Mini Metrology Lab Build (update)

[Jebnor]

I decided to build Mr. Hoffman's Mini Metrology Lab [1] as per the article.  Except for two things: I'm using a MAX6350CPA+ [2] instead of the LT1027BCH in the voltage reference and using the homebuilt rail splittler on pg 78.

I'm laying out and etching my own boards to fit in my cases, a Hammond 1455J1201 [3] for the Voltage Ref, and 1591DTBU [4] for the Null detector.  I have yet to determine what I'm going to use for the KVdivider.

I attached a picture of my parts that arrived today from Digikey.

[1] http://conradhoffman.com/mini_metro_lab.html
[2] http://datasheets.maximintegrated.com/en/ds/MAX6325-MAX6350.pdf
[3] http://www.hammondmfg.com/pdf/1455J1201.pdf
[4] http://www.hammondmfg.com/pdf/1591D.pdf

[Jebnor]

I have finished the test build of my Voltage Reference before I build a second one and use the more expensive chips.  I am testing my PCB etching process and the physical fit in the case.

I used a 1N5231B Zener diode as a voltage "Reference" in series with a 499ohm resistor across the 15V rail. Once tested, this Zener gave me 5.19V. Then I selected resistors for R4 and R5 to give me the proper feedback ratio to yield 10.0V. As you can see from the attachments, I didn't do too bad.  I used a LM741 for the test build as it has the same pinout as the LTC1050.

Before I build with 'real' parts I'm going to squeeze the IC's, R4 and R5 together horizontally. Everything else will stay the same.

Enjoy

[JP16A60]

Hi, Jebnor.

I know that this thread is pretty old, but wondering if you made any further progress.

I'm planning on building the same "Mini Metrology Lab" suite sometime soon, and would welcome any thoughts or tips on this.

JP

[Gyro]

Conrad Hoffman is actually a contributor to this forum. If he doesn't notice your post then you might want to PM him to see if he has any new tips.

[Andreas]

I'm planning on building the same "Mini Metrology Lab" suite sometime soon, and would welcome any thoughts or tips on this.

Hello,

if you want to have a 10V output I would use a AD587UQ (CERDIP-package) as reference.
Any additional resistor divider adds unnecessary temperature errors.

If you have high demands you could use a temperature compensation like on the Geller SVR-T references.

With best regards

Andreas

[Gyro]

EDIT: Probably more appropriate in the Buy / Sell / Wanted category so have moved it there...

https://www.eevblog.com/forum/buysellwanted/spotted-on-ebay-us-muirhead-10k-voltage-dividing-resistance/

[Conrad Hoffman]

That article series is long of tooth now. Learn from some of the techniques, but I'd use different chips and topology these days. The best info is right here at EEV if you read the voltage standard threads and the standard resistor threads. Whatever you do is going to rely on those two things. FWIW, the null detector circuit worked well to get people into the game with only a DVM, but I still prefer an old fashioned center-zero meter! Since then I did the quick write-up on the Hamon divider and find that to be a very handy device.

[Gyro]

Since then I did the quick write-up on the Hamon divider and find that to be a very handy device.

...and a very helpful write-up it was too! 

A short thread on mine and link to Dr Frank's here:

https://www.eevblog.com/forum/testgear/anyone-else-built-a-hamon-divider/msg776035/

Two decades is good for 'leapfrogging' up and down meter ranges to compare calibrations.

[JP16A60]

That article series is long of tooth now. Learn from some of the techniques, but I'd use different chips and topology these days.

Hi, Conrad--thanks for commenting.

I'm planning to build not because these would be the most modern instruments, but rather because it looks like it would be a fun challenge, and a way to get myself started actually building things from other than a kit or "hold-your-hand" textbook-based project.

I also find quite a bit of satisfaction in learning and building "old school".

Thanks again for your comments and for this excellent three-article series.

Regards,

JP

[Dbldutch]

I have just finished the build of the Null Detector, I wanted to use that with my "normal" 3.5 digit DMM's, but mostly I can use my bench DMM which has 1 uV resolution. This works fine for me using the original design.

However, I was mostly intrigued with the KVD, because I believe (never used one before) that it can do some things I can't do any other way. Unfortunately, for one particular task, it seems that I'm not getting the precision I want, to use the KDV to transfer a reference voltage. (it calibrates precisely, that's not the problem, but since that process goes from decade 6(output) to 1(input), it looks to me that the base it starts from is not "aligned" to get a true /10.

I have PM'd Conrad if he would be so kind to elaborate on the design constraints he made at the time, to deviate from the standard KVD decades, especially in the lower 4 ones, and what the consequences are. I suspect that is why I can't do what I want.

