Some general clarifications about a reference design based on LM723 I'm working

[Verdefluox]

NOTE: Question has been moved here from "Project, Design and Technical Stuff" since the nature of the question may be more metrology related. Hope I didn't do something wrong in the while (first time using this forum)

Hi folks,
I had been reading around on this forum for some time, and this is the first time I write a post, hope to do everything right.
Since some time I started to be interested in voltage references (I know that technically this post may fit in the Metrology section, but here my problem lies mostly on how to design a certain part which should be more general, hence I preferred posting it here). I'm quite a newbie
Some time ago I found these two "references" (not that high end, since it's based on something that is not mainly a proper voltage reference, but I'm starting just now in the field, so I think it may still be good for me. Performances I'd like to achieve are below):
- Elektor 723 Voltage Reference, from July 1978 (project no. 99 on this paper) https://www.google.com/url?sa=t&rct=j&q=&esrc=s&source=web&cd=&ved=2ahUKEwiUnYXw8of4AhUj_rsIHZeVDm0QFnoECAIQAQ&url=https%3A%2F%2Fworldradiohistory.com%2FUK%2FElektor%2F70s%2FElektor-1978-07-08.pdf&usg=AOvVaw0T7Njj0aodbUGTjAGLrvrd
- This website, in which there is another approach https://electronicprojectsforfun.wordpress.com/silly-circuits/silly-circuits-a-heated-lm723-reference/

The project
Generate that voltage (around 7V), buffer it through an op amp (OP07 should be okay for the job, I hope) and then divide it, in order to feed it into an ADS1115 (can handle max. around 6.1 V with a 2/3 gain, hence I have to use some divider for that) and monitor it through Arduino. The contour in this project is also to learn something about Python and Grafana (or some sort of software like that) in order to post process some results, but that's just a contour and explained it just to give a context. Since it's just a starting project about "precise measurements" and related stuff, I'd like to keep it as low cost as possible.

Requirements
- Manage to have readings down to millivolts (don't mind going further), hence the ADS1115 should be sufficient, with a resolution of 125 uV @ Gain = 1
- Manage to implement a correct filtering and buffer, which hopefully reduces the noise on the system and smoothen the output a little bit (unless somehow I'm doing something wrong and actually ruining the measure and/or increasing the noise, which I shouldn't, according to OP07 datasheet)

My approach
What I came up with is the following:


As a first note, on the future PCB there will be also be a regulator based on the LM350 (I have some of them lying around which I never use, and are not drawn in this schematic since this part is completely done as per datasheet indications). This because the only generator I have is an adjustable one I built and is not that much great in terms of regulation (and since here I need a fixed voltage but with relatively good regulation, I went with an on-board one). The idea is to power it with my generator, but regulated much better with the on-board one. I know this isn't optimal, but it's basically the only equipment I have.

The reference part is pretty straightforward with the Elektor project. I added a simple low pass filter in order as most high frequency AC as possible (cutoff at 1.85 Hz) using a carbon 1% resistor and a polypropylene capacitor. I'm not really sure about the buffer, however. My idea was to have a voltage follower, with a divider in order to have around 3.5V, which is the only way that comes into my mind that can be treated by the ADS1115. Note that all my following reasoning is based on this design, but I'm open to better configurations if there are any. So, here are my questions:

About the voltage follower
From the 2nd project linked above I decided to give it a try to the OP07, which, according to the datasheet (linked below) it has a low offset (max 75uV, below my requirements) and offset correction pins, which in this case should be enough, and noise shouldn't be a problem. I don't know if there are any better op configurations which could be used here for a buffer in order to feed the divider (different from the follower, maybe more used in metrology field). I've obviously used op amps in other projects, but this is quite the first time I'm trying to consider every possible behaviour, so sorry if I miss some point.
On a side note, according to the datasheet I see that they recommend connecting the central pin of the pot to VCC. Doing some general testings (done with 3.3V, and checked the two inputs with two ADS1115 pins), I actually managed to reduce the offset by connecting the pot central pin to GND instead. According to the following document from TI, from what I can understand they connect the middle pin to GND, so I don't really get why they recommend VCC in the OP07 datasheet. Is this something that depends on the opamp, or am I missing something?
https://www.google.com/url?sa=t&rct=j&q=&esrc=s&source=web&cd=&ved=2ahUKEwiM7KCln4j4AhXE7rsIHTlRDt0QFnoECAYQAQ&url=https%3A%2F%2Fwww.ti.com%2Fgeneral%2Fdocs%2Flit%2Fgetliterature.tsp%3FbaseLiteratureNumber%3Dsloa045&usg=AOvVaw2N6EOfOGY7akq1COnSqZcr

About the voltage divider
Unfortunately, I don't have any precision resistors lying around (the best I have are 1% metallic resistors, and only 10k). I don't know how much that helps, but I thought about this: since my 10k are 1%, this means that this divider in the "worst case" (maximum difference between the two) will be 10100 ohms and 9900 ohms. Supposing the output is 7V, I can have an output voltage on the divider between 3.535V and 3.465V (so, undertainity in the 10s of mV). With some 0.1% resistors (which I would have to buy), I will go between 3.5035V, so I would still be in the mV uncertainity. Note that obviously I know about the 3.5etc V; what I don't know, and that I would like to, is about the 7V above.

