opamps and local decoupling of rails, some questions

[Kingston]

maybe this has been discussed an awful lot of times, so just point me to the correct threads.

Let's assume a somewhat generic scenario. We have a mixer, or some other audio device with a large number of opamps (+10). We have a somewhat generic dual rail PSU, designed around LM317/LM337 with about 15V rails. We start feeding the opamps, and measure no ripple at the PSU end.

Now, one hears all these recommendations of locally decoupling each and every opamp in almost any design, with a standard 0.1uF or similar film or ceramic caps. "Sound improves", "really works for fast opamps" etc. gets thrown around in abandon.

Why is this? Why would these local caps help at all since we have a clean PSU like above, and when there is no ripple to measure?

Please don't be afraid to dig deep. Maybe I will understand.

Thanks,
Mike

PS. I don't yet own an oscilloscope, which is why I can't actually look in detail what is going on.

 

[volker]

The decoupling capacitor serves several purposes. Supply ripple rejection, supply stabilization for sudden demands, AC low impedance to ground...
But others who are more adept, do (and should) know way more about that topic and have written more than enough about it. I'd rather keep my half knowledge on delicate topics like that to myself . For a start:

http://www.analog.com/static/imported-files/application_notes/AN-202.pdf
http://www.analog.com/static/imported-files/tutorials/MT-101.pdf
http://focus.ti.com/general/docs/litabsmultiplefilelist.tsp?literatureNumber=sloa069
Section 7


edit: links updated

 

[steppenwolf]

I'm absolutely no expert, but as far as I know you have to consider, that the pcb traces that carry the current to the opamps will have some resistance, the longer, the more resistance you'll get. Now with all those class A/B opamps in the system, it will idle at a couple of ma with no signal, but when a signal is applied, especially a fast transient, the chip has to follow that signal and needs the appropriate current to do so. It will draw a very fast surge from your rails but because there is some resistance there, the voltage drop across the pcb trace will become large enough to prevent the opamp to follow the whole amplitude of the signal. That can cause distortion. Now with a good cap directly at the opamps pins, you have a tiny but fast and very low impedance reservoir that can feed the opamp during those surges without the drop because you minimized the impedance on the rails. On normal operation the cap will charge again and be ready for the next peak...
I really hope that makes sense, please correct me if I'm wrong;-)
It don't think those caps are used to filter the rails....

All the best,
Stefan

 

[jdbakker]

[EDIT: Volker beat me to it. Duplicate info removed]

There is no such thing as ground, there's no such thing as a perfectly solid power distribution network. All power/ground traces and planes have resistance and inductance, and even if your 317 would have zero-Ohms output impedance, the supply impedance at the op-amps is invariably nonzero.

There are two problems with this. One, an op-amp output stage driving a load will pull the op-amps power lines down. If the impedance is high enough, this can lead to oscillation and distortion. Two, power rail fluctuations will impact surrounding circuitry, just like at home when the lights dim when your oven turns on.

Read Volker's links for a more in-depth treatment.

JDB.

 

[ppa]

however, for an op amp it's sufficient using one cap only instead than two, this one cap must be connected from -Va to +Va (+Va and -Va are supply voltages) to avoid oscillations.

 

[Kingston]

While I dive into those tutorials, a quick related question:

On some (lower cost) designs I see the 0.1uF cap strapped to the positive and negative rails only, and not to ground. This saves one cap, but is there actually any benefit in doing so? Could this actually cause harm?

[edit]

oh. ppa actually suggests this above as well. My thinking says it might cause one side of the a/b configuration "sag" at a worse case scenario.

 

[jdbakker]

for an op amp it's sufficient using one cap only instead than two, this one cap must be connected from -Va to +Va (+Va and -Va are supply voltages)

Not if you're driving non-trivial loads that are referenced to ground (or anything other than V+ or V-).

JD 'no shortcuts' B.

 

[jdbakker]

On some (lower cost) designs I see the 0.1uF cap strapped to the positive and negative rails only, and not to ground. This saves one cap, but is there actually any benefit in doing so? Could this actually cause harm?

As an expansion on my earlier one-liner:

It Depends.

If your power ground is tightly coupled to your audio ground, then one such cap will keep your ground cleaner (as the op-amp's supply current fluctuations aren't coupling into your ground net). It's important to remember that a decoupling cap can also be seen as a coupling cap.

The obvious counterpoint: Don't Do That, Then. If at all possible, keep signal and power ground separated, or at least be careful what you use as a reference point between stages. Easier for new designs, harder for existing ones.

Volker's first link has the full story.

JDB.

 

[ppa]

it's obvius that the preferred version is with two caps, but when the returns to gnd from the caps are so many and the op amps' circuitries permit this one, it's better choise using one cap for each op amp instead than two to avoid many gnd return paths, considering that some times it's very hard connecting many PCB tracks to one ground point.

 

[Kingston]

Thanks for those links volker, and for your comments people. Nicely summarized the topic.

I now wish I had an oscilloscope to test a couple of real world scenarios I had in mind. I would like to see if improvements are necessary on some projects of mine.

Somewhat related, I have an ongoing mixer recap + PSU upgrade. this mixer mostly has the type of "one cap decoupling" going on. Its channel strips have no ground plane, and somewhat convoluted ground traces. I will now completely forget about "blind" decoupling of the local opamp power terminals to arbitrary points in the channel ground traces.

 

[jdbakker]

it's obvius that the preferred version is with two caps

There is no single universal preferred configuration.

