LM399AH REF 10.0000 questions

[jonpaul]

Bonjour a tous: I just got LM399AH Vref PCB from circuitvalley.com  in Germany, on a  nice little board, 50 x 100 mm, with thermal isolation cuts in the PCB and custom trimmed low TC R.

To calibrate my Fluke 8842A DVM and HP 4114/5A precision power supplies, and  Yokogowa 2554 precision V/I  using a  GenRad 1454-A 10 k Ohm voltage divider.

Will  install it in a small Pomona metal box and run from an HP lab power supply.

a few questions please

1/ there is a small voltage Variation of the 10.0000V output with input voltage.   What causes this ( no cm ground loops, differential measuring at output).
Do you suggest adding  (Lm2940-15) LDO preregulator?

2/ What  about the thermal coefficients of the connections and wire eg use ordinary binding post or must be gold plated.  Have some copper braid and silver plated wire. Most is normal tinned Cu. Seems little different voltage with gold binding post connectors or normal.

3/ What is purpose of a Styrofoam thermal shield around the LM399AH, to Reduce heater power or shield it from drafts?

It will be installed in the Pomona  metal box, lab temp environnement about 25 degrees C.  Is the box sufficient without the Styrofoam around the LM399 ?

4/Any significant difference for 5 1/2 digit CAL between use of  LM399AH and an  LTZ1000A Linear Technology part?

Many thanks again for your kind assistance

Bon weekend,

Jon

[dietert1]

To be honest, that board isn't the most sophisticated circuit one would use for a reference. I didn't search the forum, maybe the answers are already there.
- The adjustment pot in that board is definitely in the wrong place. It should sit between the two resistors of the 10:7 voltage divider, in order to have its wiper resistance out of the result.
- Then there is no output stage with a little buffer capacitance to protect the OpAmp from EMI, that may enter from the output terminals. That output filter has been shown by various people here in the forum, e.g. here https://www.eevblog.com/forum/metrology/ultra-precision-reference-ltz1000/msg3269318/#msg3269318. The Keithley circuit shown did not include the buffer cap, but it remains stable when you add one.
- The supply should be current limited at maybe 20 mA, since the output stage isn't short circuit proof.
- The reason for the supply voltage dependence of the output voltage is another error in wiring: Current into the LM399 via R1 should not come from the supply but from the 10 V output. Then a good value for R1 is 2.7 KOhm and the supply voltage dependence will be much smaller.

Maybe you can mod the board to solve the problems. The supply voltage dependence means you did not get a calibrated reference, so you should think about where you can get it calibrated.

Regards, Dieter

[Andreas]

Hello,

most of your questions are already answered in the LM399 thread.

1/ there is a small voltage Variation of the 10.0000V output with input voltage.   What causes this ( no cm ground loops, differential measuring at output).
Do you suggest adding  (Lm2940-15) LDO preregulator?

see: https://www.eevblog.com/forum/metrology/lm399-based-10-v-reference/msg441913/#msg441913
so yes the heater voltage/current has a influence.

2/ What  about the thermal coefficients of the connections and wire eg use ordinary binding post or must be gold plated.  Have some copper braid and silver plated wire. Most is normal tinned Cu. Seems little different voltage with gold binding post connectors or normal.

better material reduces time to stabilize and reduces noise from air drafts. (but you will most likely see it not with 5.5 digit instruments)

3/ What is purpose of a Styrofoam thermal shield around the LM399AH, to Reduce heater power or shield it from drafts?

It will be installed in the Pomona  metal box, lab temp environnement about 25 degrees C.  Is the box sufficient without the Styrofoam around the LM399 ?

No see:
https://www.eevblog.com/forum/metrology/project-pimp-a-keithley-2000/msg1106829/#msg1106829
Try to tilt the box in several directions with and without styrofoam.

4/Any significant difference for 5 1/2 digit CAL between use of  LM399AH and an  LTZ1000A Linear Technology part?

After sufficient run-in time the LTZ1000 has only lesser noise.
https://www.eevblog.com/forum/metrology/lm399-based-10-v-reference/msg796829/#msg796829
so let them run in for 5kHrs to 10kHrs before final calibration.

with best regards

Andreas

[jonpaul]

Rebonjour a tous, many thanks for the notes:

 Still I ask:

 What  about the thermal coefficients of the connections and wire eg use ordinary binding post or must be gold plated. 
Have some copper braid and silver plated wire.
Most is normal tinned Cu.
Seems little different voltage with gold binding post connectors or normal.

