experiment. Can you make this measurement with 1NPLC for both DMM's also? Can you show statistical data for all experiments? Thanks again.

Here you go.
Both instruments are parallel on a Fluke 731B
Both instruments are on 1 NPLC and no Filter

I is nice to see how well both instruments agree on the 10V value of my source.

[MegaVolt]

Hooray!!!! Brad O is back )
I already thought that I asked a lot of questions
Thank you for being with us.

I can try making a graph next if this doesn't answer your questions but I think I'd have to order some equipment to make a good one.  This answer comes from the hardware team:
The filter can be modelled as a simple low pass filter with a time constant of ~ 150ns putting the transition at around 1MHz. Investigate what that means to your signal input being digitized. There are higher poles also between 5MHz and 50Mhz but the upper limit response is dominated by the ~ 1MHz.

This answer is quite enough. No need to purchase equipment.

And this information is sad. I always have to think about filtering. And no matter what mode of operation is selected. In other words, the filtering issue is completely on the user.

To be honest, I expected that for at least part of this development is closed by engineers Keithley.
Especially for the DCV mode when the sampling frequency of the signal is small. And it is logical to expect that the appropriate filters are connected inside.

Suggestions for best added anti-alias additional filtering to really kill the artifacts.

Thank.

I don't think this is possible and I'm not sure why you would want to do this, wouldn't it be better to set one buffer to the maximum size than have 2 half size buffers?  Do you want all the data you're taking or are you aggregating it in some way?  If you're processing, it might be possible to use a TSP script.  If you want all the readings directly, you're limited by the bus speeds.

Oh yes I already understood this. All I could get from the device is about 300 kilobytes per second. This is 3% for LAN and less than 1% for USB. I expected more. And I hoped that I could receive all data remotely without losses.

If you want the minimum spacing between each reading: disable autoranging and autozero, and set a count or use a trigger model (Setting a count is the easiest).  Here's a small walk through on how to use the trigger model to have equal spacing between your measurements, where the spacing you want isn't necessarily the minimum.  There's an image of the final model attached.

Thank you so much for this information. I will definitely try this example.

I thought that the issue of jitter when digitizing is solved by hardware. And the program should not influence it.

And in order to use the full potential of the ADC, we must have a very low level of jitter.

For the 350kHz signal bandwidth and the ADC even 20bit (120 dB). Jitter should be no more than 0.5 ns.

And I thought that I would not need to conjure with software to get this accuracy. I was hoping that this was already done in FPGA.

------------------------

I sketched a block diagram of the device as I understand it. It is very simple and does not display many details.

But I still do not see on it fatal errors that would interfere with the transfer of all data without loss. And maintain a stable sampling rate of the ADC signals.

I am ready to correct it so that it more corresponds to the truth. I think it will be useful.

[MikeP]

HighVoltage
 Some time ago, I was surprised to find that the best accuracy for resistances at 1-2 PLC. It is very interesting to find the dependence of accuracy/PLC for voltage.
 Which PLC has the smallest error? 10-5 or 1 PLC at the same conditions. I think you know the true voltage of your 731.
 Thanks.

[Kleinstein]

The AD7982 is 18 Bits only - so the jitter requirement would be lower. Another point is that large signal bandwidth is usually considerably lower than the small signal BW. So the worst case 300 kHz signal is lower amplitude. In addition one could accept some extra error in these extreme cases when that is a large signal. So no need to have jitter effect below the quantization error even under worst case conditions.
If reasonably well programmed the trigger could still be accurate down to the clock jitter level and thus down to the sub ns range if really needed.

It is very hard to tell which PLC setting is more accurate: usually there should be very little difference. If at all there could be some extra error (e.g. waiting for settling and maybe extra INL) for the fast modes. The calibration should normally use more of a slower high resolution mode so that these modes would be the most accurate.

There might be a little higher or lower noise for some speeds, especially of one compare something like a 10 PLC mode with the average of 1 PLC conversions. The slow modes like 100 PLC are internally made as averaging shorter conversions and usually the choice here should be reasonable good for best performance.  Even without filtering enabled, there seem to be some filtering going on for the zero measurements of the AZ mode. This can be tricky when comparing different speed, as the readings are not 100% independent, bit slightly correlated. So noise estimates for the average value from the std. dev values can be a little on the optimistic side.

[MegaVolt]

The AD7982 is 18 Bits only - so the jitter requirement would be lower.

I talked about the ADC multislope.

The AD7982, with the Count parameter set, perfectly fills the buffer with zero jitter. It's fine.

But attempts to get the same with the exact ADC are not yet obtained.

Here is a picture with all disabled AUTO ...

A jitter of around 1µs is a pure 50 dB. It is very sad

[Kleinstein]

For the relatively slow multi-slope ADC, jitter is much less critical.  Having the second faster ADC in parallel, I would assume there is little sense to use it for much below 1 ms integration.
 
With so many options for triggering the jitter likely depends on the trigger type used.

Depending on the type of ADC, there could also be inherent jitter - especially the continuous integrating ADCs can have intrinsic uncertainties in the exact aperture. Some versions also have a non rectangular aperture, using a soft start and stop that represents some extra filtering.
Still 1 µs jitter would be pretty high, though in most cases not a problem, as the relevant frequencies are low  (e.g. <= 120 Hz).

[MegaVolt]

Here's a small walk through on how to use the trigger model to have equal spacing between your measurements, where the spacing you want isn't necessarily the minimum.

I checked this model. Unfortunately, it has the same 1 μs jitter.

[MegaVolt]

Still 1 µs jitter would be pretty high, though in most cases not a problem, as the relevant frequencies are low  (e.g. <= 120 Hz).

Yes, it seems so.

But at the same time I lost AZ

[Kleinstein]

There should also be a way to get a constant sampling rate with AZ mode active.  It may not be that fast (e.g. sampling rate near 24/29 Hz) but ideally should allow for a reasonable constant rate.