Metrology Meeting in Stuttgart/Germany

[notfaded1]

Did anyone find the missing 2ppm?  What kind of beer do you have over there btw?

Bill

[BU508A]

Did anyone find the missing 2ppm?  What kind of beer do you have over there btw?

Bill

About the 2ppm I cannot say anything.

The beer is most likely this one:

https://www.stuttgarter-hofbraeu.de/

[Dr. Frank]

Nope, we had another Stuttgart local, iconic beer, Vollbier Helles

[branadic]

Did anyone find the missing 2ppm?

At least we think we did, but I have to fully confirm this measuring the other references one after another.

Reason was, that K3458A was in parallel to K2002 influencing each other. I did the test yesterday: readings on K3458A can change, if you connect another meter in parallel to it. On the other hand, now that this issue was found I repeated measurements on first reference with K3458A and they now agree with the readings of HP3458A.

So this is the lesson learned for me at this summit, be careful paralleling meters and watch for changes in the readings.

Edit: Attached is the current result. In the area marked red are the findings with the meters paralleled, in the area marked green the readings with each meter individually connected to reference LTZ1000-1. The left point in the area marked green is the reading on HP3458A, the other points are K3458A, Prema 6048 as is (no adjustment) and both adjusted meters R6581D and S7081. They now agree within about +/-0.5ppm, with Prema 6048 readings somewhat higher.

-branadic-

[guenthert]

Did anyone find the missing 2ppm?

[..]
So this is the lesson learned for me at this summit, be careful paralleling meters and watch for changes in the readings.
[..]

  Uh, I thought it was well known that you're not supposed to do that.  The 'pump-out' current of a DMM might average to just a few pA, but spikes well into the nA range.  That's bound to influence other meters.

EDIT:
  Well, on 2nd thought, that's obviously a problem when measuring sources with a significant output resistance (iirc standard cells are about 1kOhm).  For low impedance sources I wouldn't know where the discrepancy comes from.

[dietert1]

To be honest, i can't understand those findings. If a nA current makes a 2 ppm shift on a 10 V or 7 V reference, that reference has an output resistance of  20 KOhm (20 uV / 1 nA). What are you doing there?
A well built voltage reference should have an output resistance of about 1 Ohm or less. Of course, then you have cables etc. But never in the KOhm range.

Regards, Dieter

[branadic]

@ Dieter

I don't llike the sound within your words

What are you doing there?

. Sounds like you know it all.

It is fact that we made this observation. There is an offset noticeable when K3458A and K2002 where connected in parallel to the references, either LTZ circuit by Andreas, LTZ circuit based on Frank's design but also on AD587 references from Andreas.
This offset noticeable changed with additional meters such as R6581D and S7081 in parallel.

The references are well build, they are buffered by either LTC2057 or LTC1052. The offset vanished, when only one meter was connected to the reference output and all meters read about the same value. If you have a better explanation for it please let us know.

-branadic-

[e61_phil]

Please repeat this experiment with a proper source, like a calibrator.

[dietert1]

A possible explanation could be RF ("Elektrosmog"). Maybe those references are OK and the DVMs are OK but those old DVMs were not made for an environment with WLAN and switching power supplies.

I remember that in high end audio people know they have to put 100 ohm to 1 KOhm resistors into all their audio cabling/device outputs/device inputs to avoid building large antenna systems. Maybe we have to do something similar when measuring references.

Regards, Dieter

[Kleinstein]

There is a chance the DIY reference may have a common weakness.  A low impedance voltage regulator may have the tendency to show ringing with only a capacitive load.

The AZ OP in the DIY reference could also react to load at the output. Some of the AZ OPs show an offset that changes with the input impedance (especially capacitance) - this could cause trouble if not filtered isolated enough.
Another point can be spikes at the OP output, that can get back to the feedback bath. Here the load impedance (e.g. > MHz range) can change things quite a bit. Here it may not need a full 2nd meter - just more cable would be enough.

The R6581 is known to have quite significant current spikes at the input, not just nA - more like µA possibly even close to mA. The Solartron and Keithley meters use a chopper stabilized amplifier. At least Solartron has quite some with filtering in front of the chopper part - so it should not cause that much interference.

[dietert1]

Already the most basic configuration of an external reference connected to a single DVM has the potential for RF problems. For example if the reference doesn't run from battery but has a mains supply or USB connection.
Our HP 3457A has two common mode chokes for its input terminals (front and scanner). For the mains connection there is a filter, i assume that one includes a common mode choke. Or one might use an external mains filter like in high end audio. For the GPIB connection there is nothing.
I think a precision reference should have a common mode choke on its output. There are ready made 3-phase common mode chokes, so one can include the guard, e.g. Schaffner RT8132-6-6M0.

Regards, Dieter

[Kleinstein]

Our HP 3457A has two common mode chokes for its input terminals (front and scanner). For the mains connection there is a filter, i assume that one includes a common mode choke. Or one might use an external mains filter like in high end audio. For the GPIB connection there is nothing.

Regards, Dieter

The GPIB is usually isolated on the ground based part, not directly connected to the analog side, only via some capacitance (parasitic and maybe intentional to ground ?). One could add at least a clip on ferrite to the GPIB cable.

[Andreas]

There are ready made 3-phase common mode chokes, so one can include the guard, e.g. Schaffner RT8132-6-6M0.