Here is a short description. I have a calibrated voltage reference, that at the 10 V setting actually produces 10.00673 V. My plan was/is to use the KVD to adjust the appropriate decades to lower the excess voltage to get 10.00000. So set decade 4 from 9-6 to 3, decade 5 from 9-7 to 3 and decade 6 from 10-3 to 7. You can then use a Null detector between de KVD output with the 10.00000 V and the output of the new reference and adjust that to have a null resulting in a 10.00000 V "transfer" without using 6+digit meters. I get close, but not close enough, and I suspect it has to do with the deviation Conrad used at the time from the 10K, 2K, 400, 80, 16 and 3.2 Ohm standard.

I'm planning to modify my KVD to the standard values later, but could use some more input on the applicability of my proposed method. I will provide a description of my KVD project with pictures etc. at a later date. Stay tuned.

[Alex Nikitin]

Here is a short description. I have a calibrated voltage reference, that at the 10 V setting actually produces 10.00673 V.

With such a small difference it would be easier just to use a simple divider, 10K + 8.2 Ohm and a 100 Ohm Cermet pot in parallel to 8.2 Ohm, to trim to the exact value . Even using standard metal film resistors with 100ppm/C tempco, the resulting output voltage change from this divider would be below 0.1 ppm/C, and if you use something better, like RC55 with 5ppm/C, the tempco from the resistors value drift would be essentially unmeasurable.

Cheers

Alex

[Dbldutch]

Hi Alex,
Thanks for the reply.

You are right of course, but in this case I wanted to verify if my method, using my new KVD, was correct. The underlying idea is to transfer a reference with say 6 digit accuracy to a new one, by using (only) a 3.5 digit multimeter,  a null detector and the KVD. It's for those that have no access to 4+ digit equipment. I have never used KVD's before so I'm on a fact finding mission for myself, and hopefully can share that with others.

I did get a reply from Conrad, and what I want to do is indeed sound, providing there is no load towards the reference, and towards the KVD. Because I started to doubt my thinking and the absolute precision (not the relative one if you know what I mean), I already changed the first decade. Originally, I used a matched set of 10K resistors (<0.01%), but the actual value was 9.950 Ohms. I recently added a 49.9 Ohm resistor to get to 10K and redid the matching, but since I was busy, did not try the transfer method yet. After Conrad's reinforcement, I tried it again this morning and low and behold, this time it worked. Question now is if that was due to my changes to the first decade? Both times I calibrated the decades to within 10uV using the bridge method.

More later.

[Conrad Hoffman]

From our PMs, I just wanted to mention for others that each decade has to be large enough so the adjustable shunt can bring it down to match the previous decade. I don't know if the slightly low resistors on the first decade are enough to prevent accurate calibration, but it sounds like the 10k fix did the trick.

FWIW, some KVDs used 11 steps on the first decade, giving 1.1 and 1.2 ratios. On Julie Research dividers they were the "H" models. It's trivial to add those steps when building one and it certainly increases the utility. I've got two Julies and one ESI, plus some Dekaviders (sp?) and none of them has that feature.

Off on a tangent, if you do AC, it's easy to get ppb (no, that's not a typo) ratios using ratio transformers. Henry Hall did a write-up on that, probably in one of the GR Experimenters.

[Dbldutch]

True, if the total decade chain can be adjusted by shunting, the calibration will work, as in my case. The resistors need to be relatively close, because there is an obvious end to shunting.

With some changes to the shunt combination to make the adjustments finer, I was able to get to a few micro volt decade to decade match using the bridge method.

But, that only means that only one condition of the KVD is met. I'm still not at peace over what the deviation from the 5x rule to get to 10K 2K 400 80 16 3.2 actually means or does. 

Thoughts?

[Conrad Hoffman]

Shouldn't mean a thing. Consider any one decade. As long as the total shunt value across the switch positions is correct, the divider works as it should, with the correct voltages at the switch points that will be the connection of the next decade.. How you arrive at that total shunt value doesn't matter. The calibration shunt to alter the total value of the next decade has no effect on that decade, as the voltages from the previous decade will be correct if the total shunt is correct. The "perfect" values were a target for designs that used no calibration shunts, but that's all. I'm not sure, and it would be a dumb design, but I think you could build a KVD with all the same resistor values, and shunt each decade.

[Dbldutch]

Thank you Conrad, another mystery explained.

I just looked at the Fluke 720 schematic again and I can confirm that they used 1K resistors for each of the last 4 decades.

Thank you for letting me sit on your lap. I learned a lot.

Paul

[Macbeth]

I can confirm that my esi RV622A which is a 6 digit divider uses shunts on the last decades.

