Homebrew standard to calibrate 3478A

[cracklincrotch]

I recently purchased an HP 3478A on fleabay.  Price seemed good for what it is.  Auction was marked as untested but it powers up and there is no calibration error.  Arrives Friday.  If it arrives as described replacing the battery is the first thing that happens to it.  The best meter I've had up to now is a 6600 count BK Precision 2709B so the 3478a is a step up.

I want to test its relative accuracy but I don't have any way to do that.  I've been reading through the forum here about people that have made their own standard, traceably calibrated or not, to compare their equipment to.  So I thought hell I'll re-invent the wheel because, well, why not?   I designed one with the 3478a range calibrations in mind.  It's based around the AD587UQ.  The circuit provides 10V (30V 1/3rd scale), 3V, 300mV, and 30mV outputs, 30 ohm through 30Mohm resistors, and 100mA and 300mA DC outputs. 

I'm not new to electronics design.  I'm self-taught (young player.)  I don't have any paper on the wall that says I'm supposed to know what I'm doing.  I learn as go and usually by example.  But I'm afraid that my design is a bit of a sausage fest because the voltage dividers.   

I paralleled a number of the resistors for improved accuracy as well as to create resistances that didn't exist for a particular accuracy.  There's a lot of resistors.  The 3478A has a 10Gohm input impedance so I wasn't worried about a voltage drop across the dividers.  But there's a LOT of resistors and I can't help but think there's gotta be another way to do this.  If there isn't well fine.

Not intending on having this circuit traceably calibrated.  I'm looking for the best out-of-the-box solution to check the calibration ranges of this (yet to arrive) DMM.  If someone here laughs and says it's not gonna work as intended and here's why, or do such-and-such and that'll fix it, that's great! 

[Andreas]

Hello,

2 things around the AD587:
- I rather would use film capacitors instead of the tantalum capacitors
  (especially on the noise reduction pin where 1uA gives around 5mV output voltage change).
- Op-Amps tend to oscillate with capacitive loads above 1nF. You might want to use a capacitive isolation cirquit.
  (see the output buffer of my AD587LW cirquit with 2 additional resistors and 1 additional capacitor).

BTW: I do not understand how you want to generate the currents with your cirquit.

with best regards

Andreas

[cracklincrotch]

I intended on having the current generated when the DMM is connected across the emitter of Q1/J16 to either J17, that has the 1k resistor for 100mA @ 10V, or  J18, that has the 3k resistance. 

[mmagin]

If you don't have access to a higher-resolution meter of known calibration or a calibration standard, something like http://www.voltagestandard.com/-.html is nice.  (And even that is not the order-of-magnitude better that you want for actually doing adjustment of the 3478A.)

As best I'm aware, you're not likely to find any voltage reference ICs with a tolerance anywhere near as good as the accuracy of the 3478A, so unless someone has calibrated your standard, it's not going to help you much out of the box (except for checking for gross problems with the 3478A, perhaps.)

[cracklincrotch]

If I were to have access to a calibrated/known device I’m only interested in what my device values are, then use that information to calibrate my 3478, yes?

[cracklincrotch]

Would it be worthwhile putting multiple AD587s in parallel improve voltage accuracy?

[Conrad Hoffman]

The 3478A is a decent 5.5 digit voltmeter, but IMHO it's a great ohmmeter, especially for low ohms and 4-wire measurements. Not much to say about standards, as this forum covers about everything. BTW, I've had mine for many years and it's never needed adjustment. It does, however, need a reasonable time to warm up for full accuracy, probably an hour or so.

[Brak]

If you're having trouble finding it, the AD587LW output buffer Andreas spoke of is at:

https://www.eevblog.com/forum/metrology/ad587lw-10v-precision-travel-standard/msg1449492/#msg1449492

Brak

[intabits]

Maybe I'm being thick here, but U3B & Q1 are trying to source current to ground via a resistance connected to J16, such that the voltage at J16 will be 10V, correct?
And if J16 were connected to J17, we'd then have 10V across 1k, giving a current of 10mA, not the 100mA described.
What am I missing here?

[intabits]

A couple more questions:-

Q1: Shouldn't the Ammeter UT really be inserted into the emitter leg of Q1, so that it's burden voltage has no effect on setting J16 to 10V?
IE: the 10v at J16, and the current through the AUT and R45 is determined solely by the precision value of R45.

Q2: The series/parallel resistor strings supplying the 300mV and 30mV references. Is there any reason to/not-to put "rungs" across the ladders?
One reason not to that I can see is that a failed resistor would cause a more easily noticed gross error in the output, but that's so unlikely. Any other reasons?

[cracklincrotch]

Maybe I'm being thick here, but U3B & Q1 are trying to source current to ground via a resistance connected to J16, such that the voltage at J16 will be 10V, correct?
And if J16 were connected to J17, we'd then have 10V across 1k, giving a current of 10mA, not the 100mA described.
What am I missing here?

