Best out-of-the-box 10V reference?

[Kleinstein]

If one has references of different noise, one can still do an average - however one should give a different weight to the references with the highest weight to the lowest noise one. If uses resistors to combine the references the best resistor value would be proportional to the square of noise (measured as a voltage).

If a references has significant higher noise it will not contribute much, and there maybe not much sense in using that one at all.

[RandallMcRee]

I don't think its that simple.

There is both drift and noise.  Averaging, at least theoretically, could improve both of these. For improving drift I would think equal value resistors are best.

Just a thought.

Randy

[pitagoras]

The application I have in mind is a Mini Metrology Lab.  Just to enter the game.

At first, it will be my best shot at calibrating the rest of the lab under construction (simple DMMs, PSUs).
I'm interested in experimenting with old-fashioned zener references and FET ones.

The rest of the game includes experimenting with sigma-delta converters ("discrete").

So whatever makes the ref "better" in some sense or another.

I agree that, at least intuitively, it looks like drift would be improved by averaging.

I will study the threads on low freq noise and popcorn.

Thanks for your valuable time.

[tggzzz]

I agree that, at least intuitively, it looks like drift would be improved by averaging.

Consider a source that drifts monotonically from 10V to 9V over the course of a year. What averaging would be necessary to improve that, and what would the improvement be?

Ditto a source that drifts during the day as temperature changes?

Averaging only works in some circumstances, and understanding those conditions will make you a better engineer.

[pitagoras]

Averaging only works in some circumstances, and understanding those conditions will make you a better engineer.

Yeah. I'd like to understand better the conditions. I understand the following:
1) "popcorn" noise is related to "imperfections" in the construction (or inherent characteristics of semiconductors). Mitigated by the manufacturer in the construction process
2) There is "climate" drift related to temperature (mainly mitigated with ovens)  and humidity (mainly mitigated with ceramic packages)
3) There is some kind of drift related to mechanical stress. Mitigated by "floating" the component.
4) There is long term drift related to aging. Mitigated by pre-aging and selection of parts.
5) There is low frequency noise, a.k.a. 1/f, 0.1 to 10Hz noise... this one I suspect this is related to some oscillation or "ringing" in the implicit control system... but I don't really know here.

What else?

[pitagoras]

hey tggzzz, funny enough, I happened to land in the link in your signature earlier today, by chance, while googling weston cells

[tggzzz]

Averaging only works in some circumstances, and understanding those conditions will make you a better engineer.

Yeah. I'd like to understand better the conditions. I understand the following:
1) "popcorn" noise is related to "imperfections" in the construction (or inherent characteristics of semiconductors). Mitigated by the manufacturer in the construction process
2) There is "climate" drift related to temperature (mainly mitigated with ovens)  and humidity (mainly mitigated with ceramic packages)
3) There is some kind of drift related to mechanical stress. Mitigated by "floating" the component.
4) There is long term drift related to aging. Mitigated by pre-aging and selection of parts.
5) There is low frequency noise, a.k.a. 1/f, 0.1 to 10Hz noise... this one I suspect this is related to some oscillation or "ringing" in the implicit control system... but I don't really know here.

What else?

Forget the physical mechanisms; averaging (and limitations) is a mathematics/statistics issue. After you understand the maths/stats, you can start see how it can/can't help you with the physics.

[tggzzz]

hey tggzzz, funny enough, I happened to land in the link in your signature earlier today, by chance, while googling weston cells

That's gratifying

I now have 4 saturated weston cells, and a couple of 10V "Trancell" zener-based references => the too many clocks problem

[pitagoras]

the too many clocks problem

Come on, one can never have too many clocks
The only problem here is not having a calibrated 3485A...

[Andreas]

2) There is "climate" drift related to temperature (mainly mitigated with ovens)  and humidity (mainly mitigated with ceramic packages)

Hello,

so for me it looks you are on a good way.

But one point: you mentioned the KR version above which should be the SO-8 epoxy-version and not a ceramics dip version (which would have a Q in the 2nd suffix).

with best regards

Andreas

[Kleinstein]

The low frequency (1/f) noise can be similar to popcorn noise, just on a different level and many processes in parallel.
Averaging helps primarily against noise: so white noise, 1/f noise and popcorn noise as well.

With drift it is well possible that most refs if the same type show drift to the same direction and thus averaging similar refs will not help very much against drift. It can help a little with linear TC - but already the second order TC is usually not at all random. For things like long term drift with multiple refs one usually looks at the ref units individually and than decides which one to trust and include in an average.