Mille mercis,

Jon

[Kleinstein]

At the 5 digit level there should be no detectable difference from different plated cables.  This is something to worry if one is looking at the < 100 nV level.  The plating is mainly there to improve corrosion resistance. Because the plating is thin there is very little temperature difference across the plating.

The problem is more the base material of the binding posts. Pure copper is too soft to be practically used. Brass is common, but can have some thermal EMF, if there are thermal gradients at the posts. With just a low power reference and thus a chance to get small temperature difference they can be still good enough here.

[Dr. Frank]

After sufficient run-in time the LTZ1000 has only lesser noise.
https://www.eevblog.com/forum/metrology/lm399-based-10-v-reference/msg796829/#msg796829
so let them run in for 5kHrs to 10kHrs before final calibration.

with best regards

Andreas

Hello Andreas,
following your link, you certainly mean, that the LM399 shows lesser noise, NOT the LTZ1000?
I don't recognize this said effect, though.. Your CH6 is noisy, and stays noisy, even after 2 years?

Frank

[jonpaul]

Rebonjour, some updates....

 power, will add the LM2940-15 prereg and use either a lab supply or 2 X 9V batteries in series.

Will add a 100 Ohm series R at the output.

 the PCB is SMD and I dont have any good capability to modify it.

For cal of  5 1/2 digit Fluke 8842A  I think its OK.

The seller of the PCB did the CAL at 15.000V input, with HP34465A DVM, the 10.0000 V should be reliable.

to  Kleinstein, I  found old copper braided gold spade tipped leads from Pomona,  and a shielded coax with gold spades. I made a shorting plug with gold plated banana and a gold plated shorting link so I am OK on thermals.

Many thanks again,

Jon

[Andreas]

following your link, you certainly mean, that the LM399 shows lesser noise, NOT the LTZ1000?
I don't recognize this said effect, though.. Your CH6 is noisy, and stays noisy, even after 2 years?

Hello,

generally the LTZ1000 has lower noise (~1.2uVpp) against a LM399 (~3-4uVpp for a good device).
And also settling time and popcorn noise is somewhat lower on a LTZ1000.

Of course my measurement is limited by the instrument (a AD586 based device) in noise so you cannot distinguish directly the noise between LM399 and LTZ1000A.

And yes LM399 #CH6 is rather noisy still some days later. It suffers obviously from some popcorn and random walk noise.
LTZ#2 had some "events" on the unbuffered output in the mean time. (some ppm shifts).

So its always good to have at least 3 stable references to observe which one is shifting.

with best regards

Andreas

[dietert1]

..
 power, will add the LM2940-15 prereg and use either a lab supply or 2 X 9V batteries in series.
..

The datasheet says that you will have a supply voltage uncertainty of +/- 0.45 V. The differential resistance of the LM399 is 1.5 Ohm max and the nominal R1 is 5.6 KOhm, so your reference will have uncertainty of +/- 0.45 V * 1.5 Ohm / 5600 Ohm = +/- 12 ppm of 10 V. That is from supply voltage versus calibration supply voltage alone, besides others. You decide whether that is good enough. I would get two precision resistors 5K and 10K and use an OpAmp to derive the 15 V supply from the 10 V output.

A LM399 with the correct circuit is good enough to calibrate a 5.5 DVM. Noise isn't any problem as long as it is an integrating ADC (like in most DVMs). At 100 PLC (2 second integration time) our HP 3456A meters had a standard deviation of about 0.04 ppm when logging a low noise source (6 V, 4x alkaline battery). Those meters have LM399 references.

Regards, Dieter

[jonpaul]

Deiter: Many thanks, very good info.

The 15V reg is reading 14.90V loaded by the Vref board, with an input from 15.12 V to 30V

With only 100 mv change, I estimate the effect on the ref will be only 2.67 PPM, well below my capabilities to read on 5 1/2 digit meter.

I may be able to trim the LM2940-15 to exactly 15V.

About current limiting, the board has an op amp LT1001ACN which is SC proof and limits curent at ~ 20 mA so I do not need any output protective series R.

Regarding AC noise, the RMS AC of the Fluke 8842A   ~ 130 uV. Fluke 8920A TRMS set on LO BW 2 MHz reads 365 uV. A Yokogawa scope measures 315-375 uV RMS in a 20 MHz BW.

Finally I tilt the unit in the box 90 and 180 deg and see little if any change in Vout.

I believe its OK for my uses, but certainly it can be improved!

