OPA197 c-load stability

On Sat, 9 Apr 2022 16:47:43 -0700 (PDT), Rich S
<richsulinengineer@gmail.com> wrote:

>On Saturday, April 9, 2022 at 5:17:38 AM UTC, bill....@ieee.org wrote:
>> On Saturday, April 9, 2022 at 2:39:42 PM UTC+10, jla...@highlandsniptechnology.com wrote: 
>> > OPA197 is a great little opamp. 36 volts RRIO, 10 MHz, pA bias 
>> > current, pretty good offset specs, 5 nV noise, EMI hardened. 
>> > 
>> > Like most opamps, it is specified to be stable up to some capacitive 
>> > load, 1 nF in this case. For bigger caps they show the usual R+C load 
>> > stabilizing idea on the data sheet. That all ignores the Williams 
>> > Effect, namely that a big enough cap will stabilize most anything.
>> It doesn't. A big enough capacitor means that the oscillation isn't driving enough current into the capacitor to produce enough voltage swing to be detectable - or sometimes not even even enough to be bigger than the Johnson noise at the oscillation frequency in the series resistance of the capacitor. At that level the oscillation doesn't take the input stage out of its linear region (+/25mV, for bipolar transistors, a volts or so for FET and MOSFET inputs), so it doesn't mess up performance enough to notice, but it is still oscillating.
>> > As a follower, handy for rail splitting and such, 
>> > 
>> > 47 nF to ground oscillates 
>> > 
>> > 1 uF ceramic rings a bit on step edges 
>> > 
>> > 4.7 uF cer or more is stable 
>> > 
>> > 56u or 180u polymer is stable 
>> > 
>> > Any tantalum cap is stable 
>> > 
>> > A 33u tantalum and a lot of ceramics looks nice. That may be the 
>> > choice for a lot of opamps. 
>> > 
>> > https://www.dropbox.com/s/bsh4i0yb2yxn2oo/Z534_1.jpg?raw=1 
>> > 
>> > https://www.dropbox.com/s/psu1qsqwsb6wi6g/20220408_120126.jpg?raw=1 
>> > 
>> > https://www.dropbox.com/s/6ukbwh9kj9pdl6i/20220408_112325.jpg?raw=1 
>> > 
>> > https://www.dropbox.com/s/s7whu9d7gszm2im/20220408_114102.jpg?raw=1 
>> > 
>> > I have a case where I want to drive a many-bypassed rail at VCC/2 and 
>> > don't want a resistor in series with the output.
>> Pity about that. 
>> 
>> https://www.analog.com/en/analog-dialogue/articles/ask-the-applications-engineer-25.html 
>> 
>> does discuss what's actually going and goes into sensible ways of tackling the problem. 
>> 
>> -- 
>> Bill Sloman, Sydney
>
>thanks for that link Bill. Although I've read the same info elsewhere
>this one also mentioned *external* compensation is useful. 
>And I thought external comp was just for "old" op-amps, a
> previous-century idea!  ;0)

Note in the opamp table above some of the amps have "unlim" capacitive
load drive capability.

In most opamps, there is a buried internal compensation pole, and
adding a cap load creates a second pole in the loop, causing
instability.

In some opamps, adding a c-load just slows the open-loop response but
doesn't add another pole. So it gets more stable, not less.

OPA197 has a number of patented features, but the data sheet doesn't
name the patents, so it's not obvious what the internal circuits are.



-- 

I yam what I yam - Popeye

On Sat, 9 Apr 2022 19:21:55 -0400, Phil Hobbs
<pcdhSpamMeSenseless@electrooptical.net> wrote:

>jlarkin@highlandsniptechnology.com wrote:
>> On Sat, 9 Apr 2022 13:44:47 -0400, Phil Hobbs
>> <pcdhSpamMeSenseless@electrooptical.net> wrote:
>> 
>>> On 4/9/22 12:39 AM, jlarkin@highlandsniptechnology.com wrote:
>>>> OPA197 is a great little opamp. 36 volts RRIO, 10 MHz, pA bias
>>>> current, pretty good offset specs, 5 nV noise, EMI hardened.
>>>>
>>>> Like most opamps, it is specified to be stable up to some capacitive
>>>> load, 1 nF in this case. For bigger caps they show the usual R+C load
>>>> stabilizing idea on the data sheet. That all ignores the Williams
>>>> Effect, namely that a big enough cap will stabilize most anything.
>>>>
>>>> As a follower, handy for rail splitting and such,
>>>>
>>>> 47 nF to ground oscillates
>>>>
>>>> 1 uF ceramic rings a bit on step edges
>>>>
>>>> 4.7 uF cer or more is stable
>>>>
>>>> 56u or 180u polymer is stable
>>>>
>>>> Any tantalum cap is stable
>>>>
>>>> A 33u tantalum and a lot of ceramics looks nice. That may be the
>>>> choice for a lot of opamps.
>>>>
>>>>
>>>> https://www.dropbox.com/s/bsh4i0yb2yxn2oo/Z534_1.jpg?raw=1
>>>>
>>>> https://www.dropbox.com/s/psu1qsqwsb6wi6g/20220408_120126.jpg?raw=1
>>>>
>>>> https://www.dropbox.com/s/6ukbwh9kj9pdl6i/20220408_112325.jpg?raw=1
>>>>
>>>> https://www.dropbox.com/s/s7whu9d7gszm2im/20220408_114102.jpg?raw=1
>>>>
>>>> I have a case where I want to drive a many-bypassed rail at VCC/2 and
>>>> don't want a resistor in series with the output.
>>>
>>> Sort of the reverse of Schawlow's law: "Anything will lase if you hit it
>>> hard enough." ;)
>>>
>>> It's worth putting a sense resistor in the supply leads to check for
>>> oscillations of very low amplitude.  THat's been known to happen even
>>> when the output looks steady on a scope.
>
>> Supply current looks OK.
>> 
>> I need to drive a net to Vcc/2, and it has a dozen 10 uF ceramics to
>> ground, and I want a low impedance drive from DC up.
>> 
>> Looks like the added 47 uF tantalum is prudent. That adds some ESR
>> damping.
>> 
>> My boss assigned me to rev this board
>> 
>> https://www.dropbox.com/s/dnkmpdzs2va6x3z/P545_Top.jpg?raw=1
>> 
>> which involves picking up a bunch of ECOs and reviewing the NEXT file,
>> where people have accumulated two pages of annoying change requests.
>> 
>> The bottom of the board is paved with parts; there's not much room to
>> add things.
>> 
>> A previous fix hung a 1500 uF aluminum cap on the rail, an ugly hack
>> on top.
>
>I recall--the lead sneaking round the edge of the board was thrillingly 
>gnarly. ;)  (Not that I'm above doing the same, when pressed sufficiently.)
>
>The C load moves the output pole to lower frequency, and when it's too 
>close to the zero-cross of the main+tail poles, you wind up with 
>instability.
>
>A large, higher-ESR cap is often a good way to stabilize switchers and 
>LDOs, too--it's a shunt version of the usual lead/lag network used in 
>feedback amps.  There's no reason that should be a problem in an op amp 
>loop, in principle.  Doing stuff outside the datasheet's guaranteed 
>limits puts the responsibility on us, but oh, well--that's where it 
>winds up anyway.
>
>Cheers
>
>Phil Hobbs

One other problem is the 12 heavy transformers. A sufficiently
aerobatic flight path to the floor, with the box landing on its top,
will break the PEMs that hold it to the bottom of the box.

There are big cutouts along the pcb edges, to let air flow up into the
fan, which don't help.