1st decade is 11 x 10k
2nd decade is 11 x 2k
3rd decade is 11 x 400 ohm
4th decade is 11 x 400 ohm shunted with 1k
5th decade is 11 x 400 ohm shunted with 1k
6th decade is 10 x 400 ohm shunted with 1k

[JS]

Shouldn't mean a thing. Consider any one decade. As long as the total shunt value across the switch positions is correct, the divider works as it should, with the correct voltages at the switch points that will be the connection of the next decade.. How you arrive at that total shunt value doesn't matter. The calibration shunt to alter the total value of the next decade has no effect on that decade, as the voltages from the previous decade will be correct if the total shunt is correct. The "perfect" values were a target for designs that used no calibration shunts, but that's all. I'm not sure, and it would be a dumb design, but I think you could build a KVD with all the same resistor values, and shunt each decade.

This is done and is good for tempco tracking (having all equal resistors) but bad for output impedance. Even if it always should work with no load with a null detector, lower the output impedance minimize the error and noise. The load error being of course the output resistance times the output current, while, if both known, could be corrected.

JS

[zlymex]

The very large(and variable) output impedance of 720A is mainly due to the 1st decade. It won't lessen much if all the resistance of subsequent decades are following the 5x rule. 720A was using 1k in all lower 4 decades because of another reason(to minimize the BOM).

One easy way to access an KVD is to buy a used Fluke 887A(or 883A or 885A or similar) differential voltmeter such as http://www.ebay.com/itm/Fluke-883A-AC-DC-Differential-Voltmeter-/201294712925

[JS]

The very large(and variable) output impedance of 720A is mainly due to the 1st decade. It won't lessen much if all the resistance of subsequent decades are following the 5x rule. 720A was using 1k in all lower 4 decades because of another reason(to minimize the BOM).

One easy way to access an KVD is to buy a used Fluke 887A(or 883A or 885A or similar) differential voltmeter such as http://www.ebay.com/itm/Fluke-883A-AC-DC-Differential-Voltmeter-/201294712925

Of course the first decade already mess things up, but having all high valuer resistors would make it even worse. After you reduced enough the value (at the third decade in this case) there are diminishing returns in lowering the value even further and lower values bring their own problems. The first decade is also limited by the contact resistance in the switch, 6 decades is a lot to chew... 1ppm in the first is a whole step in the last, of course the last decade wont have the same importance/care.

JS

[Kleinstein]

There is nothing wrong with using the same resistance as in the higher decade and an extra shunt resistor. This even has 2 advantages: the resistors don't get too small to need special switches. Also the total resistance can be trimmed independent from the steps.
 
Just for the 1st, 2nd and maybe 3rd decade one often prefers a higher impedance, not to load the input so much and to have less power dissipation. Also switch resistance is less critical at high impedance. So the values at the RV622A make perfect sense.

[Dbldutch]

Interesting comments! Thank you for taking the time to contribute.

I am greener than green with this topic, but determined to learn more about the KVD and it's applications in the hands of a hobbyist, like me. There are not many tools that you can build that, potentially, let you sniff at ppm level precision with budgets that can be pretty minimal. I said potentially!

So, while playing with this tool, help me to understand what can be done to increase the value of the instrument, without having to rob a bank.

Having said that, it is now clear(er) to me that the accuracy and applicability of the KVD is greatly depending on the accuracy and the matching of the resistor Ohm values within the decades, especially the first three.
However, it seems that a lot is also depending on the TC of the resistors as well. This is a grey area to me (as well), but it is for me under much less control than the pure Ohm-ic selection. I begin to understand the troubles Fluke and the likes went through to get it right.

You can use a batch of less precision resistors to find higher precision individuals, and match them, quite easily. Obtaining a large enough batch of resistors with a basic low TC, like below say 15 PPM/C is quickly getting very expensive, let alone 5PPM/C or lower.

Is there justifiable merit in combining inexpensive resistors with a second set of resistors with an opposite TC to reduce the total? Is that even feasible for a hobbyist? While searching, I came accross this article that seems to support that idea, but I really do not have the insight to figure out if this even makes sense.
http://web.engr.oregonstate.edu/~moon/research/files/iscas07_res.pdf

Suppose you only use the KVD occasionally, and only in a room with temperatures that do not differ more than +/_ 5 degrees C over the year. And before you make a measurement, you calibrate the KVD decade to decade shunts and you only use it in a bridge configuration (no self heating by currents). Does it then even matter, or if so, how much, if the TC of the resistors are not below say 50PPM/C?

Looking forward to learn some more...

[Dbldutch]

Never heard of this technology, but I think I found them :
http://www.vishay.com/docs/49905/_aorn_vmn-pt0449.pdf

Four 10K, 2K or 1K resistors in an SMD package with .1% precision, 0.015% ratio, 25PPM/C and if I understand it, with a 5PPM/C ratio after a long burn in period. Seems made for a KVD.

