using a voltage follower in the feedback loop of a zero drift amplifier

[dutch66]

Hello everybody,
I am new here, so please forgive and redirect me in case the metrology section is not appropriate for my question.

While studying the schematics of the wonderful CERN open hardware voltmeter I came across a design topology that I am not familiar with and so far not fully understand. At several critical points in the design the designer chose to put a voltage follower into the feedback of a primary opamp. An example would be the circuit for the common mode bias voltage buffer.


The designer chose an ADA4522-1 zero drift type as primary opamp. Obviously due to its very low offset voltage drift and very low 0.1 Hz to 10 Hz noise. The secondary opamp is an AD8065 FET type with pA input bias current. I'm assuming that this was chosen to minimize the input bias of the combined arrangement, but I would like to understand this a bit better.

I hope that some of the forum members will be able (and willing ) to help me understand this in detail.

• Is my assumption that the secondary opamp is chosen to lower the input bias current of the circuit correct?
• Why would the second amplifier not increase the noise in the low frequency range?
• Can anyone point me to some explanation of such combination?

Thanks a lot for your help.

[Kleinstein]

The important feature of the 2nd OP is not the low bias.
The important point for the 2nd OP is the high speed.  It should be faster than the AZ OP to make it easy to get the loop stable, though the extra cap and resistor make it possible even get that without the high speed.

The circuit wants a low output impedance and fast recovery even at high frequency / short times (e.g. 100 ns scale).
So the 2nd OP is responsible for the higher speed part, the AZ OP is only for the low speed part (e.g. < 15 kHz).

There is  another possible use for the extra OP: with a little filtering between the OPs, they could also reduce the chopper artifacts from the AZ OP. So they may have missed out on that option or found it not to be worth it.

[KT88]

One important aspect of this arrangement of amplifiers, AKA composite amplifier, is high gain. Simply spoken you combine the gain bandwidth products of the two amplifiers. This gives you higher (closed loop-) bandwidth at higher gain basically.
The more relavant aspect for metrology is a much reduced gain error: Let's think of an amplifier that has a gain of 10⁶ you get 1uV offset error per 1V output. With two amps this offset becomes 1pV (given both amps have the same gain) In an integrator as an example this error contributes to nonlineariy as the 1uV adds to the input voltage.
Another advantage of such a composite amp is that the heat generated by the output stage (under load) does not affect the input stage.
Last but not least it allows to create an amplifier that can be optimized for both input- and output specs like fA input bias current and 100mA output current.

Cheers

Andreas

[dietert1]

In the past they used a transistor as output stage to take the power away from the precision buffer amplifier: A cool amplifier would drift less. The good old HP 3456A lab voltmeter already had that transistor buffer for its LM399 reference. With chopper amplifiers this no longer applies as they drift much less.
I used the transistor buffer for some 10 V references as output stage. That output stage is stable under capacitive load, see here: https://www.eevblog.com/forum/metrology/lm399-based-10-v-reference/msg3567571/#msg3567571.

Regards, Dieter

[RandallMcRee]

Two good replies--
Just to add to answer #1, the speed aspect is needed to drive ADC refin. Many LT app notes mention this.

See, for example, the ADA4523-1 datasheet which has this type of circuit on the front page. In that circuit, as Kleinstein mentions there is some low-pass filtering added, as well.
https://www.analog.com/media/en/technical-documentation/data-sheets/ada4523-1.pdf

AN21 from Jim Williams is a classic reference, see his last circuit.
https://www.analog.com/media/en/technical-documentation/application-notes/an21f.pdf

[dutch66]

Hi Kleinstein, hi Andreas, hi Dieter,

many thanks for putting me in the right direction. With my bias current assumption "war ich wohl auf dem Holzweg". I just modeled the combined amp and try to get a better understanding. I still have problems to understand the noise aspect.

Again many thanks und beste Gruesse nach Deutschland 

[dutch66]

Hi RandallMcRee,

thanks a lot! I was not aware of the application note on composite amplifiers. The work of Jim Williams is always worth reading.

