Need Help with an LM399 Reference with Selectable Output Voltage

[Chance92]

Hi! I would like to build a 1mV to 10V voltage reference based on LM399. The circuit is based on fmaimon's design in the LM399 thread. I made some changes to make it a bit more versatile:

1. Added a voltage divider and a buffer to get voltages from 1mV to 10V. (I calculated the output resistance of the divider to be around 10 ohms, so I put a 10 ohm resistor on the buffer feedback. Is this resistor critical?)
2. Added a connector for battery connection.
3. Changed LM317 to 7815 to simplify the design a little bit. (Is this a good idea?)
4. Used a dedicated 7815 to power the heater in order to reduce the effect of battery voltage drop. (Is this necessary for maintaining stability?)
5. Added some input protection.

I have some used LM399s date back to 1989. Are they a better choice comparing to brand new LM399A?

I am thinking about putting the entire thing into a thermos flask. Is this a good idea?

Can I use this reference to test the stability of a 7.5 digit DMM? How can I improve the stability?

Suggestions and criticism are welcome.

[TiN]

C6 should be C0G or film, not X7R. I'd add bigger bulk cap in parallel with C5 (e.g. 100-1000uF 25V/35V).
Also some smaller bulk caps at outputs of LDOs might be nice to have (22-100uF). Check LDO datasheet for recommended/max capacitance.

You need to route heater ground separate from signal ground. Also you should modify schematics to reflect star mecca ground for all signal nodes, this is important. Your connection of zener cathode to heater pin suggests that you missed this important point.

I'd get rid of switch (as unreliable contributor in signal path) and instead have multiple individual output taps (with opamp buffers if you have budget).

Why AD706 chosen?

Also consider new ADR1399, it has noise similar to that of LTZ1000 at fraction of cost/complexity.

There are no 0.2ppm/K resistors, suggest to specify real spec, not typical specmanship .

Flask is decent idea. Make sure to protect circuit well from air drafts on both sides.

After a year of ageing with constant power and careful calibration and some history data logging you'd be able to use such reference to calibrate any 7.5-digit DMM on fullscale for base 10V and maybe lower ranges. Stability and performance of your divider chain R6-R10 would be biggest error contributor in this design. You might want to consider specially designed resistor network, likes the ones Caddock makes.

[PartialDischarge]

The offset of the AD706 is too high, not good.
The AD706 needs bipolar supply to go down to 1mV, even 1V output. For most op-amps you will need some negative rail for outputs near 100mV.
I'm missing some kind of "smallish" trimming capability
Don't use that switch, it adds too much resistance.


For an application were I need 10-100mV with at least 1% accuracy I'm building this circuit, it uses 2 gold plated pin headers for each switch range:

[Andreas]

Hello,

I personally would also use more beefy batteries than PP3 blocks.
A alkaline battery with ~200 mAh will last not long especially as the voltage quickly drops from 9V to 8V within the first 10% of discharge.
So with 20-25 mA for the heater and 5 mA for each voltage regulator + some current for the buffers the voltage will fall out of regulation before the LM399 is heated to full precision. (~15-30 minutes).

with best regards

Andreas

[iMo]

I would add an output transistor (or better two transistors for the current limiting) at the output of the U5..
PS: see below an example - it assumes your opamp is rail to rail.. You may put that stage at the U4 as well (to de-load the U4).. The max output current aprox 30mA..

[Zero999]

R10 should be 11.1111 for 1mV out with 10V in. Use 12R and 150R in parallel, with the 12R being more critical.

The output impedance of the divider depends on the tapping point. R11 isn't doing anything. The bias currents of the AD706 are low enough not to matter.

[iMo]

..another option would be to mimic the "PentaRef" voltage source.
399 Vref and an 18-20bit SPI dac plus an output opamp.
Btw, do we have schematics of the pentaref somewhere (analog part only..)?

[Chance92]

C6 should be C0G or film, not X7R. I'd add bigger bulk cap in parallel with C5 (e.g. 100-1000uF 25V/35V).
Also some smaller bulk caps at outputs of LDOs might be nice to have (22-100uF). Check LDO datasheet for recommended/max capacitance.

Shall I use ceramic or electrolytic?

You need to route heater ground separate from signal ground. Also you should modify schematics to reflect star mecca ground for all signal nodes, this is important. Your connection of zener cathode to heater pin suggests that you missed this important point.

Do you mean I should create a star point for all signal nodes and join it with the heater ground at the input? Should I avoid using a ground plane in this case? I have heard about separating grounds but I am not familiar with exact procedures.

Why AD706 chosen?

Originally fmaimon used OPA177GP, but later on codeboy2k did some calculation to show that AD706 is more temperature stable (https://www.eevblog.com/forum/metrology/lm399-based-10-v-reference/msg219394/#msg219394). I really don't know how to choose an op-amp. There are so many of them. Do you have any recommendations for my case?

Also consider new ADR1399, it has noise similar to that of LTZ1000 at fraction of cost/complexity.

Sure, if I can get my hands on them, which doesn't seem possible at the moment.

There are no 0.2ppm/K resistors, suggest to specify real spec, not typical specmanship .