Digikey sells them for $4.00 1 off.
http://www.digikey.com/product-detail/en/vishay-thin-film/AORN1002AT5/764-1149-1-ND/5138323

This could be very interesting, but now I'll need to find out how to cook them at 155 degrees C for 42 days. Glad winter is coming soon. 

Has anybody tried to use these resistors for a KVD as far as you all know?
If this is a possible option, and makes some sense with a reasonable rate of success, I'm willing to give this a try.

BTW, I'm using the Conrad Hoffman method without expensive switches at the moment.

[JS]

  Vishay does have networks with more than 4 resistors, with similar specs as the one you said. Would make more sense using one package with all you need than using 4 resistors in a package?

  Depending on the goal some trimming may be needed for the individual resistors, starting with 11x 10k resistors in series, within 0.1% of each other would allow trimming them with 100k+2k trimpot in parallel, having just enough to play with and needing more relaxed tempco as the main network. The MML from Hoffman asks for 40ppm matched resistors for the first decade, so you are not going to find that matching for 10 resistors in a single package in any cheap available product. Also, long term drift would probably take out your resistors from those 40ppm so it does makes sense having some trimming there rather than having your KVD useless (not precise anymore) after some time. Fluke 720 uses a special switch to allow the self calibration of the 2 higher decades resistors and shunts without opening the device, the third decade needs an extra screw driver to do the job. If using the pin strip instead of the switch makes the test points available, you need to have the null amp and the bridge anyway, you can take the extra step and calibrate it before critical tasks instead of relaying in the long term stability of the resistors.

JS

[Dbldutch]

JS,

I think you refer to the Vishay NOMCA series, with 7 or 8 resistors in an SMD package. I could not find others that looked suitable. The NOMCA series have similar specs as the AORN series, and cost about the same (about 5 Euro's) for twice the resistors. Go figure.

What really surprises me is the fact that these (pretty) high precision resistors with a (pretty) good TCR are so inexpensive, compared to through-hole resistors. The only downside that I see is that the power rating per resistor is low with only 100mW. I will not use my KVD with inputs higher than 30V. Correct me if I'm wrong, but I think that self heating by the current is no real issue then, certainly not on my more typical 5 to 10V inputs, let alone if the KVD is used as a bridge component.

So it still looks to me a viable alternative for a hobby application for the KVD.

Going down that path a bit further, I found that both Mouser and Digikey do not have the 2K resistor version in stock for both the AORN or the NOMCA series. You would have to use 2 x 1K in series.

As you alluded to, the 10K resistors of the first decade need to have a method for trimming each individual one, to get them within the matching specification of +/-0.0037%.

If I'm going for the NOMCA series, the absolute TCR is +/-25 PPM/C, and a tracking (I think they mean from resistor to resistor within the package) of 5 PPM/C. Putting a 100K trimpot in parallel, with a typical 100 PPM/C, as you propose, would degrade the total PPM I think.

In my opinion, it is better to use a small resistor in series (can be 50PPM/C) and a 100 Ohm 1 turn trimmer (100 PPM/C) in parallel to the small resistor. The maximum PPM influence with that combination should be less than 2 PPM/C on the 10K, if my calculation is correct. With a 1 Ohm/100 Ohm combo, you can adjust the 10K with an additional 0-0.9 Ohm, and stay close to the absolute 10K value. The value of the 1 Ohm depends on the absolute variations of the 12 resistors (for a 1.1 KVD input) that can be selected from the total of 16 that are available with 2 NOMCA16031002AT5 packages. If they vary too much, you can increase the value of the 1 Ohm.

By using the NOMCA16031001AT5 1K packages for all the other resistors, you could potentially select enough to match for 11 2K resistors and avoid the trimming. What I mean is that by using 1K resistors for decade 2, you need 22, and you need 11 for decades 3, 4, and 5 each, and 10 for decade 6, for a total of 65 resistors. This is one resistor short for 8 x 8 resistor packs, so you can add a normal 1K resistor for the last decade, or add another package. In that case you have 9 packages x 8 = 72 resistors to select from, to get a matching set (+/- 0.037%) for the second decade. Should be possible, I was able to select better matching than spec with a set of 100 1% 2K05 resistors. In this case you can also combine two 1K resistors in series to help matching.

The expenditure budget would be 11 x NOMCA1603 for 5 Euro's each, 11 trimmers for say 1 Euro each and 12 1 Ohm resistors for say 0,10 Euro's, or a total of about 66 Euro's. Not bad. I spent more on my first prototype with batches that totaled about 600 resistors of 1% metal film through-holes to match them to specifications. Actually, I was able to match them to much better than spec.

One of the advantages I think are that it will be easier to control the TC of the individual decades and the total board, compared to using individual resistors, especially when mounted Manhattan style, which is what I did.

Are there any objections to following this route or do any of you see pitfalls?

Tks!