Author Topic: Can 732As be tied together for an 'average"?  (Read 5647 times)

0 Members and 1 Guest are viewing this topic.

Offline acts238willyTopic starter

  • Regular Contributor
  • *
  • Posts: 95
  • Country: us
Can 732As be tied together for an 'average"?
« on: April 19, 2018, 04:11:43 am »
Dumb question... but since it's a matter of microvolts, can the
outputs be tied together for a 'average'? The 10mA would be
fighting among the several 732As. Will the internal resistance
of the outputs be enough, or will I need to put a resistor in series
with each output and tie the resistors together?
I plan on using my 'new' Keithley 181 voltmeter, which has a Gigohm
input... this should be enough to ignore the added resistance, right?
If resistance is necessary, could I use my pile of old L&N rams horn
4XXX resistors? I have eight 100 Ohm L&Ns and nine 1000 Ohms...
Need to do all the thermal and ground loop stuff, of course...
 

Offline montemcguire

  • Regular Contributor
  • *
  • Posts: 88
Re: Can 732As be tied together for an 'average"?
« Reply #1 on: April 19, 2018, 06:32:34 am »
This topic has been researched extensively, beyond just averaging references, to include the notion of calibration intervals, expected drift of each reference as obtained from calibration data, and also trying to predict reference drift. While it's not 'easy reading', and I'm not an expert at this, I do remember reading a paper from Fluke that directly addresses the topic of effectively using and calibrating multiple DC references to reduce uncertainty: https://us.flukecal.com/literature/articles-and-education/electrical-calibration/papers-articles/maintaining-10-vdc-03-ppm-o

I know that Fluke will gladly sell you a 'multi-pack' of references so that you can send some off at regular calibration intervals and have some at home as working references, so try to look for that as well. You're basically correct in the notion that multiple references can result in a better working reference volt, but I think it's worthwhile to study what's been done before rather than work out the details of some simple averaging system that doesn't seem to do everything that the experts have determined to be necessary or useful.

Again, I wish I could be much more specific, but I'm not an expert on this topic. Hopefully this 'tip of the iceberg' can direct your study to what's been done already, since it seems that a lot has been done that you can benefit from.
 
The following users thanked this post: TiN, cellularmitosis

Offline ap

  • Frequent Contributor
  • **
  • Posts: 282
  • Country: de
    • ab-precision
Re: Can 732As be tied together for an 'average"?
« Reply #2 on: April 19, 2018, 11:10:12 am »
The 732A being a precision instrument should not be loaded. You do not know what effect this will have. The proper way is to use a low EMF switch, such as e.g. the Data Proof ones and measure the various outputs with a 3458A (transfer uncertainty 0.14ppm).
Optionally, you could build chopper amp based buffers and mix these together (observe chopper noise issues and so on, may have other issues). Since the voltages will all be very close, low TC/low drift resistors will do the job, no bulky LNs needed (they may have bad TCs btw.). Of course this setup adds uncertainties that you will have to figure into the uncertainty budget, but if you do not have a DP switch, it is kind of a low cost DIY solution if done right.
Metrology and test gear and other stuff: www.ab-precision.com
 
The following users thanked this post: acts238willy

Offline acts238willyTopic starter

  • Regular Contributor
  • *
  • Posts: 95
  • Country: us
Re: Can 732As be tied together for an 'average"?
« Reply #3 on: April 21, 2018, 05:49:31 am »
Part of the reason for ganging 732s was to rotate one at a time
to buck the other group. The 6 together should be quiet enough
to see if the 7th one, which is doing the bucking, is noisy. The
7th digit on the 10V would be a microvolt and a 732B cal document
I've seen shows movement at the microvolt level. The only way
I can think of to show it's an individual 732 that's noisy is to buck
it against a group of them..........  or a JJA.
That's why I wanted to tie six 732s together and was wondering
if their outputs could do this.
 