Regards
Stefan

[Kleinstein]

The low frequency feedback is from the AZ OP. As a consequence the AZ also sets the lower frequency noise. As shown, the 2nd OP adds a little noise in the higher frequency range.
There is still some voidable higher frequency noise of the ADA4522 that could be reduced with a little extra filtering between the OPs (a first guess would be a RC filter with some 1 µs time constant). Only the transition region would see noise of both OPs. The circuit is still a bit tricky. It is a good idea to look at such a circuit in a simulation (e.g. with LTspice).

I think in this application, to drive the SD ADC the main point is the fast settling after current pulses drawn by the ADC.

The circuit in the ADA4523 datasheet is a bit different, with a much larger cap at the output.

[magic]

One important aspect of this arrangement of amplifiers, AKA composite amplifier, is high gain. Simply spoken you combine the gain bandwidth products of the two amplifiers. This gives you higher (closed loop-) bandwidth at higher gain basically.

Uhm, no, that's not what happens. The second chip is strictly a unity gain buffer. I'm inclined to guess like others that it was done for reduced output impedance at high frequencies, better load driving capability or taking heat off the precision opamp.

The more relavant aspect for metrology is a much reduced gain error: Let's think of an amplifier that has a gain of 10⁶ you get 1uV offset error per 1V output.

True, but auto-zero also tend to have >140dB DC gain without further tricks.

[maat]

I still have problems to understand the noise aspect.

Regarding the noise of an opamp. An op amp is an inherently noise device. Only by means of feedback it is able to suppress its own noise to the levels you see. So in essence, the first op amp is driving the second op amp in such a way, that it counteracts the noise of the second op amp. This is true within the bandwidth of the first op amp, beyond that the the second op amp is free to do whatever it likes to do.

[KT88]

@magic: although amp2 is a unity gain amplifier, it has to amplify current as well - it will show a load dependant offset. No big deal in many applications - but if one is concerned about sub-ppm accuracy, it matters.

Cheers

Andreas

[Kleinstein]

The load dependent offset of the 2nd OP would still be compensated by the 1st OP. With the high DC loop gain the AZ OP has not problem with this at low frequencies and DC.

The fast 2nd OP still has a lower closed loop output impedance than the open loop output impedance of the AZ OP. So the 2nd OP reduces the loading effect.

With the SD ADC the relevant loading is not DC but a pulsed load. A poor amplifier may here introduce nonlinear effects. The INL seen for the SD ADC chips can depend quite a bit on the amplifiers to drive the inputs and reference. In the higher frequency range the amplifiers may not be that linear anymore. E.g. the settling to a pulse load can depend on the load current or voltage. The GBW of an OP can change from positive to negative ouput current, as there are different parts of the output stage active.

[KT88]

The load dependent offset of the 2nd OP would still be compensated by the 1st OP. With the high DC loop gain the AZ OP has not problem with this at low frequencies and DC.

Yes, that's the point of having a composite amplifier. Athough in this case the second amp has a closed loop gain of one, under load the finite open loop gain is still affecting accuracy.
The high gain of AZ amps helps with another issue in a voltage follower amp (buffer), which is CMRR. An amplifier that is able to drive the load can be chosen without the need of highest CMRR. This refers to my comment of creating a composite amp that is optimized for both input- and output requirements.

Cheers

Andreas

[blackdog]

Hi,

The OpAmp that is set as 1x gain(buffer) is there to relieve the first OpAmp of the high load.
Because the first OpAmp sees no load you get the maximum performance.
The 1x OpAmp AD8065 is located within the loop of the OPA388, so its offset and bias specifications do not count.
What does help of the AD8065 is its relatively high output current it can deliver.
Allen this OpAmp is not suitable to drive with a 1 Ohm resistor at the output 10nF....

I am not happy with this setup, because in my opinion this circuitry setup is asking for instability and looks ill-conceived.
There is in my opinion no account taken of the phase margin which is reduced by adding an extra OpAmp in the loop of the OPA388 and the very heavy load with the 10nF capacitor.
And then also no resistor between the OPA388 output and the +input of the AD8065.
Also, almost certainly a compensation capacitor would have to be included between pin-6 and pin2 of the OPA388....

Maybe I have it all wrong, but I've built enough of these types of circuits to know that this one is not optimally stable.

Kind regarts,
Bram

[Castorp]

Pretty much everything was covered, there isn't much to add.