Mille mercis encore, pour votre gentil aide,

Jon

[Kleinstein]

The resistor at the output is not just for current limiting. The usual reason for a resistor at the ouput is that OPs generally don't like capacitive load and they may start to oscillate. For some of the faster OPs it only takes some 100 pF (or 1 m of coax) to make them oscillate. Instead of just a resitor there can be special output stages that give low DC ouput resistance and can still tolrate quite some capacitance (e.g. like some 1 µF).

With the noise it is usually not so much about the higher frequency noise, but more about low frequency noise, e.g. 0.1 -10 Hz as the usual standard band for low frequency noise. The actual relevant noise may be even lower (like 1 - 100 mHz), but this is slow and sometimes difficult to measure. The higher frequency noise can be reduced with filtering, but this does not work well below some 1 Hz.

[bdunham7]

A LM399 with the correct circuit is good enough to calibrate a 5.5 DVM. Noise isn't any problem as long as it is an integrating ADC (like in most DVMs).

I have to disagree when it comes to the 8842A unless some conditions are met.  The 8842A is a 30ppm device, despite being only 5.5 digits/200K counts.  It also has a very different type of ADC and part of the calibration adjustment is a particularly fiddly, iterative calibration of the ADC linearity.  I strongly recommend leaving that part alone unless you have very high end equipment and a lot of patience. 

If you were to amend your statement to say a reference using a properly aged and characterized LM399 and calibrated with a reference at least as good as a 3458A, I might agree.  Here we have an inexpensive reference (I haven't looked at it) with unknown aging and history calibrated with a DMM that has about the same uncertainty as the 8842A. 

[bdunham7]

To calibrate my Fluke 8842A DVM and HP 4114/5A precision power supplies, and  Yokogowa 2554 precision V/I  using a  GenRad 1454-A 10 k Ohm voltage divider.

1/ there is a small voltage Variation of the 10.0000V output with input voltage.   What causes this ( no cm ground loops, differential measuring at output).
Do you suggest adding  (Lm2940-15) LDO preregulator?

2/ What  about the thermal coefficients of the connections and wire eg use ordinary binding post or must be gold plated.  Have some copper braid and silver plated wire. Most is normal tinned Cu. Seems little different voltage with gold binding post connectors or normal.

4/Any significant difference for 5 1/2 digit CAL between use of  LM399AH and an  LTZ1000A Linear Technology part?

I can only answer as to the 8842A as I have no experience with the other two.  I would think long and hard before you press that CAL button on the 8842A.  What you have is probably OK for checking it, but IMO is nowhere near good enough for an actual calibration adjustment.  I doubt your 8842A needs adjustment in any case and the process is very complex and fiddly if you are calibrating the ADC linearity.  There's nothing to indicate that either the voltage reference or the divider is anywhere near accurate enough to calibrate a DMM with 30 ppm basic accuracy. 

1) as others (dietert1 I think) have mentioned, this is a design issue and though it can be solved overcome with a precise input voltage, you need to make sure that the input voltage doesn't vary with heater current.

2) I don't know about plating, but you will have thermal voltage issues at the lowest ranges.  The maximum resolution of the 8842A is 100nV, and if you have drafts or don't let the connections settle, you'll have a dynamic display.  The input connections are shielded bananas and not some metrology grade binding post, so there's only so much you can do.  That low voltage range is a bit noisy no matter what and the accuracy specs are much wider, so you need not obsess over every last count, but you should shut the windows and turn the ceiling fan off.

4) I think the actual implementation is more important than the particular reference used.  A properly aged, selected and dialed-in LM399 should be pretty good.  I had a quick look at the one you have (I think) and I'm a bit puzzled--they claim 0.5ppm/C but have 10ppm/C divider resistors and I don't see any trim pot?  And I don't see anything about actual uncertainty specs, just that it is 'calibrated' with a 34465A.  Am I looking at the right unit?

Perhaps you should limit yourself to some testing for now--see what those instruments tell you about the current state of calibration and then decide whether you believe them.

[jonpaul]

Rebonjour a tous!

To precise the reference 10.0000  I have in hand, attached are a few photo's.

My mods are off the PCB:

Added LM2940-15 LDO prereg trimmed to 15.00 V to feed the PCB

Added 2 x 9V batteries and ext PSU input

Place PCB in Pomona metal cast box with battery, switch and gold binding posts.

Finally, I plan to use a GenRad precision divider, 1454-A to calibrate the other equipment, it has proven very accurate.