Offline TheSteve

  • Supporter
  • ****
  • Posts: 3743
  • Country: ca
  • Living the Dream
Re: Can 732As be tied together for an 'average"?
« Reply #4 on: April 21, 2018, 06:27:48 am »
The 732A is short circuit protected and is supposed to fully recover without loss of stability so you wouldn't think its outputs are that delicate. The Datron 4910 uses 200 ohm resistors in series with each of its 4 outputs when in average mode. So if you did want to try it I would copy how Datron does it with the series resistors.
VE7FM
 
The following users thanked this post: acts238willy

Online Kleinstein

  • Super Contributor
  • ***
  • Posts: 14076
  • Country: de
Re: Can 732As be tied together for an 'average"?
« Reply #5 on: April 21, 2018, 08:51:30 am »
Combining with resistors should be OK. There is no need to very low TC resistors, as the voltages should be relatively close. This also makes the current flowing there small, even of just 100 Ohms are used. The main source of error would be likely possible thermal EMF. So keeping the resistors shielded from thermal gradients might help more than low TC resistors.
 
The following users thanked this post: acts238willy

Online RandallMcRee

  • Frequent Contributor
  • **
  • Posts: 540
  • Country: us
Re: Can 732As be tied together for an 'average"?
« Reply #6 on: April 21, 2018, 03:08:50 pm »

My experience has been that the standard deviation over 100 to 200 readings of a voltage source is well correlated with noise. I have four different references, one is five LT1027's heated and averaged using 100 ohms resistors, three are LTZ1000 sources. The five 1027s show a standard deviation in the 800nV range whereas the LTZ1000 and (also a 731B) show deviation in the 1uV range. I also averaged eight LT6655's and confirmed a similar trend of lower SD with more refs averaged.

Its important to minimize temperature shifts during measurement, of course.

I use a Keithley 2015, nothing fancy.
 
The following users thanked this post: acts238willy, zhtoor

Offline Andreas

  • Super Contributor
  • ***
  • Posts: 3222
  • Country: de
Re: Can 732As be tied together for an 'average"?
« Reply #7 on: April 21, 2018, 04:26:46 pm »
Hello,

I do not understand the correlation.

A LTZ1000 has 1.2uVpp = 200nVeff = 200nV std deviation low frequency noise
(the higher frequencies are filtered by the integration time)

If you measure 1000nV then 980nV are from your instrument (K2015) since they add as a square law.
So how can you see a correlation?

The LT1027 has about 3uVpp so 5 in average should give around 3/sqrt(5) = 1.3 uVpp which is in the same ball park than a LTZ.

with best regards

Andreas
 
The following users thanked this post: RandallMcRee, acts238willy

Offline MisterDiodes

  • Frequent Contributor
  • **
  • Posts: 457
  • Country: us
Re: Can 732As be tied together for an 'average"?
« Reply #8 on: April 21, 2018, 04:46:05 pm »
Combining with resistors should be OK. There is no need to very low TC resistors, as the voltages should be relatively close. This also makes the current flowing there small, even of just 100 Ohms are used. The main source of error would be likely possible thermal EMF. So keeping the resistors shielded from thermal gradients might help more than low TC resistors.

Not really the best idea with 732's, and this is NOT the best recommended method by Fluke to get most accurate results.

These units are pretty tough, but that doesn't mean you'll still get good results if you use them incorrectly.

Acts238 - You want to take a look at the 732a/b manual and schematic and learn what to avoid.

A) The 732 10V Hi output is source only thru the current limiter.  You start paralleling these and your average 10V high will just be sitting at the highest voltage of the group (balance resistors or not) - no real averaging going on.  Once that 732 10V output pin is driven at all higher than it's Vref output amp reference, the output loop basically tends to just turn off.  It takes a few seconds for that loop to restabilize once the overdrive is removed; an effect usually mistaken for a noisy Vref.  Head's Up: A corroded contact on 10V Hi stud on the feedback line can also cause this effect.

B)  Learn why the 10V Lo pin is part of the 10V current error compensater circuit - That's on its own eco system.  Whatever current 10V Hi is sourcing should be about the same returning on 10V Lo.  Notice that 10V Lo is not on the same circuit node as the other LO terminal on the front of your unit..  Both Lo returns are on same node, but 10V Lo should always be the return from 10V Hi.

You can try the cheesy way and run load resistors between 10V Hi and 10V Lo, and then balance resistors between your Hi Common and Low Common "average" rails but that adds noise and is not most accurate.

The preferred method is to make use of your low thermal switch array and cycle through your Vrefs, then get your average Vref value in data.  You can also use low noise isolator amps to prevent the 732's outputs from seeing each other directly.

