DIY Precision Reference - How far one can go?

[Paulo Peres]

Hi all!
I was recently doing some experiments with voltage references and found a resolution limit to test it in my instruments.
My initial goal was investigate how far I could go with "simple" voltage references.
So, I used a REF5030, a integrated resistor network to divide it by 3 and a very low drift op. amp. (OPA4192) to bufferize the output. I added some extra circutry at output to make it more rugged, too. I also used the trimming scheme suggested in the REF datasheet (and trimed it to 1.0000V at output).
To provide thermal stability, I physically attached a small aluminum block to the main active components and controlled heated it to 40C (the famous LTZ1000 have this arrangement).
As result, I obtained a 1.0000V in my Fluke 8050A. There is a 200uV fluctuation that disappears as the Al block heats.
And here my doubts begins.
The voltage output stabilizes in the resolution limit of my instruments. Now I dont know what happens beyond it.
It will take a while until I get a voltmeter with more resolution, so I ask to more experimented colleagues: how far you reached doing this kind of arrangement?
I know I can reach very good resolutions with a LTZ1000, but my intention with this experiment was to know how far I can go WITHOUT a LTZ1000.
Any thoughts?

[Kleinstein]

One can go quite a bit lower than the limits of the Fluke 8050.
For a direct reading with a DMM the reading is always a combination of the external reference and the reference inside the meter and the meters amplifier/ADC.
As an alternative, at least for a limited frequency band, like the classic 0.1-10 Hz one can use an AC coupled amplifier and this way elimitate the reference.
There are some plans for such amplifiers around here in the forum.

Another option is to build 2 equal references and look at the difference with a DC coupled amplifier.

The specs for the ref 5030 call for some 3 µV/V of peak to peak noise for 0.1 to 10 Hz.  The OPAx192 also adds some noise and there are lower noise alternatives, like OP07 or OPA202  with less LF noise. The OPA4192 is aleady better than the reference specs, but the frequency dependence could be different.

[iMo]

The limits of your effort are
1. the ref5030's long term stability
2. the ref5030's smd epoxy packaging (ie. humidity inside the package) and the related hysteresis when switched off/on
3. the way of mounting of the ref5030 on the pcb
4. the long term stability of the resistive dividers
5. TC of the ref5030 and "ratio TC" of the resistive dividers
6. short and long term stability of your heater system (ie. temperature regulation stability, enclosure type, insulation effectiveness, etc)
7. etc.
Frankly, I doubt you will get better results than with a "standard" design with a 399, for example.

[GigaJoe]

basically as far as 10 microvolts resolution,  if you use a differential measurement between known  stable reference with the same voltage output as is yours made, as it basically the lowest resolution  of 8050.  as your lowest 200.00 mV range , this is basically difference limit ...

[Paulo Peres]

Another option is to build 2 equal references and look at the difference with a DC coupled amplifier.

I liked the idea.
It could give more precise idea of the influence of my heater.

[Paulo Peres]

The limits of your effort are
1. the ref5030's long term stability
2. the ref5030's smd epoxy packaging (ie. humidity inside the package) and the related hysteresis when switched off/on
3. the way of mounting of the ref5030 on the pcb
4. the long term stability of the resistive dividers
5. TC of the ref5030 and "ratio TC" of the resistive dividers
6. short and long term stability of your heater system (ie. temperature regulation stability, enclosure type, insulation effectiveness, etc)
7. etc.
Frankly, I doubt you will get better results than with a "standard" design with a 399, for example.

Good points.
You are probably right about the 399 and similar designs.
But the datasheet does not say to heat the REF5030. I just wondered what the result might be if I did it like some old designs when stable components weren't available. What could be the result with some already stable components?
As it will take a while to test it myself, I asked on the forum if anyone had already tried it.

[iMo]

When talking the reference itself, the issue number one is the epoxy package of the reference. My experience is it has none sense to spend your time with epoxy packages. All those epoxy references has not been developed for more than 12bit ADC/DAC applications, imho. That is 732uV (1LSB) resolution in your case.  When you look into the datasheet, you may see the "long term stability in 1k/2k hours" varies +/-150ppm (even 200ppm), that is +/-450uV in your case.
With an extreme care and luck you may create an 1V reference, which when the ref will be always on will show 1.00000V +/-1 in your last digit, longer term.
Edit: added a "0" not to be too pessimistic

[Kleinstein]

Heating the reference to a stable temperature can help. It can essentially eliminate the thermal effects and with the higher temperature the humidity is lower (about a factor 2 for every 10 K above ambinent). This can help against humidity effects on plastic cases, but it also means more settling effect after being powered off for some time as it takes days to weeks to get the humidity out.

For the 1 V ref. level the divider can well be the bigger issue than the ref chip. Startung with 3 V or 2.5 V one could use a resistor array (e.g. ORN, MORN, LT5400)  with 4 equal resistors for the divider. These give quite stable divider ratios at moderate costs.

[iMo]

The DIL and smd references (and most probably resistor arrays too) are sensitive to mechanical stress coming from the epoxy pcb as well. Thus the "dead bug" mounting style (with rather thin wires) may help as well.