You're not missing anything.  I was. 

[cracklincrotch]

A couple more questions:-

Q1: Shouldn't the Ammeter UT really be inserted into the emitter leg of Q1, so that it's burden voltage has no effect on setting J16 to 10V?
IE: the 10v at J16, and the current through the AUT and R45 is determined solely by the precision value of R45.

I think I see where you're going with that.  But if I changed the circuit to reflect that, that be leaving U3B in open loop.  Wouldn't that be an issue? 

Q2: The series/parallel resistor strings supplying the 300mV and 30mV references. Is there any reason to/not-to put "rungs" across the ladders?
One reason not to that I can see is that a failed resistor would cause a more easily noticed gross error in the output, but that's so unlikely. Any other reasons?

So long as everything is working the rungs wouldn't matter.  But I agree with you.  It would be easier to find an issue without them there.  I've removed them.

[cracklincrotch]

Schematic updated, as attached.

[SvanGool]

10/300 = app. 33.3 not 300 mA

[cracklincrotch]

10/300 = app. 33.3 not 300 mA

Those resistors should have been in parallel.  Fixed.

DMM arrived a day early!  Meter looks in great shape externally.  Most important part of all is "SELF TEST OK"

[SvanGool]

DMM arrived a day early!  Meter looks in great shape externally.  Most important part of all is "SELF TEST OK"

First thing to do is a rough sanity check, just connect some resistors, voltages etc.
I had a 3478a that said "SELF TEST OK", but the hybrid input circuit was defect (located before the self test relay), so it was still not good.

[GigaJoe]

10V with 1LSD error, produce 3LSD on 30V scale; i had 5 such devices (now 2) , and none of them drift more then 1LSD in 3 years on 30V voltage range. they drift most  on a resistance range.

black HP label - usually on the latest devices; earliest units has a dual color
you may need replace battery, there sequence in a forum somewhere ;  i suggest to replace capacitors on a power filter. shorten on 30 mv range should produce noise within 0.3-0.7 micro-volts; if it over 1.5 -3 micro volts - 100% caps.  replace it ; and double it capacity . a huge cap on bottom , it GPIB filter. - i didn't touch it ...

you may find a range resistors better TC like 5PMM to constrict 30ohm 300ohm etc ... and send someone for the favor they can measure it. 
then construct 10V ; then 30V; then build a voltage dividers with low TC, then 5.5 became not enough and you buy HP3456A  (use this service: www.crossborderpickups.ca) 

and finally you done ....

[TERRA Operative]

First thing to do is a rough sanity check, just connect some resistors, voltages etc.
I had a 3478a that said "SELF TEST OK", but the hybrid input circuit was defect (located before the self test relay), so it was still not good.

Same here. AC ranges don't work.. The rest are spot on accurate.

[cracklincrotch]

10V with 1LSD error, produce 3LSD on 30V scale; i had 5 such devices (now 2) , and none of them drift more then 1LSD in 3 years on 30V voltage range. they drift most  on a resistance range.

black HP label - usually on the latest devices; earliest units has a dual color
you may need replace battery, there sequence in a forum somewhere ;  i suggest to replace capacitors on a power filter. shorten on 30 mv range should produce noise within 0.3-0.7 micro-volts; if it over 1.5 -3 micro volts - 100% caps.  replace it ; and double it capacity . a huge cap on bottom , it GPIB filter. - i didn't touch it ...

you may find a range resistors better TC like 5PMM to constrict 30ohm 300ohm etc ... and send someone for the favor they can measure it. 
then construct 10V ; then 30V; then build a voltage dividers with low TC, then 5.5 became not enough and you buy HP3456A  (use this service: www.crossborderpickups.ca) 

and finally you done ....

Quick sanity test, after the unit has been powered up for a bit more than an hour.  Temperature in the room has been a steady 27.65C, an average from two side-by-side 1-wire probes in 12-bit mode.  Relative humidity is about 45%.

Ohms tests:
30K 1% through hole resistor was the closest I grabbed.  DMM reads 29.9646k.
1k 5% through hole resistor next measured 0.97591k
60.4ohm 1% through hole measures 61.339 on 2-wire, 60.503 on 4-wire
30.2ohm (60.4 1% || 60.4 1%) measures 31.853 on 2-wire, 29.6518 on 4-wire
22ohm 1% 0805 measures 23.2127 on 2-wire.  Can't get on to it with 4 probes.

DC tests, using my DP832:
PSU 10mV reads as 13.0579mV
PSU 100mV reads 103.039mV
1V reads 1.00247V
10V reads 10.0024V
30V reads 30.0035V

DC 30mV range shorted produces 0.00000V on the display

DCA tests using 100ohm 5% 1/4W + DP832:
5V = 48.975mA on the 3478a
1V = 9.757mA

ACV 28V reads 27.9020VAC.  BK Precision 2709B reads 28.09.

I've read about shorted power supplies/blown fuses on these due to failed caps.  It's on my list.