That's the typical TCR of the ultra high precision Z foil resistor by Vishay:
https://docs.rs-online.com/3cdd/0900766b80d9a540.pdf
But now I am a bit reluctant to use them since I just found out they cost £55 each!

After a year of ageing with constant power and careful calibration and some history data logging you'd be able to use such reference to calibrate any 7.5-digit DMM on fullscale for base 10V and maybe lower ranges. Stability and performance of your divider chain R6-R10 would be biggest error contributor in this design. You might want to consider specially designed resistor network, likes the ones Caddock makes.

I didn't find the ratio I need, but I didn't look very hard. I'll try again. If I am to use individual resistors, do you think if it is necessary to find a way to keep them at the same temperature? How about mounting all of them on a circular copper block with insulation on the outside? Do I need to worry about the difference of thermal loss produced by the resistors?

[Chance92]

The offset of the AD706 is too high, not good.
The AD706 needs bipolar supply to go down to 1mV, even 1V output. For most op-amps you will need some negative rail for outputs near 100mV.
I'm missing some kind of "smallish" trimming capability
Don't use that switch, it adds too much resistance.


For an application were I need 10-100mV with at least 1% accuracy I'm building this circuit, it uses 2 gold plated pin headers for each switch range:

I didn't know that. My knowledge on op-amps is very limited. Do you have any recommendations for op-amp?
I want to build the reference with maximum stability. I am planning to calibrate the reference afterwards and use those voltages as they are, so that I can avoid the instability introduced by the trimmers. What type of trimmers are you using?

[Chance92]

Hello,

I personally would also use more beefy batteries than PP3 blocks.
A alkaline battery with ~200 mAh will last not long especially as the voltage quickly drops from 9V to 8V within the first 10% of discharge.
So with 20-25 mA for the heater and 5 mA for each voltage regulator + some current for the buffers the voltage will fall out of regulation before the LM399 is heated to full precision. (~15-30 minutes).

with best regards

Andreas

Hi Andreas. I am worried about that as well. I am thinking about put more batteries in parallel or in series. Apparently none of these two options are perfect. Putting them in parallel still can't harness most of the energy. Putting them in series will probably cause more energy waste and temperature rise on the regulators. I was also considering using lithium batteries, but that would require additional charging circuit. Do you have any suggestions? Shall I go for something like CR123A?

[Chance92]

I would add an output transistor (or better two transistors for the current limiting) at the output of the U5..
PS: see below an example - it assumes your opamp is rail to rail.. You may put that stage at the U4 as well (to de-load the U4).. The max output current aprox 30mA..

Thanks! I need to think about how much current I need.

[PartialDischarge]

I didn't know that. My knowledge on op-amps is very limited. Do you have any recommendations for op-amp?
I want to build the reference with maximum stability. I am planning to calibrate the reference afterwards and use those voltages as they are, so that I can avoid the instability introduced by the trimmers. What type of trimmers are you using?

For the output opamp I can recommend an OPA189 for the low temp drift, low offset and low input bias, but there are many others.
I use multi-turn pots but my application is not very stringent, and anyway it is configured only to have a small adjusting range.

[Chance92]

R10 should be 11.1111 for 1mV out with 10V in. Use 12R and 150R in parallel, with the 12R being more critical.

The output impedance of the divider depends on the tapping point. R11 isn't doing anything. The bias currents of the AD706 are low enough not to matter.

I was having trouble finding resistors below 1 ohm with TCR. Paralleling them is a good idea. I will try to find the suitable resistors.

[Kleinstein]

The staility of the LM399 and ADR1399 is limited. So the AD706 would be good enough and no real need for 0.2 ppm/K or similar grade resiistors.
For the resistors one could consider a ready made resistor array (e.g. LT5400, maybe MORN ) to get a good ratio stability at moderate costs.
For the divider to very low values one can use a 2 stage divider to get a low voltage without needing 10 ohms or lower. So like 10 V to 1 V and 100 mV and than 100 mV to 10 mV as a an extra divider.  Ideally the divider would start from the 7 V and not the 10 V, so that the lower votlages avoid the 7 to 10 V gain step.

Other alternative OPs would be the OPA202, OPA207 and LTC2057, maybe MCP6V51.

The ADR1399 alterntive would like an extra RC for damping and the LM399 would be also be Ok with this.

Ideally there would be some extra measures to allow a capacitive load at the output - many OP-amps have a limited capacitor to drive capacitive loads.

[Chance92]

For the divider to very low values one can use a 2 stage divider to get a low voltage without needing 10 ohms or lower. So like 10 V to 1 V and 100 mV and than 100 mV to 10 mV as a an extra divider.  Ideally the divider would start from the 7 V and not the 10 V, so that the lower votlages avoid the 7 to 10 V gain step.

Thanks for your suggestions. May I ask how I should put the two dividers together exactly? Do I put a buffer between them or do I directly put the second divider in parallel with part of the first divider?

The ADR1399 alterntive would like an extra RC for damping and the LM399 would be also be Ok with this.

Do you mean I should put a low-pass filter across the LM399? What cutoff frequency should I aim at?

[iMo]

Here is the 1399 filter, the same for 399 (from the DS)..

[Chance92]

Here is the 1399 filter, the same for 399 (from the DS)..

Thanks! I'll add it to my design.