Calibratory D-105 DC Precision Voltage Reference Standard

[engineer_in_shorts]

5V is in the 10V range and therefore within 24hrs the relative accuracy is 15ppm plus 2ppm.

Have you made an uncertainty budget ?

Your calculation is not correct, either!

uncertainty of 5V on 10V range:
15ppm of  reading (5V)  => 75µV
plus 4 ppm of range (10V) => plus 40µV

=> 115µV / 5V = 23ppm

To be exact, you also have to add the A/D non linearity, which is 2ppm of range => 20µV => gives additionally 4ppm, for a  Grand Total of 27ppm.
And then also some ppm for the temperature coefficient, which I'd like to omit now, because it's already included, if you stay within +/- 1°C.

On a 6 1/2 digit DMM, 27 ppm of 5V or 135µV uncertainty on 10V range are visible on the last two digits.


Frank

Yes, thank you for the correction.
I am also pleased to see there is someone who has read the rest of the specifications i.e non-linearity and temp effect.  I call this 'the small print' it's a phrase used in the UK regarding small print of adverts stating terms and conditions etc etc.

[Dr. Frank]

Yes, thank you for the correction.
I am also pleased to see there is someone who has read the rest of the specifications i.e non-linearity and temp effect.  I call this 'the small print' it's a phrase used in the UK regarding small print of adverts stating terms and conditions etc etc.

Yes, it's worthwhile to study the specs precisely, even the 8 1/2 digits DMMs have their weaknesses, if you dig deeper, and they also are not specified correctly or completely, in some aspects..

But another question: Your nickname indicates, that you only wear short trousers during your work?

Respect! 

Frank 

[HighVoltage]

Ok, I will leave the reference and the 3 multimeters running for 48 hours. Since I will not be in the lab tomorrow, I will post a picture in about 48 hours.

BTW, the reference I have hooked up, is the 0.02% by VoltageStandard.
This reference was calibrated by VoltageStandard about 1 year ago.

http://www.voltagestandard.com/New_Products.html

From their website:

Model Vref5-002 Precision Voltage Reference

The Vref5-002 is a 5.000V, 0.0025% accurate precision voltage reference.  Housed in an anodized aluminum enclosure with rubber feet, this reference will be a great addition to your test bench. Based upon the Cirrus Logic VRE305A (used to be Thaler), proprietary laser trimmed circuitry provides excellent stability and accuracy without an oven!  Output terminals are gold plated banana jacks.  Input power is provided by included 24VDC, UL listed power supply.  Input power connector has locking ring to prevent accidental unplugging.  Red LED adjacent to output terminals provides visual indication that reference is powered-up. At time of shipment, the Vref5-002 is adjusted to be within 5uV of 5.000000V as measured by our calibrated 0.0008% accurate HP 3458A 8.5 digit DMM.  Free calibration is included for first two years after purchase- user just pays shipping/insurance both ways; subsequent calibrations are $7.50 plus $8.50 shipping.

Specifications
Accuracy: 0.0025% 
Stability: Reference output is guaranteed to be within 125uV (0.0025%) of 5V for 6 months.  Ambient temperature must be within 6 oF of calibration temperature (typically 68 oF)
Load regulation: 3ppm/mA
Temperature coefficient: 0.83ppm/oF, typ.
Output noise (0.1 to 10Hz): 3uV p-p typical
Size: 5" W x 4" D x 2" H
Power requirements: 110 - 130VAC 60Hz to power included 24VDC power supply.
Warm up time: 30 minutes

[HighVoltage]

I took another picture with a thermometer that I placed on the meters as reference.

[JohnnyBerg]

I am also pleased to see there is someone who has read the rest of the specifications i.e non-linearity and temp effect.

There are more out there 

What I find intriging is the TC specification. What to think from the 18° to 28° range?

Tsja .. resolution and accuracy ..

[TiN]

Since this thread went off rails long ago, here are my 5c.

Doing 20ppm/year will be very challenging IMHO.