-- 

I yam what I yam - Popeye

On Sat, 09 Apr 2022 11:18:19 -0700, John Larkin
<jlarkin@highland_atwork_technology.com> wrote:

>On Sat, 09 Apr 2022 11:05:29 -0400, legg <legg@nospam.magma.ca> wrote:
>
>>On Fri, 08 Apr 2022 21:39:25 -0700, jlarkin@highlandsniptechnology.com
>>wrote:
>>
>>>OPA197 is a great little opamp. 36 volts RRIO, 10 MHz, pA bias
>>>current, pretty good offset specs, 5 nV noise, EMI hardened.
>>>
>>>Like most opamps, it is specified to be stable up to some capacitive
>>>load, 1 nF in this case. For bigger caps they show the usual R+C load
>>>stabilizing idea on the data sheet. That all ignores the Williams
>>>Effect, namely that a big enough cap will stabilize most anything.
>>>
>>>As a follower, handy for rail splitting and such,
>>>
>>>47 nF to ground oscillates
>>>
>>>1 uF ceramic rings a bit on step edges
>>>
>>>4.7 uF cer or more is stable
>>>
>>>56u or 180u polymer is stable
>>>
>>>Any tantalum cap is stable
>>>
>>>A 33u tantalum and a lot of ceramics looks nice. That may be the
>>>choice for a lot of opamps.
>>>
>>>
>>>https://www.dropbox.com/s/bsh4i0yb2yxn2oo/Z534_1.jpg?raw=1
>>>
>>>https://www.dropbox.com/s/psu1qsqwsb6wi6g/20220408_120126.jpg?raw=1
>>>
>>>https://www.dropbox.com/s/6ukbwh9kj9pdl6i/20220408_112325.jpg?raw=1
>>>
>>>https://www.dropbox.com/s/s7whu9d7gszm2im/20220408_114102.jpg?raw=1
>>>
>>>I have a case where I want to drive a many-bypassed rail at VCC/2 and
>>>don't want a resistor in series with the output.
>>
>>Just because you don't see a voltage across a 1uF cap, doesn't mean 
>>that the thing driving current into the node isn't going unpredictably
>>nuts. Case temperature?
>>
>>RL
>
>The 1u case seems to have a limit-cycle oscillation that dies out
>pretty fast.
>
>Oscillation would increase supply current, and I don't see that.
>
>The data sheet has a chart of recommended damping resistor vs cap
>load, table 3. The last entry is 2 ohms and 1 uF. Why did they stop
>there? The next step could have been 10 uF and zero ohms.
>
>And why the 47r and 100 pF point?
>
>Makes no sense.

Bifurcation?

RL

On Saturday, April 9, 2022 at 4:22:06 PM UTC-7, Phil Hobbs wrote:
> jla...@highlandsniptechnology.com wrote: 
> > On Sat, 9 Apr 2022 13:44:47 -0400, Phil Hobbs 
> > <pcdhSpamM...@electrooptical.net> wrote: 
> > 
> >> On 4/9/22 12:39 AM, jla...@highlandsniptechnology.com wrote: 
> >>> OPA197 is a great little opamp. 36 volts RRIO, 10 MHz, pA bias 
> >>> current, pretty good offset specs, 5 nV noise, EMI hardened. 
> >>> 
> >>> Like most opamps, it is specified to be stable up to some capacitive 
> >>> load, 1 nF in this case. For bigger caps they show the usual R+C load 
> >>> stabilizing idea on the data sheet. That all ignores the Williams 
> >>> Effect, namely that a big enough cap will stabilize most anything. 
> >>> 
> >>> As a follower, handy for rail splitting and such, 
> >>> I have a case where I want to drive a many-bypassed rail at VCC/2 and 
> >>> don't want a resistor in series with the output. 

In short, you don't want a power supply, you want ground, but you're making
it with an amplifier from  power supply as input.   Are there current surges on this
pseudo-ground?   If not, current sources (by the dozen, if necessary) into 
parallel RC loads are a way to get well-filtered voltage levels without power 
supply ripple sensitivity.  Takes one op amp and a pass transistor per branch.

Then again, why not use the ECL trick of +3.2V and -2.0V power supplies?
Signal circuitry  ought to be well-characterized against power supply noise,
but the signal-splitter application has a gain of 0.5 on power noise, and
adds another round of filter capacitance, multi-branched, to boot.   It's arguably better to use
split supplies (like, +1.8 and -1.8V), instead, and a single-point ground topology
like the old guys did 50  years ago.

