A dual LM399 reference based on the 3456a U200 resistor network

[gmac34]

I wanted to make my own 1V and 10V references, and I started to check my parts bin for some goodies.

I have two Mettler scales boards with some rather old LM399s, they should have a lot of hours on them so hopefully quite stable, I also have a hp fineline resistor form the HP3456A, that I have bought a while back to repair my meter, this was not needed after all, so I’m stuck with it.

The design I proposed is based on other that were shared here, in particular from a board shared by JodelJonny

Two LM399s averaged, amplified to 10 V, with a 10:1 divider for a 1 V output. What makes it slightly unusual is that every ratio-critical resistor comes from the HP 1QD4-0044 "fineline" network (the U200 signal-conditioning array) rather than from discretes.

How the network is used:
* Two 40K halves with a centre tap do the averaging.
* 90K and 200K sharing a common pin set the gain of 1.4388, with a select-on-test resistor plus a 100 Ω trimmer if I really want 10.00000.

* For the 1V divider there are a few options, I have chosen the 67.4985K over 7501.5 for  a ratio of 8.9980, so 200 ppm off a decade, I think this needs trimming to be useful

Other choices:
* Zener bias is bootstrapped from the buffered 10 V through 3.01K (about 1 mA each), with 1 M startup resistors from +15 V.
* Heaters run from +15 V to −15 V rather than to ground.  Has anyone actually run LM399 heaters across a split supply? The datasheet permits it but I've not seen it done and I'd like to hear from someone who has.

* LTC2057 chopper for the gain stage, OPA202s for both Kelvin output buffers, the 10 V one with a BC817 emitter-follower booster and Vbe current limit so the load dissipation happens outside the op-amp package.
* TMP117 on a separate header supply so I can log board temperature against output.
* I have an external ±15 V linear supply, so nothing on-board regulates.

Any opinions and feedback is welcome

here the schematics (the layout is not done at this point, what you see is from JodelJonny)
https://kicanvas.org/?repo=https%3A%2F%2Fgithub.com%2FpignoniG%2FU200_dual_LM399%2Ftree%2Fmain%2FU200_dual_LM399

[Kleinstein]

I have used LM399 with a +-15 V supply. Quite some meters, e.g. HP34401 also do this. So no issue there. One may want a schottky diode from the negative supply side to GND to prevent it from doing too positive, e.g. during start up / turn off.

The 40 K resistors are relatively large for averaging, though still OK. Similar the gain resistors are on the high side for a LTC2057.
One could use the averaging resistors to allow at least some minimal filtering (e.g. a 100 nF film capacitor to ground).
I would prefer the standard 8 pin case of the LTC2057 to have to option to change to a different amplifier if needed.
For the start up I would prefer the 2057 powered from GND and +15 V and a resistor (some 2.5 K) from the 15 V to the OP-output to provide most of the current. This way the 2057 should start up without the extra 1 M resistors to the reference.


Not sure if it is really worth it using the HP resistor array: the resistor values are a bit on the high side an one would likely need trimminging for the 10 V gain anyway. The HP array would make more sense if more of the resistors could also be used for trimming (e.g. with jumpers, similar to the 3456 12 V ref. circuit). Otherwise the mix with array and external resistors (and especially trimmers) is not so good.
Trimmers are often not very stable, especially not the wiper resistance.
A 1:10 divider can be optained as a 3 S : 3 P array with 6 equal resistors from a standard array, possibly with a check option at 3 S : 3 S like in the classic Hammon divider. So not real need for the custom array there.

[guenthert]

There are still quite a few -hp- 3456A in use (they tend to offer a good price/performance ratio for those who care about stability).  Sooner or later someone might need such a spare part.