Appreciate any comments or feedback!

Mille mercis,

Jon

[Kleinstein]

The LM399 circuit looks not the very best: the Zener current from the main supply is not ideal.

The reference has so seprate drive and sense terminals -  the cable / contakt resistance may become a small issue with the 10 K divider. This is especially true with an added resistror at the output to reduce the capacitive loading. Even with inductor and resisor in parallel there is still some resistance.

[jonpaul]

Bonjour Herr Kleinstein!

1/ The PCB has a 4 term layout and separate Vin and 10 V out.

2/ The opamp os LT1001, the spec sheet says  stable over a wide range of cap load. I saw no oscillations with 1 M RG/59/U several 100 pf. It is alos short ckt protected so I do not believe  series R is needed.

3/ Zener runs from the +15V regulated out of LM2940-15.

4/ Any comments re the Gen Rad divider?

Kind Regards,

Jon

[Kleinstein]

The 10 K divider needs quite some current and without seprate sense and drive terminals one would see a little error from the cables. With a fixed 10 K load, one could consider an only approximate compensation with just 2  +/-1 mA current sources of moderate accuracy that would provide most if the current. The reference circuit could than be left unchanged.

In the old days of mercury reference cells and null meters, one used an auxiliary source to drive the divider and used the reference only to compare with a null meter.

The swtiches of theGenrad divider look really good.

[jonpaul]

Mr Kleinstein, many thanks again!

1/ The ref as is seems to be close to the Yokogawa calibrator and the Fluke 8842A. It is also near the HP4114A and 4115A precision supplies so I have some confidence in it.
The designer and cal man in Germany seems a serious guy....

2/ RE the 1/F noise, I am aware of that, but my main goal is just to CAL the Fluke, which is integrating.

3/ RE  GenRad 10K divider, was from LLNL decades ago.
Previously,  I discussed the cleaning and lube, label residue and paint stencils.

±0.005% of full scale  ... for input  voltages below one half of rating   ±0.04% of reading 
https://www.ietlabs.com/pdf/Manuals/GR/1454-A,%20-AH%20Decade%20Voltage%20Dividers.pdf

Testing the ratios, tap to end resistance , seems  much better than the spec.

I am aware of the standard cells and null measurements.

Will try a CAL today on the Fluke.

Bon Journee,

Jon

PS: Epoch 1965-7  was  in college and worked as tech at Julie Research Labs in NYC, I had worked with Loeb Julie....
We were winding and then  testing Poseidon missile resistors in a precision temp oil bath.
I  saved a few   JRL resistors but sadly,  no other JRL equipment.

[bdunham7]

the Fluke, which is integrating.

Is it?  An analog filter isn't exactly the same as an integrating ADC.  And you won't have the option of selecting the analog filter (reading rate) during the CAL process.  I don't know whether it is engaged or not at that time.

[Kleinstein]

At least if the analog filter is activated the Fluke 8842 with it's recirculating reminder ADC is close to an integrating ADC. The higher frequency noise of the LM399 is not great ( ~ 100 nV/sqrt(Hz)), but it is still better than most (HP3458 and similar may be about that level) DMMs in the 10 V range. So for the reference noise there is not much difference between a ture integrating ADC and the ADC in the Fluke meter.
For the lower frequency noise the Fluke ADC with multiple readings will behave like an integrating ADC.

Unless there is an absolute need to do so (e.g. a lost calibration), I would not use such a 10 V ref for a calibraion with adjustment of the Fluke 8842: it is not really up to the task and there are more things to go wrong.
The reference and dividers may still be used for a plausibility test - so measure the values but not change the meter internal constants.  Most calibrations would be of this type anyway: measure the performance and not the difference, but not adjustment. So not even a meed to call the meter internal cal procedure, just normal measurements at quite a few test points.

[jonpaul]

Hello again all

Many  thanks to Kleinstein :

 The original 8442A DVM CAL was not preserved it had bad zero offset and bad CAL on all DCV ranges, so a CAL was required.

That was the motivation to get the 10.000V ref.   I have done the AD CAL a few times with better and better results.

For the  VDC CAL: Fluke needs accurate sources  100, 1000V, I do not have any of the big CAL boxes, Fluke, Keithley, etc.

Plan:

1/ Use 10K GenRad divider 1454-A and 10.00 REF to  CAL 10.000V  and below.

2/ Use   JRL resistors 0.01 and 0.025% and   GenRad for accurate 10:1 and 100:1 dividers.