Difficulty for 20ppm/y might be little overstated. Even without unobtanium parts something like 5-10 ppm/year should be pretty achievable (especially if one have thermostat box to keep stuff constant over temperature) with home-brew resources and ~300-500 USD for parts, given you already have stable DMM to make sure of drift coming from reference, not a measurement tool.

For statements mentioned about long term stability, i have some data logged in almost month time apart.

21 January 2015

+23.0°C

REF1 : 7.1367921 VDC
REF2 : 7.1365940 VDC
REF3 : 7.1304962 VDC

17 February 2015

+23.6°C

REF1 : 7.1367090 VDC, -11.64 ppm off
REF2 : 7.1366613 VDC, 9.43 ppm off
REF3 : 7.1304826 VDC, -1.91 ppm off

Yes, it's not divider/multiplied output, but do note that references and measurement KI 2002 were powered off/on multiple times and moved from one place to another, even
wiring was subject to change as well. Definitely, better figures are achievable if everything kept constant. No custom parts were used (PCB not counted).

[engineer_in_shorts]

Since this thread went off rails long ago, here are my 5c.

Doing 20ppm/year will be very challenging IMHO.

SNIP ............
 given you already have stable DMM to make sure of drift coming from reference, not a measurement tool.
............... SNIP

That's what we are saying!!!!!! You need a 8.5 digit class DMM to verify a 20ppm/year reference and the peeps on here trying verify the drift of a reference device quoted at 2ppm/year using a 6.5 digit class DMM are just dumb. 

Also a 20ppm/year device is not good enough to verify a 6.5 digit class DMM, so whats the point? as 6.5 DMM are cheap and plentiful.

[paulie]

Those suggestions are so laughable it's hardly worth the time to respond.....if you can't do the simple math to reach a conclusion that you won't be able to even verify these "cheap standards"....then best of luck....

It's very hard to take anything you say seriously

Not one of these devices comes with ANY form of trustworthy traceability.....it's just Awesome14 all over again.....I suspect something about conspiracy and angels will come out next....good grief....

no 6 1/2 digit "lab meter" is going to verify stability to PPM range.  The least significant digit is not trustworthy and @ 10V your least significant digit is 1ppm....

and YOUR suggestion for measuring vREF is this....?    why even bother speculating about "ppm stability".....there is simply ZERO data....

There you go again with all that traceabilty, NIST, 1ppm crap which makes no sense in relation to my posts. You don't seem to get it. Maybe just not capable of understanding what the phrase "5 digit" means. And totally lost in regard to "hobby" or at least pretending so.

I've been noticing your posts since you came on. And the other volt-nuts too. Many wanna-bes but in my estimation probably 2 maybe 3 individuals at most justified in getting into that area. You are not one of them. Not everybody who owns a precision meter (or two) is able to pull their head out of their ass. The phrase "More money than brains" comes to mind.

How's that for elite superiority? I can be a little abusive regarding technical matters but not usually into personal attacks. Thought maybe time for an exception here.

As far as Awesome14 I don't feel too bad being compared there. Even with all the vampire/assassination stuff he still seemed to have one of the more balanced attitudes. Too bad he was banned from the site.  Firing squads worked for Mao and Mussolini so apparently very effective quelling opposition here.  No chance for a defense now. The detractors don't seem too happy either with nothing to strike out at. So I guess anything else that might be in range.

LOL.

[paulie]

Fluke adds some proprietary temperature compensation to the reference circuitry, and this DMM's performance over reasonable temperature swings is very good, so that rumor might be true.

What? Secret method for squeezing extreme performance out of otherwise mundane reference chip? Where did I hear that bef... oh yeah!

[alhoop]

..................................
As far as Awesome14 I don't feel too bad being compared there. Even with all the vampire/assassination stuff he still seemed to have one of the more balanced attitudes. Too bad he was banned from the site. 

..................................

"Awesome14 banned form the site". Who made that decision and why? Foul language sure doesn't get one banned and I don't want to get banned.
AL

[splin]

Your calculation is not correct, either!

uncertainty of 5V on 10V range:
15ppm of  reading (5V)  => 75µV
plus 4 ppm of range (10V) => plus 40µV

=> 115µV / 5V = 23ppm

To be exact, you also have to add the A/D non linearity, which is 2ppm of range => 20µV => gives additionally 4ppm, for a  Grand Total of 27ppm.
And then also some ppm for the temperature coefficient, which I'd like to omit now, because it's already included, if you stay within +/- 1°C.