If you've got multiple 732's on hand then you should have multiple Low Thermal Switch banks as well - use them wisely.  Your meter/s should see only one 732 Vref at a time - make sure you aren't accidentally paralleling Vrefs in your switches via sneak paths, etc.

If you're looking for noise spec, the best way to do that is to use a decent LNA filtered to .1 to 10Hz.  For lowest noise your BPF filter will have a wet tantalum cap on the input to remove DC (The electrolytic cap input filter designs here on EEVblog tend to add a bit more noise, which might be important to you) and then look at that filtered noise data on your 'scope, low freq spectrum analyzer, fast DAC system, strip chart recorder, etc.

That's how we do it for best results.





 



« Last Edit: April 21, 2018, 05:45:03 pm by MisterDiodes »
 
The following users thanked this post: ManateeMafia, Edwin G. Pettis, acts238willy, MegaVolt

Online RandallMcRee

  • Frequent Contributor
  • **
  • Posts: 540
  • Country: us
Re: Can 732As be tied together for an 'average"?
« Reply #9 on: April 21, 2018, 06:00:56 pm »
Quote
The LT1027 has about 3uVpp so 5 in average should give around 3/sqrt(5) = 1.3 uVpp which is in the same ball park than a LTZ.

Thanks--lets forget about the comparison between the 1027 versus LTZ1000, for now. You are correct that we cannot be directly measuring the noise, but could there be a correlation?

When you successively measure one, two, three...five references (using averaging via 100 ohms say) and the standard deviation is lower for each, do you think that is significant? I thought the logical conclusion would be that noise is being reduced.

No? Yes?

P.S. I certainly agree that the proper way to do this is using an LNA designed for this...I just got one from Pipelie, so hopefully I can soon see for myself what is what.
 
The following users thanked this post: acts238willy

Offline Andreas

  • Super Contributor
  • ***
  • Posts: 3222
  • Country: de
Re: Can 732As be tied together for an 'average"?
« Reply #10 on: April 21, 2018, 06:34:08 pm »
but could there be a correlation?

When you successively measure one, two, three...five references (using averaging via 100 ohms say) and the standard deviation is lower for each, do you think that is significant? I thought the logical conclusion would be that noise is being reduced.

No? Yes?
Hello,

to see a real correlation the noise of the instrument has to be at least a factor 3 lower.

Did you also disconnect one after the other of the five references to see if the correlation is only over time or does directly correlate to the number?
For one LT1027 I can imagine that you can see a larger noise than with the LTZ alone. But with 3-5 in parallel it will undergo into statistics.
(because noise level is around equal between LT1027 and the internal LM399 ref, so they may add with sqrt(2))

By the way: what I have recently detected on my K2000 with long term measurements: I have up to around 1 ppm popcorn noise on a 7V reference (LTZ#4).
(which I do not see on a parallel connected 34401A in ratio mode LTZ#4 against LTZ#6).
Never mentioned this before. So maybe it is a environment temperature dependand effect.
So standard deviation can be different depending on when you measure something.

with best regards

Andreas

 
The following users thanked this post: acts238willy

Offline TiN

  • Super Contributor
  • ***
  • Posts: 4543
  • Country: ua
    • xDevs.com
Re: Can 732As be tied together for an 'average"?
« Reply #11 on: April 22, 2018, 04:11:09 am »
How about switching + nullmeter. Even very best DVMs have worse stability/tempco spec than 732, so one could use 732 with null measurement, to get uV-delta between the units and then calculate output average in post-processing as result. This should give more useful resolution, right?  :=\
YouTube | Metrology IRC Chat room | Let's share T&M documentation? Upload! No upload limits for firmwares, photos, files.
 
The following users thanked this post: acts238willy

Offline MisterDiodes

  • Frequent Contributor
  • **
  • Posts: 457
  • Country: us
Re: Can 732As be tied together for an 'average"?
« Reply #12 on: April 23, 2018, 05:27:28 pm »
TiN - The Nullmeter method still has a pitfall and requires another piece of equipment unless the 732's are within uV of each other - To properly use a null meter you need a KVD or similar sitting the higher voltage 732, because a Null Meter's strength and max accuracy if when it's reading a NULL, or very close to null.  That's the point when the null meter has least effect on the circuit in terms of injected current flow.  Even a uA flowing INTO a 732 10V HI terminal is going to throw its feedback loop off, so you want to always watch out for that.

