Opamp + buffer, how to bypass buffer at HF for unity gain?

[Christer]

Consider the standard setup, noninverting op amp followed
by buffer, buffer inside the feedback loop. It is usually
recommended to close the feedback loop around the opamp
for HF, by a capacitor from the op amp output to the point
between the feedback and gain resistors, ie. the non-inverting
input. That is straightforward and makes sense. My question
then is, what if we have unity gain, that is the feedback is
a straight wire (well, not so straight, perhaps )? Surely the
capacitor won't hurt, but will it make much sense? The inductance
in the feedback path from the buffer might cause HF problems,
of course, which motivates the capacitor. I would suspect,
although I confess I have not tried to calculate on this, that
the inductance is usually so small that we would need quite a
large capacitor to get a crossover frequency about the same
as for reasonable values of feedback resistor in the other case.
A very large capacitor at the op amp output seems not a
good idea, especially since we buffered it in the first place to
avoid such problems.

The context of this question is that I will have said setup, but
the option to switch between unity gain and a higher gain
by shortcutting the feedback resistor with a jumper (for the
unity gain case, of course). Should I just calculate a suitable
capacitor for the high gain case, and then use that also for
the unity gain case, or should I have jumpers to also switch
in a larger capacitor, and if so how to choose the value?

[peranders]

If you plan to use BUF634 I image you just follow the instructions at the end of the datasheet.

You will get more trouble at low gain than high, so concentrate your efforts there.

I have tested OPA134, OPA627 and AD8610 together with a BUF634 with excellent results, but I don't mention where. I'll guess everybody know it by now?

[Christer]

In this case I will use a discrete diamond buffer. Walt Jung
recommends the capacitor, but it is obvious from the formula
he states that he assumes a non-zero feedback resisitor, and
does not say anything about the unity gain case.

One might, perhaps, go by the recommendations for BUF634,
but they seem to suggest the capacitor is usually not necessary,
which may apply for compact layout. For certain reasons my
feedback paths will not be so straight as one would like, and
will pass at least one switch. I do thus assume a capacitor
is very wise in my case, in the high gain case. The problem
with uníty gain case is not really mentioned in the datasheet
for BUF634. It is very cook-bookish, suggesting a value of
200pF for certain opamps, not even suggesting the value
be adjusted according to feedback resistor value!! Besides
their stability discussions seems also to assume one does not
isolate the feedback path from the load with a resistor as
Jung suggests. It seems to me the recommendations in the
BUF634 datasheet are too simplified to be very reliable even for the BUF634.

[ppl]

Maybe this is what you had in mind. SW1 allows the gain to be changed from 2 to a higher value and still retain your compensation capacitor. if you just want to compensate the Buffer only you can place a small cap from the output to the input of the buffer. See the Circuit at the top of pg. 10 of this app note. thay call it the capacitor bootsrap method.
http://www.intersil.com/data/an/an548.pdf

[Christer]

PPL,

It is similar to what I have in mind, but for certain reasons I
want unity gain, not a gain of 2, so I can see no other solution
than shorting the feedback resistor. In the worst case, I could
consider using a gain of 2 and a voltage divider at the input,
but I would like to avoid that.

I looked at that linked document. Maybe I am missing something,
but I don't quite see how it solves the problem. I suppose,
without having really thought about it, that it is intended to
compensate for phase shift in the buffer. In my case, I am more
worried about my not so straight feedback path, and inductance
in that, which may be harder to compensate for.

[jcx]

using PPL's drawing above, why not elimiante R2, with S1 open you have unity gain and don't have to switch R3

R3*C1 form an integrator with respect to the ouput, the idea is to have the integral time constant below the frequency that you get excessive phase shift from the buffer+load (especially useful with some modern op amps which are designed to have only 30 degree phase margin to push up the advertized bandwidth in "unity gain stable" mode)

[peranders]

Christer, this capacitor value is important only if you want the most out of the amp+buffer combination at very high frequencies.

I have tested without this capacitor and it still works. This capacitor needs to be trimmed only if the opamp is too fast for the buffer.

[Christer]

Thanks jcx,

Of course that's the way to do it. Stupid of me not to think of
that. That's even the standard way to do unity gain feedback.
Must be some mental blocking that remains from some earlier
phase in the design when things looked different. Or maybe
it is just that I have a bad head ache today? Or maybe I am
just trying to blame my stupidity on someting?

[Christer]

Originally posted by peranders
I have tested without this capacitor and it still works.

On your PCB and with your combination of op amp and buffer
yes.

This capacitor needs to be trimmed only if the opamp is too fast for the buffer.

And what if it is?


Anyway, Jcx seems to have solved it for me.

[ppl]

Thanks John for the note regarding making my circuit unity gain.

Christer> It is true the compensation method described in the app note I linked to is to compensate the Buffer only. If you noticed the illustration circuit it was used in had the Buffer driving a Converter input. This is almost a purely capacitive load with an almost non existent parallel resistance. Emitter followers do not like this very well and can peak or become unstable driving purely capacitive loads. Several methods are used to deal with this, with the most common being adding a small resistor in series with the output of the Buffer to isolate the load capacitance. This method has the drawback of higher output impedance. An alternative method I use a lot is to put a ferrite bead in parallel with this output isolating resistor so as to bring down the output impedance at lower frequencies while still letting the impedance rise at higher frequencies to maintain capacitive load isolation. This is similar to the R/C network on the output of loudspeaker Amps. Also in addition to C1 you can put another small cap across R3 to compensate the overall loop including the Buffer. Finally you could additionally add what I call Noise gain compensation. This is a series RC/ network connected from the Inverting input to the non inverting input of the op amp. I must point out that noise gain compensation requires at least C1 in the first schematic illustration I posted. I don’t think you will have to go that far however and Noise gain compensation can get real tricky under some situations.

The Attached Image is of the AD825 op amp. notice the R/C network on the output stage. this is similer to the capacitor placed between the Buffers output to input as depicted in the app note i linked to in my first reply.