How I thought to solve the divider problem
Now, I know that I can simply analyze the 3.5V and see how is the behaviour of that (aka. having a sort of "reference" at 3.5V), but I'd like to reconstruct the full 7V considering the variability of the resistors. For this purpouse, I thought to mount the 10k first and add various pins both on the output and on the divider, in order to "test" the resistors with some known voltage, like the 5V from Arduino. I know that even those 5V aren't perfect, but since in this case I can both recover 5V and 2.5V with the ADS1115 with sufficient precision, I could theoretically easily recover the ratio of the two outputs, and hence the ratio of the resistors. I don't know if this is an overkill, a wrong procedure or something else, but isn't this conceptually right? I know that there are more in-depth considerations (like tempco), but as a starting point, a proper PCB design (keep the divider distant from other heat sources, like in this case the 723 itself, and keep them close to have the most uniform temperature) should be enough, right?
I honestly can't think of any other alternative for now, since for this I thought that maybe the simplest approach (assuming that I can somehow solve its uncertainities) is just enough, but I'm open to alternatives of there are any.

COMPONENTS DATASHEETS
- ADS1115 https://www.google.com/url?sa=t&rct=j&q=&esrc=s&source=web&cd=&cad=rja&uact=8&ved=2ahUKEwjU9rSIr4j4AhWfQvEDHTc3BOYQFnoECAQQAQ&url=https%3A%2F%2Fwww.ti.com%2Flit%2Fgpn%2Fads1114&usg=AOvVaw1uY1d0MAXekJgc17vRTqG6
- OP07 https://www.google.com/url?sa=t&rct=j&q=&esrc=s&source=web&cd=&cad=rja&uact=8&ved=2ahUKEwiTnvSVnYj4AhVmhf0HHW6hBeYQFnoECAUQAQ&url=https%3A%2F%2Fwww.analog.com%2Fmedia%2Fen%2Ftechnical-documentation%2Fdata-sheets%2Fop07.pdf&usg=AOvVaw1Hrllw_v2B6WbReKi_KBjS

[RoGeorge]

Nice project! 

There are plenty of NOS 723 from decades ago, so most of the age drift of voltage must have been settled by now.

From the Elektor article:

Obviously the chip temperature must
be maintained considerably higher
than ambient for correct operation
(since the circuit cannot cool the
chip to a temperature below ambient

Back then, heating was the only practical solution.  Nowadays there are cheap Peltier elements that can either heat or cool.  I wonder how much performance improvement can get by cooling for thermal stabilization, instead of heating.

With lower temperature, lower noise would be expected, though no idea if it will worth or not to do that in practice.

[Verdefluox]

There are plenty of NOS 723 from decades ago, so most of the age drift of voltage must have been settled by now.

Even though I don't understand completely what this implies, I suppose that this means that the zener (which for reference is a RM723 -should be from Raytheon- in a TO-100 package) has reached a stable state on its internal reference. Don't know if I'm right, though. Thank you for your opinion, however.
Main problem here is mostly on the final divider, since I don't really know if my procedure in doing the "exact" calculations is valid or if I'm missing a big point. I am aware that my method isn't optimal, but for now this is all I got. I'd like to ask if in your impression I'm losing my time doing this, and then a prolonged power on will completely change the situation. I know these are just simple resistors in the end, but I'm starting to understand that on mV level and below there are more in-depth considerations, like drift and such (or maybe they start at a level even lower -uV maybe?-, so I don't have to worry about that?).

P.S. Maybe I should post on Metrology section for a more detailed answer?

[RoGeorge]

has reached a stable state on its internal reference. Don't know if I'm right, though.
...
P.S. Maybe I should post on Metrology section for a more detailed answer?

Yes, that's what I meant, but I don't have much practice about voltage references.  Seen forced aging for other voltage references, but can't predict how much this will affect the results for a 723 voltage ref.

Duplicate posts are not allowed.  You can move the topic to the metrology section if you want to.  There is a button with a drop down menu at the bottom of the webpage, available for own started topics.

[iMo]

The schematics above is the "self heating" version of the 723 ref source, we experimented here some time back:

https://www.eevblog.com/forum/projects/playing-with-723-voltage-regulator-oldtimer/msg1782875/#msg1782875

I think the 723 is not suitable as the ref source for a 16bit ADC, anyway..