So you could carefully divide down the higher voltage 732 on your 720a, but remember you're not quite as accurate at those dial settings very close to "1" on the KVD... so the preferred method for lower uncertainty on 732 average values is the switch scanner method - as long as you don't take too long cycling between meter readings on your 732's you should get to lowest uncertainty for the average value between any group of 732's.  It also helps if you've got several 3458a's to keep measuring confidence high.
 
The following users thanked this post: acts238willy

Offline TiN

  • Super Contributor
  • ***
  • Posts: 4543
  • Country: ua
    • xDevs.com
Re: Can 732As be tied together for an 'average"?
« Reply #13 on: April 23, 2018, 05:40:47 pm »
Just to be clear, I'm not the one asking a question. Null method just shifts requirements from having very good meter, or from very good scanner with proper low-thermal connections to more convoluted setup with very good reference instead :).
YouTube | Metrology IRC Chat room | Let's share T&M documentation? Upload! No upload limits for firmwares, photos, files.
 
The following users thanked this post: MisterDiodes, acts238willy

Offline MisterDiodes

  • Frequent Contributor
  • **
  • Posts: 457
  • Country: us
Re: Can 732As be tied together for an 'average"?
« Reply #14 on: April 23, 2018, 07:03:07 pm »
...Oh Yes!  As long as you watch that reverse current flow into the 732 it can be done with more than one approach.  Fluke recommends low thermal switch scan method - if possible, for lowest uncertainty.  Null meter methods can work too as long as you keep that meter close to null for best accuracy.  If you had literally no better way you could strap load and balance resistors onto your group of 732's but that can increase noise.  Sounds like OP has several 732's, so a low thermal switch scanner approach seems like it would be in the available asset inventory...but whatever works.
 
The following users thanked this post: acts238willy

Offline lars

  • Regular Contributor
  • *
  • Posts: 132
  • Country: se
Re: Can 732As be tied together for an 'average"?
« Reply #15 on: April 24, 2018, 05:36:06 pm »
I have to ask if only a few uA reverse current into the 10V output can upset the loop of a Fluke 732A? As far as I understand the two outputs 1V and 1.018V is made from two resistor strings that each takes about 1000uA from the 10V output. If the designer at Fluke didn't do anything stupid I would guess the 10V output should be able to handle some reverse current?

Lars
 
The following users thanked this post: Andreas, MisterDiodes, acts238willy, MegaVolt

Offline MisterDiodes

  • Frequent Contributor
  • **
  • Posts: 457
  • Country: us
Re: Can 732As be tied together for an 'average"?
« Reply #16 on: April 24, 2018, 09:50:29 pm »
Lars:  The 732a/b is 10V out is spec'd for 0 to 12mA out, sourcing only.  In actual use you probably want to stay under 5~8mA max sourcing current if you're chasing ppm.  Any reverse current should be kept very low, otherwise it's going to start letting that 10V Hi sense line change a little higher even with the secondary output divider load (the feedback amp is not quite perfect) - so in essence yes you're right there IS some wiggle room for reverse current - we like to stay under a few uA - which might be a little extreme because we're trying to preserve absolute value as much as possible.  I wouldn't go over a few tens or 100 uA reverse current; you'll begin to see effects on your DMM.   In any case when using a  pair of these in practice you want to try to keep reverse current low - or ADD a load resistor between 10V Hi and 10V Lo to create a bit more sourcing current flow to handle your setup- if doing so doesn't increase noise too much.

Again - you aren't going to hurt the 732's when they are paralleled via balance resistors, just be on the lookout for reverse current situation and make sure that's not messing with your average.  That's one of the reasons why Fluke recommends the low thermal scanner switch method (if available).

The 732a manual does show the 732's connected back to back via null meter, but you're looking for a NULL when -adjusting- the 732's - and at that point you should be very close to zero reverse current when they are aligned to each other, even without external load resistors.  That's OK.  I think the OP was going for was creating an AVERAGE value across several 732's - and in that case I'd be careful to make sure not too much reverse current on any one 732 that could effect the result - too much reverse current on one 732 would mean that it's not really contributing to the average value as it should.