Hello,

if I look at the data-sheet: the self resonant frequency of these inductors is around 400 kHz.
So they are optimized for switchmode power supply noise.
At least the meters should have analog supplies.
So I think a different common mode choke (51 uH "CAN" choke) should give better results over the whole frequency range.

At least my references run all from NiMH batteries during operation.

A possible explanation could be RF ("Elektrosmog").

This is only one explanation; another could also be the current spikes from auto-zero cirquits (like in the neighbour thread https://www.eevblog.com/forum/metrology/prema-8017-(sick-)-vs-tektronix-dm5120-vs-hp-34401a/msg3094592/#msg3094592 ) or both.

But I think we will find the final reason. It simply takes some time.

Our HP 3457A has two common mode chokes for its input terminals (front and scanner).

Do you have a specification of those chokes (inductivity, core material ?)

with best regards

Andreas

[dietert1]

The choke in the HP 3457A has about 2.4 to 2.7 uH at 10 KHz, similar to a 2 m wire. The higher number is for the scanner input which has 40 cm extra wire length. The ferrite ring has about 24 mm OD, 15 mm ID and a length of 7 mm with 4 turns. One can try this one: KEMET ESD-R-25D-8
Now, that inductance is similar to what you can get with a clip-on ferrite (easier to mount). I measured two of those and got 3.4 and 4 uH. Würth has this nice set: http://www.farnell.com/datasheets/1789626.pdf

Yes there is no perfect wide band common mode filter part. For high end audio i have seen combinations of that with a wound choke like the one from Schaffner. I think 400 KHz means it works well for pulses of about 1 usec, while beads work better at the nsec time scale. A charge injection pulse from a CMOS MUX would be in the nsec time scale, if the resistance is 50 Ohms and it drives a stray capacitance of 50 pF.

Regards, Dieter

[Kleinstein]

The auto zero current spikes from the DMMs would be from JFET switching and they have some filtering inside. So this would be relatively slow, more like µs and longer.

For the 3458 it's a JFET turned of with some 1 mA maximum gate current and filtering with  some 120 pF - 5 K - 82 pF - 5 K . So the really high frequencies would be there only from parasitic effects in the filter.

[Dr. Frank]

The spikes look exactly like this. DCV 10V, zero input, 10µsec APERTURE, AZERO ON.

Scope is set to 1MOhm, 1:1 input, therefore 5nA/div. The AZ measurement happens right in the middle of both group of spikes.

It's been discussed in the book of Rado Lapuh "Sampling with 3458A", page 111, 112.

Frank

[dietert1]

If the source impedance is 5 or 10 KOhms like Kleinstein wrote, these spikes are very easy to suppress with a blocking capacitor. A well built reference should be able to drive a capacitor without ringing or oscillating.

Regards, Dieter

[TiN]

I also did see shifts when F8846A DMM connected in parallel measuring same LTZ1000A-based reference. But 8846 is very far from being serious meter anyway.

Said that, I do not see any measurable difference from connecting tandems of 3458A and 2002s together. I do often run big tandems like 3x3458 + 2x2002 + other meters for group logging on DUT DMM stability (not calibration obviously). So 2ppm shift is also unclear to me here.

[branadic]

Turned out to be a simple issue with one button in the wrong place. Guard button on K3458A was in "low" position and not in "open". Guard of the cable was connected to case of the reference and Guard jack of K3458A, but on GND on K2002. This created a Guard loop, which created the differences.
I was able to recreate this problem today. Having multiple meters in parallel can then influence the readings.

So it's not the references and no E.M.I by WiFi or anything like this.

-branadic-

[dietert1]

Assume by "GND on K2002" you meant the Lo input terminal. Looked through some manuals, but no mention of "Gnd" nor "Guard". Maybe a K2002 doesn't have a separate guard terminal?

I respect your findings, but can you explain a little how you arrived at the conclusion that no EMI was involved? Isn't the guard/shield all about EMI? How can guard connections affect a DVM reading in a lab without EMI?

Regards, Dieter

[branadic]

Exactly, K2002 doesn't have separate Guard connection.

I respect your findings, but can you explain a little how you arrived at the conclusion that no EMI was involved?

I simply recreated the setup and repeated the measurement, switching the Guard button on K3458A from Open to Low, which resulted in changes of the readings on both meter, K3458A and K2002. I can show this measurement tonight for clarification.

EDIT: Attached is the corresponding measurement.

-branadic-

[branadic]

Hi folks,

we are planing the next Metrology Meeting for september 2021. By then hopefully temperatures as well as vaccination for people at risk allows us to have a live meeting.
I will try to get the lecturer that were planed for MM2020. More infos coming soon.

-branadic-

[Dr. Frank]

that's great, branadic
I'll bring the 7000 with me, at least
Frank

[HighVoltage]

Hello branadic
Thanks for letting us know, lets hope times will improve and the meeting will happen.

[branadic]

Save the date: 11.09.2021

We are already working on a program incl. talks.

Thankfully Nikolai Beev already agreed to join us live and talk about HPM7177.

Wekomm is also in and talks about resistor standards, which was already planned for MM2020.

Propably also Mr. Zirpel from the PTB could be with us and talk about calibration, but that has to be confirmed first.

John Pickering agreed to join us at least remotely and give a talk too.

I'm sure we will also have a nice discussion about LTFLU, LTZ, 2DW23x and similar, but also some very new "lowest drift/noise voltage reference". Teaser attached.

-branadic-