On a 6 1/2 digit DMM, 27 ppm of 5V or 135µV uncertainty on 10V range are visible on the last two digits: 4.99987 .. 5.00014 V.

That's not so good, more a 4 4/5 digit result, than a 5 1/2 digit one.

It's mainly caused by the mediocre stability specification of the LM399H reference plus the assorted amplifier-resistors.

Frank

Just being a bit picky, but right answer, wrong reason according to my version of the datasheet:
 'A-D Linearity: 0.0002% of reading + 0.0001% of range' - so 2ppm * 5Vrdg + 1ppm * 10Vrange = 20uV.

Splin

[Dr. Frank]

Your calculation is not correct, either!

uncertainty of 5V on 10V range:
15ppm of  reading (5V)  => 75µV
plus 4 ppm of range (10V) => plus 40µV

=> 115µV / 5V = 23ppm

To be exact, you also have to add the A/D non linearity, which is 2ppm of range => 20µV => gives additionally 4ppm, for a  Grand Total of 27ppm.
And then also some ppm for the temperature coefficient, which I'd like to omit now, because it's already included, if you stay within +/- 1°C.

On a 6 1/2 digit DMM, 27 ppm of 5V or 135µV uncertainty on 10V range are visible on the last two digits: 4.99987 .. 5.00014 V.

That's not so good, more a 4 4/5 digit result, than a 5 1/2 digit one.

It's mainly caused by the mediocre stability specification of the LM399H reference plus the assorted amplifier-resistors.

Frank

Just being a bit picky, but right answer, wrong reason according to my version of the datasheet:
 'A-D Linearity: 0.0002% of reading + 0.0001% of range' - so 2ppm * 5Vrdg + 1ppm * 10Vrange = 20uV.

Splin

Sorry, I'm well over 50, so I did not remember the linearity spec correctly.

THX for correction.

Frank

[Dr. Frank]

Fluke adds some proprietary temperature compensation to the reference circuitry, and this DMM's performance over reasonable temperature swings is very good, so that rumor might be true.

What? Secret method for squeezing extreme performance out of otherwise mundane reference chip? Where did I hear that bef... oh yeah!

Ha ha, very funny...

Truth be told though, the LM399 [and it's TempCo tested brother the LM399A] is actually a *VERY* good reference--- far better than what the data sheet says.  And with burn-in and testing/selection, only accepting parts in the lower 10th percentile, well, it can be *exceptionally* good.  The data-sheet spec for an LM399A is 0.3ppm/K [heated of course], and Bob Dobkin told me that the *average* part is going to be more like 0.1ppm/K [they have to state a higher value on the spec sheet for the exceptions].  Now, if you are testing them yourself for TempCo, and destroying 90% of the parts, then it's a good bet that you will end up with a reference that is far better than the "average" 0.1ppm/K.  With a ONE YEAR burn-in at 150C, you are artificially aging the references about 64 years-- so, then you get on the "linear" portion of the logarithmic curve where the drift is 1ppm/year to 2ppm/year.  A single LM399A would kick *any* unheated reference's ass...

There is speculation that the drift in a subsurface Zener is due to the micro-cracks and micro-strain in the silicon die after separation from the wafer.  Supposedly, these crystalline imperfections show up as increased LF-noise [DC to 10Hz].  So, if the rumor is true, if you select for very low LF-noise, it's a good bet that you will end up with a reference that is more stable over time.  If this is true, then I would imagine that it is being kept as a closely guarded "trade secret" by interested parties.  Note that this has nothing to do with the on-chip intentional and unintentional resistors that will account for 99% of the drift, and the burn-in [and subsequent temperature cycling between -55C and +150C] should take care of most of that.  That leaves only the die-attach, which for these chips will most likely be high temperature epoxy that is filled with "micro-balloons".  As the die [heater] is turned on and off, the strain will cause some of these balloons to "pop", and you will see about a 1-ppm jump in the output voltage temporarily.  So, after the burn-in and temp-cycles, the reference heater should be turned on and then off [1-minute on, 1 minute off] for 90 days, but in a cold environment [like a freezer].  This will "exercise" the die attach, and any micro-voids or "ready to pop" micro balloons will have already done whatever it is that they are going to do, and these temporary 1-ppm jumps in voltage will be gone.