It should be noted that the other 1.018 and 1V outputs on that 732a divider string on NOT buffered, they are just secondary references.  Since those are just resistive dividers you use those in a way that creates NO current flow, or at least keep it under 20pA (about the current on a 3458a input).
« Last Edit: April 24, 2018, 10:00:57 pm by MisterDiodes »
 
The following users thanked this post: acts238willy

Offline acts238willyTopic starter

  • Regular Contributor
  • *
  • Posts: 95
  • Country: us
Re: Can 732As be tied together for an 'average"?
« Reply #17 on: April 25, 2018, 05:42:09 am »
Info... I have a 705 Keithley scanner and a 7168 nanovolt scanner card.
Wish I had a 'low thermal' instead of a 10 Ohm JFET switch group..
 

Offline acts238willyTopic starter

  • Regular Contributor
  • *
  • Posts: 95
  • Country: us
Re: Can 732As be tied together for an 'average"?
« Reply #18 on: April 25, 2018, 06:05:41 am »
Mr Diodes... Thanks for all of your input! The uV spread on the 732s from low to high is:
732A...   -56uV
732B...    -2uV
732A...  +27uV
732A...  +89uV
732A...  +96uV
732A... +122uV
732A... +185uV
The +89uV was Fluke cal'd at +94uV 8 years ago. None of these 732As have ever been
'adjusted' since I got them. The 732B has had its clickers reset a few times to get closer
to 10V.  All of these are within a few ppm of when I got them and began to watch them
with some 3458s I was repairing, 5 years ago.
 

Offline acts238willyTopic starter

  • Regular Contributor
  • *
  • Posts: 95
  • Country: us
Re: Can 732As be tied together for an 'average"?
« Reply #19 on: April 25, 2018, 05:33:08 pm »
I see now what the problem is...
The 732 output can be easily pulled up, so an 'average' isn't possible
by paralleling the outputs, even through a resistor. The 'high' 732 would
just pull all the others up and ruin the attempt to get an average.

Would this work...? (sorry about my graphics)

Take the 'lowest' output voltage 732 and get its current up to say, 1mA,
with a load resistor.
Then the 'higher' 732 couldn't pull the 'lower' 732's output up. since its
higher voltage would be going through the Keithley 181, which has
a Gig of resistance..........??????????????

Then the 181 would read the true difference - and I could use the 7168
scanner to run all of the 'higher' 732s, bucked by the lowest 732.
 

Offline MisterDiodes

  • Frequent Contributor
  • **
  • Posts: 457
  • Country: us
Re: Can 732As be tied together for an 'average"?
« Reply #20 on: April 25, 2018, 06:30:43 pm »
Hmmm...Yes that might work.  If you were null-aligning aligning the 732's to each other you wouldn't need the load resistor - but you have to be very careful trying to measure the low unit against the higher unit.  The Keithley 181 could work here - I haven't tried that personally.

Remember that on a 732 you have an equivalent output impedance sourcing of around 1mOhm or less.  So if you have a difference of even a few uVolts between two 732s you start talking about mA current flow  (or trying to) if you just direct-connected a pair of 732's without any type of balance resistor setup, or with at least some sort of meter in between.  Watch current flow on both 10V Hi and 10V Lo terminals. And thermal EMF's.  Use Guards and Shields correctly.  Let each 732 operate at its own defined voltage level without being influenced too much by a bank-mate.

Do you have a KVD?  In your situation that is a good method:  You divide down the higher-voltage 732 and null against the lower unit. As described in the manual.  We use Keithley 155's and 720a KVD if we're after low error direct comparison; or just use a 3458a to measure each 732 on its own.  Whichever works better for your situation.

Do you have a 3458a or similar available?  What I would try is just one-by-one direct read all of your references on the 3458a, calc the average, and use that as your offset on your 3458a readings.  Now you could go back thru and monitor your references compared to that average value "offset".  That might show if you have a drifty unit.  Or think about other techniques like that.