I'm sorry to keep harping on this, but the optimal configuration always depends on the surrounding circuitry. For every decoupling scheme you (or anyone else) proposes it is possible to find a valid, existing circuit where this scheme is not optimal. Especially hard are situations where the driving/receiving end is ill-defined; a common example in audio is the op-amp balanced line output driver.

Sure, if you know that

- all signals are single-ended
- power ground and signal ground are separated
- signals and load are ground-referenced

you can start making shortcuts. But these conditions aren't always true.

(By the way, this is one of the reasons people complain about the 'sound' of electrolytic bulk supply decoupling caps. Current must flow in a loop; if you don't or can't put decoupling directly between your driver and the load's return point, the return current will find another path -- through those nasty electrolytic caps, through the power/ground paths of surrounding circuits)

JDB.
[and what happens with dual op-amps? Two outputs, one single set of power supply pins. Which current paths will dominate?]

 

[tv]

Depends on how stubborn your opamps are. Bitches like lm6172 like 3 caps (rail-to-rail and rails-to-gnd)

With fast opamps that flow more current, you must take care that they don't pollute your ground connections.

If your opamps are lighter and less picky, perhaps it would be better to stick f.e. 1uF xr7's (not 0,1uF) rail-to-rail and see if you can get away without rail-to-ground decoupling...

But every trace in combination with a cap is a LCR resonant circuit and every opamp acts as an exciter for that resonant circuit, when speaking of psu rails.

ymmv.

 

[ppa]

it's obvius that the preferred version is with two caps

There is no single universal preferred configuration.

I'm sorry to keep harping on this, but the optimal configuration always depends on the surrounding circuitry. For every decoupling scheme you (or anyone else) proposes it is possible to find a valid, existing circuit where this scheme is not optimal. Especially hard are situations where the driving/receiving end is ill-defined; a common example in audio is the op-amp balanced line output driver.

I haven't write that exist an "universal" solution. I intended with "Preferred" that very often this was and is the choise preferred (the choise with two caps per each op amp), like is very easy to know. But, i think, it's right to report the other solution with the one cap per op amp, that, by the way, is used from SLL and others.
I think, it's possible use the one cap solution (when it's possible to use, considering it's possible use it very often) to reduce components, layout and the problems of many return to ground by grounding caps.
The one solution in a console is very interestig for the space reduction like, I repeat, SSL has done and used extesively.

 

[Kingston]

Bitches like lm6172 like 3 caps (rail-to-rail and rails-to-gnd)

How would the extra rail-to-rail cap help when you're already dumping hf to gnd from both sides?

 

[tv]

Like shunting the "capacitive divider" formed by both rail-to-gnd caps?

Like acting as an additional reservoir of "power" to keep things more in balance (i.e. not referenced to gnd directly).

Iow, depending on circumstances and your load, you may "round things up" this way - and lessen the ground pollution.

 

[ppa]

However, it's important to say that many audio op amps accept even tracks not so short before they make oscillations or bad response. Infact, several audio circuits that use many op amps for audio have only a few of decoupling caps.

 

[tv]

>>>> that use many op amps for audio
EEEEEEK !!!

 

[abbey road d enfer]

One thing that some hinted at, but I like to put it in a different light.
PS decoupling and grounding are two intricately linked facets of the audio art. The only way to master them is to analyse the complete circuit, with all ground connections replaced by resistors (and inductors if needed); then one sees all the interactions, feedback paths (negative AND positive), feedthroughs, crosstalk, noise current injection.
Most of the circuits can be gleaned from cookbooks, but the fine details of implementation need analysis.

 

[ppa]

While I dive into those tutorials, a quick related question:

On some (lower cost) designs I see the 0.1uF cap strapped to the positive and negative rails only, and not to ground. This saves one cap, but is there actually any benefit in doing so? Could this actually cause harm?

[edit]

oh. ppa actually suggests this above as well. My thinking says it might cause one side of the a/b configuration "sag" at a worse case scenario.

there are two needs:

1- to avoid op amp oscillations gived by supply tracks' impedance.

2- grounding some parts of the board circuitry.

in these boards to avoid op amp oscillations (point 1) are used these small caps of 0,1uF (with different small values in other schematics) at the positive and negative rails, because these caps have the function to shortcut at high frequencies the op amps' supply pins. Are used the one cap grounding solution to reduce the components and reduce the ground paths needed by the two caps solution,

Moreover, (point 2) for grounding these some parts of the board circuitry are often used two small caps parallelized with bigger electrolytic caps to shortcut the supply rails to ground at high and audio frequencies, to reduce the regulators' out impedance and improve the transients response of these regulators.

for example, I have see this design in ssl schematics.  

 

[Kingston]

a strange episode with some of the above put to test:

I have an R*E HDSP card with 4X output expansion. I had always wondered why the expansion card outputs sounded and measured better. They are the same chips, same configuration as the main card outputs. The expansion has less noise basically, and I could see an extra THD peak when comparing the main outputs to the expansion.

I inspect the cards and notice the expansion has independent rail bypass caps (220uF) for the outputs. Main card just has everything (input, output, some of the conversion) under the same rails with mere 100uF bypasses. And there is no local decoupling to be found! Neither on the expansion.

I thought that was a bit odd and decided to start by strapping 1uF films across the I/O opamp rails locally. Voila! That cleaned it. The difference can actually be heard! Of course I will now proceed to decoupling the whole damn lot, and increasing the bulk caps too.

I have to say I find this whole episode extremely surprising, maybe even a bit disturbing. Would think R*E designers know better? Maybe they thought it was "good enough" already and didn't bother with optimisations like this?