[pitagoras]

Thank you for your answers.
Yes, I got epoxy version because of current budget restrictions.
Could/Should I ovenize the reference?
I will have 2 units so I can experiment with one ovenized.
But to check drift I need something more stable, like building another reference based on LM399...

[Andreas]

Could/Should I ovenize the reference?

Hello,

that depends on your stability needs and the individual T.C. near room temperature of your references.
You can also try a temperature compensation by NTC like in the Geller SVR-T reference.

The DIP8 epoxy package has after my experiences a humidity coefficient up to 0.5 ppm/%rH. (LT1027CN)
Lars sent me some measurements for AD587LN with up to 0.3ppm/%rH.
According to a MAXIM datasheet the SO8 packages may be less stable by a factor of 2-3.

With best regards

Andreas

[not1xor1]

Could/Should I ovenize the reference?

Hello,

that depends on your stability needs and the individual T.C. near room temperature of your references.
You can also try a temperature compensation by NTC like in the Geller SVR-T reference.

The DIP8 epoxy package has after my experiences a humidity coefficient up to 0.5 ppm/%rH. (LT1027CN)
Lars sent me some measurements for AD587LN with up to 0.3ppm/%rH.
According to a MAXIM datasheet the SO8 packages may be less stable by a factor of 2-3.

With best regards

Andreas

I have 3 MAX6250 I bought a few years ago and never used
they are in DIP package so would probably be affected by ambient humidity too
do you know if anybody ever succeeded in solving/reducing the problem by enclosing the circuit with some sort of sealant?
thanks

[Gyro]

It's really not hard to make a hermetic enclosure. Some copper sheet (foil) or even PCB material. Hermetic feedthroughs - either feedthrough caps or glass-metal seals off ebay and fully solder all the seams. Fix a silica gell sachet inside if you want a dry, rather than constant humidity, environment inside. Just make sure you've done all your circuit tweaking first.

[ap]

There is a difference between a hermetic enclosure and a hemetic case for an electronic part. There are different mechanisms, one is e.g. the epoxy case 'moving', causing voltage changes, the other is the continuous invluence of remaining humidity causing chip changes on die/crystal level. Silica e.g. gets humidity down to say 20 or 30%, the remaining humidity inside, especially with temperature changes, is enough to have a continuous impact on the reference. There is no way arround a hermetic case on die level for a stable reference. A hermetic enclosure may make things a little better, but the same money is easier spent on the right part.

[tszaboo]

It's too bad the metal can references are getting scarce.

Well, TI sells the REF50xx series as a DIE option, so technically nobody is preventing us from buying that, and package it in a metal can. It even has a TEMP pin. One of my "virtual" projects is making an external heater with temperature controller for that chip. In the meantime, I have this:

Good enough. Unfortunetly it has a fatal flaw, so  I should be working on version 2 already.

[Andreas]

Hello,

and you think you can get a single DIE without buying a complete wafer?

On the other side: a REF5050 has typically a factor 5 more noise (3 ppm/V) than a comparable buried zener (0.6 ppm/V)
And there are still some of them available in hermetically package in single quantity. (AD586LQ/AD587UQ/LT1021CCH...)

with best regards

Andreas

[tszaboo]

Hello,

and you think you can get a single DIE without buying a complete wafer?

On the other side: a REF5050 has typically a factor 5 more noise (3 ppm/V) than a comparable buried zener (0.6 ppm/V)
And there are still some of them available in hermetically package in single quantity. (AD586LQ/AD587UQ/LT1021CCH...)

with best regards

Andreas

Probably not. It should be a group buy! 
I am aware of the issues with modern IC voltage references, I had my share of problems with it. Noise is a problem, if you cannot set the integration time high. At 100 NPLC, 10 Hz noise is not an issue.
The biggest issue I have with buried zener is the driving strength. You need to buffer it, if you want to use it as a reference for any ADC, and you will end up with similar noise than the IC references. And tempco is worse.
And I would write more, but it is almost 1 PM...

[pitagoras]

You can also try a temperature compensation by NTC like in the Geller SVR-T reference.

The link to the schematic for SRV-T is broken here
http://www.gellerlabs.com/Voltage%20References.htm

Does anyone have it at hand?

[lars]

You can also try a temperature compensation by NTC like in the Geller SVR-T reference.

