Vref driver circuit

[miro123]

Hello analog design engineers.
I've just finished the tests of one of my Vref  module/ PCB.
I am using the classical dual loop buffer circuit. Fast /High BW inner loop. and slow out loop. For outer loop i have chosen the classical inverting integrator. such solution has many advantages.
 - Allow Kelvin connection
 - integrator can be a zero drift opamp.
 - Output sink and source currents

Then I am confronted from other design approach - the DC servo. I have no experience with such circuit. I started with some simulation.
I have many questions
 - which circuit you will choose and why?
 - for which application you will choose the dual loop buffer and when servo DC.

I have attached the ltspice sim file . Note the sim file are not exact design files.
Ideas suggestions or remarks are welcome.

[miro123]

My input
I like the DC servo idea of injecting of zero drift integrator with 1:100 ratio - so it reduces the chopper signals at least 40dB.
However stability and flatness gives advantage on classical dual loop buffer.

[Jacques]

Then I am confronted from other design approach - the DC servo. I have no experience with such circuit. I started with some simulation.
I have many questions
 - which circuit you will choose and why?

DC servos are a particular popular solution in audio designs, so over at the diyaudio.com forums you will find many examples. I've attached one from an article by Bruno Putzeys in Linear Audio:



Edit: I also recently came across a composite op amp were the nested op amp was very fast (outer one was DC precision). If memory serves me correctly, it was implemented like this:

[Victorman222]

Good day. Some feedback

What is R4 and R8 for? Seems like they will worsen DC feedback depth.

If its a 10V input, both OPA189 or OPA140 can't be powered by +12V, since the allowed common mode input range is (V+)-3.5 V (OPA140) and (V+)-2.5V (OPA189).

The bottom circuit seems to be a composite amplifier where main purpose of U2 is to make it so U1 doesn't dissipate power.

It should be possible to make the DC servo stable and without peaking in frequency response by playing with opamp speed and output RC if that is what you mean by stability and flatness.

The chopper "scary thing" are current pulses at the inputs, not on the output. Check the ADA4523 datasheet current noise paragraph

Also check stability with load capacitance, since for a DC reference there is likely to be capacitance somewhere. Its not a bad idea to put some capacitance on the output too

What are your requirements for this circuit? I'd work from those and try the simplest circuit that would meet them

[miro123]

Good day. Some feedback

What is R4 and R8 for? Seems like they will worsen DC feedback depth.

If its a 10V input, both OPA189 or OPA140 can't be powered by +12V, since the allowed common mode input range is (V+)-3.5 V (OPA140) and (V+)-2.5V (OPA189).

The bottom circuit seems to be a composite amplifier where main purpose of U2 is to make it so U1 doesn't dissipate power.

It should be possible to make the DC servo stable and without peaking in frequency response by playing with opamp speed and output RC if that is what you mean by stability and flatness.

The chopper "scary thing" are current pulses at the inputs, not on the output. Check the ADA4523 datasheet current noise paragraph

Also check stability with load capacitance, since for a DC reference there is likely to be capacitance somewhere. Its not a bad idea to put some capacitance on the output too

What are your requirements for this circuit? I'd work from those and try the simplest circuit that would meet them

Thanks for the replay.
Maybe i was not clear from previous post this is only simulation not real design. Whit this in mind here is my answers.
 - R4 and R8 are used to sasisfy the simulator - to find initial dc point. 100M shoud not influence simulation. in real design they are not present.
 - +-12V normaly i use 7V but yes in real design I use +-15V for other reason.
For clarity, I will show the actual design used on my current Vref board.

Short explanation: The outputs of U8A and U8B are used as feedback/monitor  to assess the confidence interval of the voltage transfer. For example, the charge-injection errors of the Keithley DMM6500 and the two HP 3456A units are close to the noise level. The DMM6500 performs slightly better. The HP 34401A is not as good, but its performance is still acceptable. The Pico M2000 performs worse. Some flukes are bad too while Brymen 869s is beter than any fluke

Lessons learned:

I previously used Vrefs with a simple, classical buffer consisting of an op-amp and an NPN transistor. At that time, I did not know what I was actually measuring, so I was 100% confident.
Attaching handheld multimeter At U8B output and measuring AC and DC part give reasonable good result, but attaching of active  high speed probe and 12-bit scope is more fun.


The most important lesson is:

The less you know, the fewer problems you see. The more you know, the less confident you become.

My questions
 - which circuit you will choose and why?
 - for which application you will choose the dual loop buffer and when servo DC?
I dont need to tune existing circuit. I need good fundamental block  E.g. Jacque give me very valuable answer.

[miro123]

What is R4 and R8 for? Seems like they will worsen DC feedback depth.

R4=R8=100M are needed for Spice simulation, just to define the initial DC point of the circuit.

If its a 10V input, both OPA189 or OPA140 can't be powered by +12V, since the allowed common mode input range is (V+)-3.5 V (OPA140) and (V+)-2.5V (OPA189).

In AC simulation i use DC=0 - i make only the AC bode plot. In real circuit i use +-15V and 7V Vref

The bottom circuit seems to be a composite amplifier where main purpose of U2 is to make it so U1 doesn't dissipate power.

U2 is fast  to handle transcients,  while U1 is zero drift. I used this circuit in many SAR and SD ADC designs.

It should be possible to make the DC servo stable and without peaking in frequency response by playing with opamp speed and output RC if that is what you mean by stability and flatness.

Thanks
The chopper "scary thing" according to TI must be sloved with RC at output.
https://www.ti.com/lit/wp/sboa586a/sboa586a.pdf?ts=1785948200584&ref_url=https%253A%252F%252Fwww.ti.com%252F

What are your requirements for this circuit? I'd work from those and try the simplest circuit that would meet them

Good question. Unfortunately, I do not have the technical requirements.

There are many factors involved—not only capacitive loading or frequent charge injections. Transformer cross-coupling can also create plenty of phenomena. I still do not understand why some Fluke DMMs generate noise as well. They are battery-powered. The Brymen 869s performs better, but it is still not as good as high-end, A-brand DMMs. Anyway the disturbance created from those handhelds are orders of magnitude lower than their resolution  and specs.

From the user’s perspective, the requirement is simple :-) It must provide a confidence interval of less than 0.3 ppm at a confidence level of 99.6% with most DMMs.

Translating this simple requirement into technical specifications is another piece of cake. :-)