> >> Sort of the reverse of Schawlow's law: "Anything will lase if you hit it 
> >> hard enough." ;) 
> >> 
> >> It's worth putting a sense resistor in the supply leads to check for 
> >> oscillations of very low amplitude. THat's been known to happen even 
> >> when the output looks steady on a scope.
> > Supply current looks OK. 
> > 
> > I need to drive a net to Vcc/2, and it has a dozen 10 uF ceramics to 
> > ground, and I want a low impedance drive from DC up. 
> > 
> > Looks like the added 47 uF tantalum is prudent. That adds some ESR 
> > damping. 

That 'been known' and 'looks like' means  off-the-spec-sheet design.   
The thought makes me... itch.   Maybe it's reminding  me of bugs?

> > My boss assigned me to rev this board 
> > 
> > https://www.dropbox.com/s/dnkmpdzs2va6x3z/P545_Top.jpg?raw=1 
> > 
> > which involves picking up a bunch of ECOs and reviewing the NEXT file, 
> > where people have accumulated two pages of annoying change requests. 
...
> > A previous fix hung a 1500 uF aluminum cap on the rail, an ugly hack 
> > on top.

Yeah, sounds like the itchy feelling isn't just me.

> The C load moves the output pole to lower frequency, and when it's too 
> close to the zero-cross of the main+tail poles, you wind up with 
> instability. 
> 
> A large, higher-ESR cap is often a good way to stabilize switchers and 
> LDOs, too--it's a shunt version of the usual lead/lag network used in 
> feedback amps. There's no reason that should be a problem in an op amp 
> loop, in principle. Doing stuff outside the datasheet's guaranteed 
> limits puts the responsibility on us, but oh, well--that's where it 
> winds up anyway.

The big reason to avoid split supplies, is ... the mindset of the student with a
small project to complete for class.   He will always design a negative ground system,
usually with a microprocessor/ADC that accepts no negative signal input.
Next year, he'll be writing up applications literature, based on his 'experience',
which will guide the next generation of students.

On Sun, 10 Apr 2022 11:16:13 -0700 (PDT), whit3rd <whit3rd@gmail.com>
wrote:

>On Saturday, April 9, 2022 at 4:22:06 PM UTC-7, Phil Hobbs wrote:
>> jla...@highlandsniptechnology.com wrote: 
>> > On Sat, 9 Apr 2022 13:44:47 -0400, Phil Hobbs 
>> > <pcdhSpamM...@electrooptical.net> wrote: 
>> > 
>> >> On 4/9/22 12:39 AM, jla...@highlandsniptechnology.com wrote: 
>> >>> OPA197 is a great little opamp. 36 volts RRIO, 10 MHz, pA bias 
>> >>> current, pretty good offset specs, 5 nV noise, EMI hardened. 
>> >>> 
>> >>> Like most opamps, it is specified to be stable up to some capacitive 
>> >>> load, 1 nF in this case. For bigger caps they show the usual R+C load 
>> >>> stabilizing idea on the data sheet. That all ignores the Williams 
>> >>> Effect, namely that a big enough cap will stabilize most anything. 
>> >>> 
>> >>> As a follower, handy for rail splitting and such, 
>> >>> I have a case where I want to drive a many-bypassed rail at VCC/2 and 
>> >>> don't want a resistor in series with the output. 
>
>In short, you don't want a power supply, you want ground, but you're making
>it with an amplifier from  power supply as input.   Are there current surges on this
>pseudo-ground?   If not, current sources (by the dozen, if necessary) into 
>parallel RC loads are a way to get well-filtered voltage levels without power 
>supply ripple sensitivity.  Takes one op amp and a pass transistor per branch.
>
>Then again, why not use the ECL trick of +3.2V and -2.0V power supplies?
>Signal circuitry  ought to be well-characterized against power supply noise,
>but the signal-splitter application has a gain of 0.5 on power noise, and
>adds another round of filter capacitance, multi-branched, to boot.   It's arguably better to use
>split supplies (like, +1.8 and -1.8V), instead, and a single-point ground topology
>like the old guys did 50  years ago.
>
>> >> Sort of the reverse of Schawlow's law: "Anything will lase if you hit it 
>> >> hard enough." ;) 
>> >> 
>> >> It's worth putting a sense resistor in the supply leads to check for 
>> >> oscillations of very low amplitude. THat's been known to happen even 
>> >> when the output looks steady on a scope.
>> > Supply current looks OK. 
>> > 
>> > I need to drive a net to Vcc/2, and it has a dozen 10 uF ceramics to 
>> > ground, and I want a low impedance drive from DC up. 
>> > 
>> > Looks like the added 47 uF tantalum is prudent. That adds some ESR 
>> > damping. 
>
>That 'been known' and 'looks like' means  off-the-spec-sheet design. 