So, if you own an older "big name" DMM that you are junking, at least pull the LM399 out of it-- as that reference is likely to be very special indeed.

-Ken

Anyhow, National and LT both specified a typical drift rate only, about  50ppm/year typical, if I remember correctly.
You also could order pre-monitored references, with 20 or 50ppm/year maximum drift.

As most semiconductor suppliers tend to brighten their specs (National and LT may be different from that), these limits really represent, what the LMx99 refs are capable of.
Otherwise, if they would be in reality "much better than specified", then HP or Fluke won't skip 90% of them.

Joe Geller has a very instructive article on his site, how 78 LM299 behave over 1008h:

http://www.gellerlabs.com/LM299AH-20_Case_Study.htm

The corrected diagram shows the usual statistics, some samples are rock stable, others drift wildly, so there's a maximum envelope of about +/- 10ppm after 1008h, where all references stay within.
http://www.gellerlabs.com/LM299/LM299AH_Corr_Time_Drift%20b.pdf

Hardly to predict, how that would look after 1 year, if these drifts grow linearly, or if they all stabilize to something like +/- 50ppm..
For sure, the LTZ1000, SZA263, LTFLU on 45°C would show an at least 10 time smaller envelope, I would bet.


Anyhow, there's no way out to monitor such references long enough, to be able to specify upper limits for the drift, which would be the only reliable parameter..

Same goes for the 6.5 digit (or even higher grade ) instruments.
All of them MAY perform better than spec, or as given by TYPICAL values, but nobody can count on that, if it's going for measuring uncertainties..

Frank

[timb]

Did Bob say how long of a warm up the LM399 needs to stabilize? I notice in the datasheet there's an app circuit for a "portable calibrator" where it's only on as long as you hold down the push button. This would tell me it is stable within a few seconds.


Sent from my Smartphone

[blackdog]

Hi Group,

Just started a new measurement, with two 34461A an a 34401A, all three more than 24 houre's on before the measurement started.

First this, this is a nice link on the website of Gellerlabs about tempco of DMM's
http://www.gellerlabs.com/DMMtempco.htm

Some pictures of 4 instruments measuring a 10V Reference, is uses a quad heathed LT1021-5V TO ic's
First picture meters are in "Null" state, showing the difference for 2 hour.
2x 34461A, The top meter drift a llittledown and the bottom meter drift a little up.
Cal december 2014 and cal 30-jan 2015



One of my TEK DMM4050 DMM's.
Almost no drift, only noise, verry stable meter on 10V range.
Cal Nov 2013, 10 V DC, stil spot on.


This is my old 34401, a view months late for its cal.
It a verry stable meter, but for the last digits it has to be on fore at least 16 hours.


The meters are compared bij a freshly calibrated 3458A.

Now I have a 3458A, and two calibration reports of these meter, one from the beginning of december 2014 and one at the end of December 2014.
You can almost never be sure, that if you measure 10V DC reference, is accurate to within 3 ppm of NIST and PTB.
There are so many variables, cables, connectors, DMM drift etc.

Probably you can really have a little confidence, if you have 3x 3458a and a Fluke 732a. 
If you look at the drift specifications of the calibrators used for my 3458a, you will not be happy...
I know the specifications and the reputation of the drift makes it a lot better than specifications.
But still, it gives me no good feeling.
All I can do, is difference measurements, with multiple instruments to see if my circuit drift. Live sucks 

Kind regarts,
Blackdog

[JohnnyBerg]

Did Bob say how long of a warm up the LM399 needs to stabilize? I notice in the datasheet there's an app circuit for a "portable calibrator" where it's only on as long as you hold down the push button. This would tell me it is stable within a few seconds.

Interresting!