3/  1000 V source HP 6515A  use  100:1 divider,  with 10.000 REF to calibrate

My null detectors: D'Arsonval center meters  and a TEK AM501 diff amp

4/ 100 V sources: HP6115A precision PS and Yokogawa 2554 Voltage Cal.

Repeat process above  with 100 V sources and 10:1 divider

Interested in your comments,

Bon Journee,


Jon

[bdunham7]

The original 8442A DVM CAL was not preserved it had bad zero offset and bad CAL on all DCV ranges, so a CAL was required.

That was the motivation to get the 10.000V ref.   I have done the AD CAL a few times with better and better results.

For the  VDC CAL: Fluke needs accurate sources  100, 1000V, I do not have any of the big CAL boxes, Fluke, Keithley, etc.

OK, so if the cal was off it will be interesting to see if recalibration fixes it.  Do you know why it was lost?

If you have survived the ADC CAL procedure and you think it is reasonably accurate, congratulations--the rest of it should be easy!

For your proposed procedure this may not matter, but keep in mind that the 8842A is able to accept non-standard reference sources.  You don't need an exact voltage, just a stable and known voltage.  This is typically done with a stable power source of some sort and a reference meter.  This is especially helpful when calibrating the resistance ranges.

Best of luck!

Edit:

3/  1000 V source HP 6515A  use  100:1 divider,  with 10.000 REF to calibrate

Unless I missed it, you are using the GenRad 1454A as the 100:1 divider?  That would seem to be too much voltage for a 10K divider.

[Kleinstein]

If the Fluke 8842 calibraion is really lost, it is OK to use a not so ideal resource to do at leat a crude calibration. Just keep in mind that the accurcy may be lower than the specs.

For the higher Voltage ranges, one could also consider to use a Hammon type 1:10 divider. This may work better than 0.01 %. It would be at least a second, independet path to check.
Another option may be to measure some 10 nominalle equal resistors for a divider, even if lower accuracy. The meter may be more accurate for small differences than 0.01 %.

Instead of relying on the specs of accurate resistors, one may consider using the meter to measure the resistance and maybe use a few in series to avoid very large values. It can still be tricky for the 1000 V divider. It may still be worth the math if measuring 10 x 1 M is better than the accuracy one gets from a ready made 10 M resistor.

[mawyatt]

We use 6 resistors measured as close as possible to each other. Then 3 in series and 3 in parallel give a 1/10 voltage reduction with a simple divider.

Best,

[jonpaul]

Bonjour a tous,

Please note that I am relatively new to  metrology, having  only a modest CAL lab, accumulated randomly over decades of Ham Radio fleas and ebay.
GenRad divider,  few capacitance, resistance  and inductance standards, several HP RLC bridges, and the power supplies and Yokogawa CAL as mentioned.

I was very lucky to find some of these CAL devices, which sat unused and unrecognized until recently. 
The voltage standard was my labs weakest point, I had only an ancient Epley standard cell.

NOTES and RESPONSES:

1/ 8842A was unused since 2010, unknown last CAL or history but in excellent physical condition in and out and with TRMS AC.
Poor CAL due to  backup NVRAM or battery ?

2/  GenRad 1454-A is for use at <10V.  10.000K checked with a Leeds and Northrup 10000 ohm standard 4 terminal (4040-B) 0.001%

https://www.surplussales.com/equipment/testequipment/pdf/eqp-ln-4025b-ln-4045b_PDF.pdf

3/  RE For 100V and 1000V supplies CAL  from the 10.000V source: 
 GenRad in series with R decade below, set   90K  10:1 or 990K  100:1 ratios

4/ Decade Resistor is  ZM-16B/U from Aimes Engineering, 0.1 ohm - 100MOhm, in 9 decades. It is NOT super accurate BUT is very stable.

https://www.liberatedmanuals.com/TM-11-5102.pdf

I setup the decade to  ~ 90K and 990K  and  then trim   lower decades  for exact rations 10:1 and 100:1.

Aimes is rated 7 mA on the 0-100K decade (for the 100V source) and 2 mA on the 0-1000K decade (for the 1000V source)

5/ The Fluke 8842A has no settings for analog filters.
Fluke calibration is iterative, and improves with each pass.  I have three passes on AD and two passes on VDC so far.


I welcome any comments or  observations on Yokogawa calibrator 2554 orHP 6515A high voltage precision power supply (max out is 1600V)

Bon Journee,

Jon