Spec sheets are often incomplete or downright wrong. Experimenting and
thinking are both worthwhile. Dremeling and soldering and measuring
are a break from screens and mice too.

Spice models of opamps are typically not realistic. And this is TI,
who have their own version(s) of Spice.

  
>The thought makes me... itch.   Maybe it's reminding  me of bugs?
>
>> > My boss assigned me to rev this board 
>> > 
>> > https://www.dropbox.com/s/dnkmpdzs2va6x3z/P545_Top.jpg?raw=1 
>> > 
>> > which involves picking up a bunch of ECOs and reviewing the NEXT file, 
>> > where people have accumulated two pages of annoying change requests. 
>...
>> > A previous fix hung a 1500 uF aluminum cap on the rail, an ugly hack 
>> > on top.
>
>Yeah, sounds like the itchy feelling isn't just me.
>
>> The C load moves the output pole to lower frequency, and when it's too 
>> close to the zero-cross of the main+tail poles, you wind up with 
>> instability. 
>> 
>> A large, higher-ESR cap is often a good way to stabilize switchers and 
>> LDOs, too--it's a shunt version of the usual lead/lag network used in 
>> feedback amps. There's no reason that should be a problem in an op amp 
>> loop, in principle. Doing stuff outside the datasheet's guaranteed 
>> limits puts the responsibility on us, but oh, well--that's where it 
>> winds up anyway.
>
>The big reason to avoid split supplies, is ... the mindset of the student with a
>small project to complete for class.   He will always design a negative ground system,
>usually with a microprocessor/ADC that accepts no negative signal input.
>Next year, he'll be writing up applications literature, based on his 'experience',
>which will guide the next generation of students.

Our synchro box workes nicely with a single +24 supply from a big
wart, without a big + to - converter. So it's handy to reference
signals to a clean +12 rail.

The original version had some channel-to-channel crosstalk via that
rail, and it was fixed with a gigantic aluminum cap to ground. I
thought I'd do something more elegant for the next rev.

No students were involved.

I found this TI patent, but it may not apply to this opamp.

https://tinyurl.com/2p9ement

The first fig is interesting. There are four comp caps, but they all
hang on the output node.




-- 

I yam what I yam - Popeye

On 4/10/2022 22:14, jlarkin@highlandsniptechnology.com wrote:
> ....
> 
> Our synchro box workes nicely with a single +24 supply from a big
> wart, without a big + to - converter. So it's handy to reference
> signals to a clean +12 rail.
> 
> The original version had some channel-to-channel crosstalk via that
> rail, and it was fixed with a gigantic aluminum cap to ground. I
> thought I'd do something more elegant for the next rev.

So why do you not want a 20-30 Ohm resistor (plus one say 1k and a
few pF of a cap) if you have 12V headroom? The opamp is fast enough,
what it cannot do in 1-2 uS will be done by the bypass caps you have.
If this is your original setup and it took the huge aluminium cap
to filter the crosstalk I very much doubt shorting the opamp's
output to all the bypass caps will buy you anything. Did it?