3456... Don't know that I could sneak that by the wife

[GigaJoe]

"DC 30mV range shorted produces 0.00000V on the display"   -  sure     30mV  : 00.0000  - last digit fraction of microvolts - always has a noise. that you state are 3V range

[cracklincrotch]

Yep.  Wrong range. 

With the probes clipped together the display fluctuates from -00.0006 to +00.0006mVDC.

[GigaJoe]

it seems calibration good in overall, null offset are great, and voltage seems all ranges offsetting on +3mV - seems you "BK" are incorrect ...
all my devices , drift offset (30mV range)  on 2-3 microvolts for an year ... ; this one may have a recent calibration then it decommissioned ...


the only 2 negatives in overall for this device:  display, and measure speed;  often I switch autozero-off and 4 digits ...

[cracklincrotch]

Thought it was time to take it apart.  Newest date code I could find on a part was 8926.  Date code on the board sticker was 09-8930.

The unit has had the power switch and battery changed soon after production.  Whoever did the work didn't clean the flux off.

[GigaJoe]

first models that large capacitor around the battery, it was shielded , then became a regular; latest model has some kind chip instead isolated coils; in overall nothing special

another thingi i love: literally no warmup time ....

in one unit i use 4 CR2450;  you may made a holder for a few 2032 and change it every year ... simple thing when few in parallel.

[cracklincrotch]

Updated the schematic. 

Q1: Shouldn't the Ammeter UT really be inserted into the emitter leg of Q1, so that it's burden voltage has no effect on setting J16 to 10V?
IE: the 10v at J16, and the current through the AUT and R45 is determined solely by the precision value of R45.

^^^ Done.

Added an option to add two more AD587 references to increase accuracy. 

Attached PDF doesn't reflect that I terminated the 4th unit of the U3 opamp by connecting a 2k resistor divider  pin 12 and connected pin 14 to pin 13.

[intabits]

Updated the schematic. 

^^^ Done.

My post was more a question/suggestion rather than any sort of knowledgeable recommendation, as I've never played with that sort of stuff (but intend to soon).
It just seemed to be a problem, but I don't know if what I suggested would have other ramifications. As it's all DC, I would think not, but maybe you should just breadboard that part and see that it behaves. (maybe a little capacitance somewhere might help keep things stable?...)

Otherwise, it's all looks good to me.

[cracklincrotch]

I thought about it and I don't know if the burden of the meter would affect it or not either.  This way it definitely doesn't.  No point finding out afterwards that it does.

[cracklincrotch]

I made some other modifications to the schematics to fix a couple other issues with "stacking" voltage references, added TVS on the test points, and reduce BOM cost a bit.

Schematic and layout attached.  It's a four layer stack up.  1st: signal, 2nd: ground, 3rd: +15V supply, 4: nothing.

[e61_phil]

I would try to use a configuration in which the lower resistor/voltage divider is included in the next stage.

For example:

100R as smallest
900R as the next which gives 1k in sum
9000R gives 10k together with 900R and 100R
and so

It is also possible to do it in your 3x steps. This way you can save some resistors and you can do some transfer measurements (ohms and voltage).

Perhaps you can use the same resistor chain for volts and ohms

[try]

Hello cracklincrotch,

I have no idea what resistor values are available in SMD.

The equivalent resistor ladder with regards to Phil's proposal would be as follows:

(all values are Ohms)

TP0-30-TP1-270-TP2-2700-TP3-27k-TP4-270k-TP5-2,7M-TP6-27M-TP7

Those values give you a decadic ladder measured against TP0:

TP1: 30
TP2: 300
TP3: 3k
TP4: 30k
TP5: 300k
TP6: 3M
TP7: 30M

Lack of availability of 2,7M and 27M could require manual assembly.
Unfortunately TP6 and TP7 will be unstable due to a high TK.

I wouldn't use 3 references.

I would build the portable calibrator as described in the LM399 datasheet and adjust the orginal scheme for a x4 multiplier in the feedback path (and replace the 200k by 390k and the 7,5k with a 22k)
This gives you approx. 28V.
You need roughly 30V supply power.
Behind the 28V you place an off-shelf decadic divider (Vishay, Caddock).
Use a switch to connect the different output ratios to a buffer.
The decadic divider provides the fractions 2,8V, 280mV, 28mV, 2,8mV, 280µV...

You can realize the x4-Multiplier with 4 discrete resistors or use some ratio-stable LT5400.

Giving up a little bit of stability for an output value closer to 30V you would combine 4x10k in series with 2700 and 150 Ohm giving you a divider of
10000k - (3x10k + 2700 + 150) for an output of 29,995V with an input of 7V for instance.
The decadic fractional values follow accordingly (2,995V; 0,2995V ...).

Regards
try

[cracklincrotch]

Those are all great ideas. 

I'll make changes based on those suggestions and post back when I've finished.

[Fujiguy462]

Ant update on this project?