I am doing some experiments with a Lipo and the temp output of a REF5025. I expected it would take quite some time to get a stable output.

Do you have a LM399 at hand?

[blackdog]

Hi JohnnyBerg,

I have about 20 pieces at hand of the LM399, what do you wanna know?

Kind regarts,
Blackdog

[JohnnyBerg]

I have about 20 pieces at hand of the LM399, what do you wanna know?

How fast do they reach there set temperature, and how fast do they settle?

I know from the CO forum that you also experimented with references in ovens. How fast did they settle? (give or take, do we talk seconds, minutes or hours) 

[blackdog]

Hi JohnnyBerg, :-)


The time depends on the thermal mass, the small ones about 2 minutes and the big one's 15 minutes to 1 Houre depending on the temperature difference.
Tomorrow i wil boot up two LM399 fore you en measure until it is stable.

Kind regarts,
Blackdog

[Jay_Diddy_B]

Hi JohnnyBerg, :-)


The time depends on the thermal mass, the small ones about 2 minutes and the big one's 15 minutes to 1 Houre depending on the temperature difference.
Tomorrow i wil boot up two LM399 fore you en measure until it is stable.

Kind regarts,
Blackdog

If you measure the heater current you will know when the temperature has stabilized.

The LT datasheet shows:







So this gives you an idea.

It doesn't say how long it takes to stabilize to ppm levels. I know an HP 3456A stabilizes pretty quick.

Jay_Diddy_B

[blackdog]

Hi Jay_Diddy_B 

I know the datasheet, i wil do tomorrow a test with both KEYSIGHT 34461A DMM's i have, one fore the zener voltage and the other for the current at 12V, 15, 20, 24DC.
How quick it will reach PPM level, we shall see...

Kind regarts,
Blackdog

[SteveyG]

It works well. Try doing better than that on a proto board. You are attempting to do me intentional harm. Shame on you! Why are you doing that? I have done nothing wrong. I manufacture a superb product. If you are one to judge only by appearances, you can't see the truth! And if you're so concerned with doing me harm, I hold you captive, because you really can't harm me. The product performs superbly when used according to even the most basic common sense.

You can get reasonable results on prototype board.

[blackdog]

Hi,


First a picture of the two 34461A DMM's measuring my 10V reference.
Distinctly different behavior, this is also due to the difference of connectors and connectors that are not shielded.



I have done some testing on a couple of LM399 ICs.
With a test jig and a current source, but i measured big variations in output voltage.
It looked like popcorn noise, but the noise pulses lasted too long.
Then it got time to properly shield the LM399 ICs and use a good quality coax cable to connect the LM399 to the DMM.

It only works if your Reference is mounted in a kind animal box.



Mounted in de metal box the strange noise pulses where gone.



Here you can see that the LM399 is in a socket, YES this is BAD!!!
On the left side of the board you can see the temp compensated 1,3mA current source (not in use for the last measurements)



Now this is why a socket is bad...
If you wiggle the LM399 you get big shifts in output voltage.
I good not get the voltage to the same level...



If the LM399 is cold, it takes about 15 seconds to come back whitin 1PPM.
Its like the data sheets tells you, if you look at the stabilized current, just like Jay_Diddy_B told us.
So a pushbutton can be used, but be sure to hold 15 seconds 


Kind regarts,
Blackdog

[paulie]

It works well. Try doing better than that on a proto board. You are attempting to do me intentional harm. Shame on you! Why are you doing that? I have done nothing wrong. I manufacture a superb product. If you are one to judge only by appearances, you can't see the truth! And if you're so concerned with doing me harm, I hold you captive, because you really can't harm me. The product performs superbly when used according to even the most basic common sense.

You can get reasonable results on prototype board.

What is all that circuitry on the right?

[codeboy2k]

It looked like popcorn noise, but the noise pulses lasted too long.
Then it got time to properly shield the LM399 ICs and use a good quality coax cable to connect the LM399 to the DMM.

Interesting, and extremely relevant.   Thanks for posting, I will remember this. If I ever get into ppm and sub-ppm requirements, I will shield my references and use coax to bring it to a meter in the future.