On Mon, 11 Apr 2022 16:49:05 +0300, Dimiter_Popoff <dp@tgi-sci.com>
wrote:

>On 4/10/2022 22:14, jlarkin@highlandsniptechnology.com wrote:
>> ....
>> 
>> Our synchro box workes nicely with a single +24 supply from a big
>> wart, without a big + to - converter. So it's handy to reference
>> signals to a clean +12 rail.
>> 
>> The original version had some channel-to-channel crosstalk via that
>> rail, and it was fixed with a gigantic aluminum cap to ground. I
>> thought I'd do something more elegant for the next rev.
>
>So why do you not want a 20-30 Ohm resistor (plus one say 1k and a
>few pF of a cap) if you have 12V headroom? The opamp is fast enough,
>what it cannot do in 1-2 uS will be done by the bypass caps you have.
>If this is your original setup and it took the huge aluminium cap
>to filter the crosstalk I very much doubt shorting the opamp's
>output to all the bypass caps will buy you anything. Did it?

At low frequencies, the closed-loop output impedance of the opamp
follower will be less that the impedance of any reasonable cap. Adding
20 ohms, well, adds 20 ohms.

And why not do what's simplest? And learn something along the way?





-- 

I yam what I yam - Popeye

On 4/11/2022 17:27, jlarkin@highlandsniptechnology.com wrote:
> On Mon, 11 Apr 2022 16:49:05 +0300, Dimiter_Popoff <dp@tgi-sci.com>
> wrote:
> 
>> On 4/10/2022 22:14, jlarkin@highlandsniptechnology.com wrote:
>>> ....
>>>
>>> Our synchro box workes nicely with a single +24 supply from a big
>>> wart, without a big + to - converter. So it's handy to reference
>>> signals to a clean +12 rail.
>>>
>>> The original version had some channel-to-channel crosstalk via that
>>> rail, and it was fixed with a gigantic aluminum cap to ground. I
>>> thought I'd do something more elegant for the next rev.
>>
>> So why do you not want a 20-30 Ohm resistor (plus one say 1k and a
>> few pF of a cap) if you have 12V headroom? The opamp is fast enough,
>> what it cannot do in 1-2 uS will be done by the bypass caps you have.
>> If this is your original setup and it took the huge aluminium cap
>> to filter the crosstalk I very much doubt shorting the opamp's
>> output to all the bypass caps will buy you anything. Did it?
> 
> At low frequencies, the closed-loop output impedance of the opamp
> follower will be less that the impedance of any reasonable cap. Adding
> 20 ohms, well, adds 20 ohms.
> 
> And why not do what's simplest? And learn something along the way?
> 
> 
> 
> 
> 


Well learning something is always worth it of course. But the 20 Ohms
closed in the loop does not mean you add 20 ohms to the output
impedance, especially with all the 12V headroom that you have.
To make sure we are talking about the same thing: 20 ohms between output
and load, 1k between load and - input, a couple of pf between output and
- input to ensure stability.

On Mon, 11 Apr 2022 18:56:30 +0300, Dimiter_Popoff <dp@tgi-sci.com>
wrote:

>On 4/11/2022 17:27, jlarkin@highlandsniptechnology.com wrote:
>> On Mon, 11 Apr 2022 16:49:05 +0300, Dimiter_Popoff <dp@tgi-sci.com>
>> wrote:
>> 
>>> On 4/10/2022 22:14, jlarkin@highlandsniptechnology.com wrote:
>>>> ....
>>>>
>>>> Our synchro box workes nicely with a single +24 supply from a big
>>>> wart, without a big + to - converter. So it's handy to reference
>>>> signals to a clean +12 rail.
>>>>
>>>> The original version had some channel-to-channel crosstalk via that
>>>> rail, and it was fixed with a gigantic aluminum cap to ground. I
>>>> thought I'd do something more elegant for the next rev.
>>>
>>> So why do you not want a 20-30 Ohm resistor (plus one say 1k and a
>>> few pF of a cap) if you have 12V headroom? The opamp is fast enough,
>>> what it cannot do in 1-2 uS will be done by the bypass caps you have.
>>> If this is your original setup and it took the huge aluminium cap
>>> to filter the crosstalk I very much doubt shorting the opamp's
>>> output to all the bypass caps will buy you anything. Did it?
>> 
>> At low frequencies, the closed-loop output impedance of the opamp
>> follower will be less that the impedance of any reasonable cap. Adding
>> 20 ohms, well, adds 20 ohms.
>> 
>> And why not do what's simplest? And learn something along the way?
>> 
>> 
>> 
>> 
>> 
>
>
>Well learning something is always worth it of course. But the 20 Ohms
>closed in the loop does not mean you add 20 ohms to the output
>impedance, especially with all the 12V headroom that you have.
>To make sure we are talking about the same thing: 20 ohms between output
>and load, 1k between load and - input, a couple of pf between output and
>- input to ensure stability.