The link to the schematic for SRV-T is broken here
http://www.gellerlabs.com/Voltage%20References.htm

Does anyone have it at hand?

Got a request offline so here is my copy. /Lars W

This schematic is based on the original SVR that I just added the NTC + series resistor to. I have done temperature compensations to many AD587, REF102, LT1021 and LT1236 and my preferred way today is to have a separate trimpot for temperature compensation with a NTC+ series resistor.

The AD587JQ from Digikey (or Mouser) is right now my preferred DIY10V ref as it isn´t humidity sensitive. The last years I have bought AD587JQ´s a couple of times and no one have been over 5ppm/C (probably been lucky but am not sure). So it should be quite easy to get them down below 0.5ppm/C with some temperature tests and NTC compensation.

As pin 5 on the AD587 is at about 7V I have had two 20 k trimpots in series between 0 and +10V (pin 6). The upper has been connected to the NTC (680+680k SMD near pin  5)) in series with a resistor of 1.5M. This gives a range of about +-10ppm/C adjustment if I remember correct. If the temperature coefficient is positive I sometimes calculates a higher value resistor to get some second order compensation, this also gives lower sensitivity to the components (TC, res, pot). The other pot is connected with a 10Mohm resistor to pin 5 to give about +-135ppm trim. A third resistor is connected between 0 or 10V to get the trim range for the pot.

This compensation scheme works well with the other IC´s mentioned but values will be different. If the trim pin is at +5V (eg LT1021+1236) don´t put the trimmers in series but instead in parallell. To set up your own calculation just hook up a 1Mohm resistor to the 0v respective 10v and see the change in ppm. Use these values together with say values for the NTC at 20-25-30C and put in a spreadsheet.

Another important thing that is missing in the schematic is a 100nf capacitor on the output (pin 6 to 0V). Without that the SVR is very EMI sensitive as I have seen a lot of times in my lab before I understood that. I also normally use ferrite beads to the connecting pins (in and out)

[Andreas]

Hello,

made some T.C. measurements.

First on a AD587UQ#1 date code 1631
I had bought 2 samples but unfortunately I killed one of them by wrong polarity.
So no statistics here.
T.C. is linear with 7.4 ppm/K


Then I decided to go for the JQ grade since lars reported having below 5 ppm/K.
Yesterday they arrived and the very first measurement looks promising.
AD587JQ#01 date code 1613
T.C. is near linear with -1.5 ppm/K
so a factor 5 better than the UQ -version near room temperature.

On the first measurement there is some drift. But this should stabilize after a few cycles.

With best regards

Andreas

[HighVoltage]

My most stable 10V reference at this time is an original Geller SVRT, about 5 years old and was turned ON most of the time.
The picture here shows a recent measurement over night with "cheap normal cables" and around 2 °C change over night.

Actually I am surprised how stable the SVRT really is.
When it is turned off for a few days it needs about 30 h to stabilize again.

[Andreas]

Actually I am surprised how stable the SVRT really is.
When it is turned off for a few days it needs about 30 h to stabilize again.

Hello,

it also depends a bit on luck how good the final cirquit is.
you can really get better results from a device that has < 1ppm/K without external compensation than from one with > 5ppm/K.
That´s what I´m looking for: a device like this sample.

AD587JQ #03 Datecode 1342

#03 has very low T.C. around 25 deg C.  -0.14 ppm/K

But unfortunately this device has some pop corn noise of up to 3uV measured on oscilloscope
 which is also visible in the T.C. measurement (e.g. near 25 deg C)

with best regards

Andreas

[orin]

Another important thing that is missing in the schematic is a 100nf capacitor on the output (pin 6 to 0V). Without that the SVR is very EMI sensitive as I have seen a lot of times in my lab before I understood that. I also normally use ferrite beads to the connecting pins (in and out)

This can show up when you reverse the leads at the DMM... From my discussion about it at the time: "To my surprise, I read -9.989XX (I don't recall the XX).  With the leads the normal way, I'm seeing 9.9998X with the X depending on temperature drift in the 'lab'".

After using twisted pair leads for both power and output to the DMM (an HP 3455A), I found a 4.7 nF film capacitor on the SVR output was sufficient to fix it in my case, but before the twisted pair leads, using standard mini-grabber leads, it took 0.33uF to make the 3455A read the same with the leads swapped!

I never really worked out why the AD587/HP 3455A did that - I had no problem with an HP 3456A.