Handy Tip:  If you accidentally short out or overdrive a 732 while you're screwing around with comparisons (it happens) makes sure to wait at least the spec'd 2 minutes delay time after you remove the short to let the feedback loop and current limiter re-stabilize.  Sometimes you'll need to wait 3 or 5 minutes.  Just a head's up.  Your 732 will recover, but sometimes you have to be patient.  Sometimes they pop back right away, it just depends on what stupid thing you did to 10V terminals.  We've all been there.

Acts:  Your 732's are all over the map for value - which is OK if you're maintaining calibrated drift records on each.  Not sure if that's your goal - but normally you want the spread across your bank under 10ppm or so.  NORMALLY if you're maintaining a cal'd lab you'd want to send your 732b in to Fluke for cal (since its easier to ship), and then when you get it back you could align your other 732's to your cal'd unit.  Then don't ever touch their adjusters while you build drift records on each unit.  Keep at least one 732 in cal (your best lowest noise unit or probably the B but that depends on your shipping distance)   At that point the 732 bank is considered your standard reference that you set your Master DMM to, not the other way around.  You could also work with Fluke to be part of the Direct Volt Maintenance Program where they periodically send YOU a reference to keep your bank of 732's compared and in a known absolute value state - and that is usually the best way if you're after lowest uncertainty volt measures. 

Good luck!


 
The following users thanked this post: acts238willy

Offline acts238willyTopic starter

  • Regular Contributor
  • *
  • Posts: 95
  • Country: us
Re: Can 732As be tied together for an 'average"?
« Reply #21 on: April 25, 2018, 08:01:23 pm »
Randall, I have a few 3458s and a program called Labview Home. It allows
for any number of readouts and then will do statistics on them.
Looks like the noise.... and the spread on hundreds of readouts...
is about the same, you say?
If so, I'll read each 732 out and confirm any 'noisy' ones that way, too. 
 

Offline acts238willyTopic starter

  • Regular Contributor
  • *
  • Posts: 95
  • Country: us
Re: Can 732As be tied together for an 'average"?
« Reply #22 on: April 25, 2018, 08:09:33 pm »
Mr. Diodes, There are a lot of warnings I've read about messing
with the 732A  'adjustment' resistors.

I do have good history on the drifts and the 732As and B are solid.

Would it be worth the risk to trim the 732As closer to 10V??

The 732B is no problem with its click adjustment.
 

Offline MisterDiodes

  • Frequent Contributor
  • **
  • Posts: 457
  • Country: us
Re: Can 732As be tied together for an 'average"?
« Reply #23 on: April 25, 2018, 09:38:28 pm »
I wouldn't touch any adjustment on any 732 if you're building drift records - just leave them be.  But usually when you doing that you're getting at least one or more 732 cal'd per year.  Otherwise you don't know what the yearly drift rate really is to low uncertainty.  Measuring on a DMM usually isn't low enough uncertainty - but all of that is up to your requirements.

If you're doing a 732 7 or 10 yr maintenance and have the unit torn apart for new caps & batts, then we might choose that time to re-align a 732 if it's drifted pretty far out and we know the annual drift rate is stable and predictable.  But you want to wait at least a few days if not a few weeks after adjustment before an adjusted 732 is put back into service.

If you send a unit back to Fluke they won't touch the adjustment unless its out by more than 15~20ppm, and then they will contact you to find out if you want the adjustment or not.  They usually won't ever touch the 1V / 1.018V outputs unless you pay extra...but we really never use those anyway.

Fluke will NOT touch anything on a 732a beyond the adjustment pot so you need to make sure the unit is within range on the coarse adjust board (with 10V adjust pot in center position) before sending it in for cal - if you want it aligned to their definition of 10.000000VDC.  Which is THE definition since they are aligned to a JVS, just like NIST.  It will probably be closer than your 3458a cal, unless you send yours to Loveland.

 
The following users thanked this post: acts238willy

Offline acts238willyTopic starter

  • Regular Contributor
  • *
  • Posts: 95
  • Country: us
Re: Can 732As be tied together for an 'average"?
« Reply #24 on: April 26, 2018, 05:12:21 am »
My fleabay junkyard 732B from years ago is the traveling standard.
I hand carry it to the cal lab after their 732B comes back from cal.
So, I have a poor man's 734A/B
A732B and six 732As.
 


Share me

Digg  Facebook  SlashDot  Delicious  Technorati  Twitter  Google  Yahoo
Smf