I've done that, but it will still present a higher bus impedance at
some frequencies... assuming that the opamp doesn't peak, which it
seems not to do. The real test is to snoop the transient response to a
small load step.

1K and a couple of pF is a tau of a couple of ns.



-- 

I yam what I yam - Popeye

On 4/11/2022 19:40, jlarkin@highlandsniptechnology.com wrote:
> On Mon, 11 Apr 2022 18:56:30 +0300, Dimiter_Popoff <dp@tgi-sci.com>
> wrote:
> 
>> On 4/11/2022 17:27, jlarkin@highlandsniptechnology.com wrote:
>>> On Mon, 11 Apr 2022 16:49:05 +0300, Dimiter_Popoff <dp@tgi-sci.com>
>>> wrote:
>>>
>>>> On 4/10/2022 22:14, jlarkin@highlandsniptechnology.com wrote:
>>>>> ....
>>>>>
>>>>> Our synchro box workes nicely with a single +24 supply from a big
>>>>> wart, without a big + to - converter. So it's handy to reference
>>>>> signals to a clean +12 rail.
>>>>>
>>>>> The original version had some channel-to-channel crosstalk via that
>>>>> rail, and it was fixed with a gigantic aluminum cap to ground. I
>>>>> thought I'd do something more elegant for the next rev.
>>>>
>>>> So why do you not want a 20-30 Ohm resistor (plus one say 1k and a
>>>> few pF of a cap) if you have 12V headroom? The opamp is fast enough,
>>>> what it cannot do in 1-2 uS will be done by the bypass caps you have.
>>>> If this is your original setup and it took the huge aluminium cap
>>>> to filter the crosstalk I very much doubt shorting the opamp's
>>>> output to all the bypass caps will buy you anything. Did it?
>>>
>>> At low frequencies, the closed-loop output impedance of the opamp
>>> follower will be less that the impedance of any reasonable cap. Adding
>>> 20 ohms, well, adds 20 ohms.
>>>
>>> And why not do what's simplest? And learn something along the way?
>>>
>>>
>>>
>>>
>>>
>>
>>
>> Well learning something is always worth it of course. But the 20 Ohms
>> closed in the loop does not mean you add 20 ohms to the output
>> impedance, especially with all the 12V headroom that you have.
>> To make sure we are talking about the same thing: 20 ohms between output
>> and load, 1k between load and - input, a couple of pf between output and
>> - input to ensure stability.
> 
> I've done that, but it will still present a higher bus impedance at
> some frequencies... assuming that the opamp doesn't peak, which it
> seems not to do. The real test is to snoop the transient response to a
> small load step.
> 
> 1K and a couple of pF is a tau of a couple of ns.
> 

There will always be some transient of course, this is where the bypass
caps come in. If it takes a "huge aluminium cap" to filter out to levels
you need then I understand your experiment but I doubt it will bring
much of an improvement, the opamp will still have to respond with
current etc. (And I'd be nervous about having a batch work out of spec
and not knowing if the next one will behave the same but well, it
may be of no concern in many cases).
The way to reduce the size of the aluminium cap I would go to would
be the 2 resistor and a cap circuit and a faster opamp and keeping the
compensation as close as practical. But whatever you do you will need
enough capacitance to filter out the transients, it is just a matter of
how much is enough. With this opamp I'd say 100uF would be plenty.
opamp,