There's a good old Bob Pease reference on output loading and its effect on linearity. Here it is, formatted as a TI application report:
https://www.ti.com/lit/an/snoa471b/snoa471b.pdf

The circuit snipped shown in the first post is actually the only such composite amp in HPM7177 that doesn't drive an ADC pin. It's the 2.5 V common-mode buffer, a fully static voltage that doesn't seen any charge kickback. Still is has a somehow high 10 nF capacitive load. The output impedance at DC makes a little difference for the gain of the analog frontend.

[magic]

@magic: although amp2 is a unity gain amplifier, it has to amplify current as well - it will show a load dependant offset. No big deal in many applications - but if one is concerned about sub-ppm accuracy, it matters.

That's not open loop gain, it's output impedance.
To get 1V out, the first opamp still needs to output 1V.
In your example with 120dB OLG per chip, the offset error is still 1µV rather than 1pV.

[Kleinstein]

The 1x OpAmp AD8065 is located within the loop of the OPA388, so its offset and bias specifications do not count.
What does help of the AD8065 is its relatively high output current it can deliver.
Allen this OpAmp is not suitable to drive with a 1 Ohm resistor at the output 10nF....
....
And then also no resistor between the OPA388 output and the +input of the AD8065.

I am a bit confused about the circuit in question: the circuit shown uses an ADA4522, though an OPA388 would be suitable in the circuit too.

With the 2nd OP so much faster (the AD8065 is 145 MHz GBW) I see no real issue with stability, as there is extra local feedback to slow down the AZ OP. The version in the ADA4523 datasheet has extra lokal FB at the AD8065 to improve it's ability to drive the capacitve load.

A totally agree with the missing resistor between the AZ OPs output and the AD8065 - some filtering (even if only 100 Ohms and 100 pF) there could help to reduce the small switching spikes that AZ OPs typically produce.

[iMo]

Pretty much everything was covered, there isn't much to add.

There's a good old Bob Pease reference on output loading and its effect on linearity. Here it is, formatted as a TI application report:
https://www.ti.com/lit/an/snoa471b/snoa471b.pdf
..

Great paper!

..These were accomplished mostly with the use of symmetry, and not with the use of computers. That is because computers are not generally suitable for analyzing the heat flow among the millions of points inside a silicon die, not to mention the thousands of points in time, when a thermal transient occurs. ..

Is this still true?

PS: Thus even the OP07 gets pretty warm (ie the hottest two opamps - like 44degC on the picture of the DIY voltmeter in the 399 RefV) the issue is the local temp gradient between the two input transistors on the die..

[KT88]

That's not open loop gain, it's output impedance.

To get a very low output impedance in it requires high open loop gain. The output stage of a typical opamp usually has a couple of Ohms of output impedance...

In your example with 120dB OLG per chip, the offset error is still 1µV rather than 1pV.

I admit that my example was poorly chosen as I was referring to a composite amplifier that combines open loop gains with an over-all gain setting feedback and not to the circuit shown, with a buffer as a second stage. However the two gains still work together...

[magic]

That I can agree with, although the exact parameter that you really want from a buffer is not as much gain as transconductance

There's a good old Bob Pease reference on output loading and its effect on linearity. Here it is, formatted as a TI application report:
https://www.ti.com/lit/an/snoa471b/snoa471b.pdf

Who else found this excerpt funny?

The basic old LMC662 is the dual version of the LMC660, Texas Instruments first CMOS amplifier.

Clearly, somebody substituted all mentions of National with TI and didn't quite proof read the result

[mawyatt]

That AN by Bob Pease is one of the very best analog op amp ANs ever written!! All the old analog designers back then would just "brush" any non-linearity under the table of open-loop gain, the younger ones would run to Berkley SPICE for another "wrong" answer. Bob's AN showed you just can't do that in either case!!

Bob's use of the DUT itself as it's own high gain preamp is just brilliant, typical of Bob Pease and Jim Williams circuit design creativity

I vividly remember Solomon's IEEE paper and how it showed the 741 gain reverseing due to thermal feedback, this meant the + and - inputs reversed under thermal loads, also shown in Bob's AN

Back in early 90s we had developed primitive thermal models for bipolars that would run in SPICE, much later VBIC, MEXTRAM and HICUM bipolar models incorporated thermal effects, the later ones included dual thermal time constants.

Best,