Decoupling capacitors & rationalizing their use.

[Optikon]

All, I have a general question that I would like some comments/feedback on.

When considering operational amplifier circuits, it is very common to provide decoupling capacitors on the respective power supply pins.

Without getting into circuits/schematics (cause I dont have one to give), consider an amplifier circuit that is precision and VERY low bandwidth (like less than 1KHz down to DC)

What are good reasons that one should still provide (or not) decoupling capacitors on power supply pins?

There presumably are no high speed circuits operating in the amplifier so the usual use of decoupling caps to provide a local supply (decoupling from the far away/high impedance power supply) does not seem to apply. No fast changing currents, no large surges of current, so why have them at all?

Very often I see "blind" use of them on low bandwidth circuits. In practice, I have removed them with no noticable difference buit maybe I shouldn't?

Again, very slow, precision amplifier applications only - any thoughts on why or why not to keep decoupling caps on them?

Thanks!

 

[Styx]

depends,

a recent board I layedout I did not put dedicated OPAMP decoupling caps on the supply pins, because the board only had two OPAMPS and the DCC converters (+ their decoupling) were very close

Yes decoupling is used to minise the effect of of fast-edges and as you say you dont have any, but they are really needed for analogue IF the supply rails to the chip are noisy, if the rails are noisy then the output signal is going to be more noisy then you would otherwise expect.

for the hassle of two 100nF, I would put them down

 

[Optikon]

depends,

a recent board I layedout I did not put dedicated OPAMP decoupling caps on the supply pins, because the board only had two OPAMPS and the DCC converters (+ their decoupling) were very close

Yes decoupling is used to minise the effect of of fast-edges and as you say you dont have any, but they are really needed for analogue IF the supply rails to the chip are noisy, if the rails are noisy then the output signal is going to be more noisy then you would otherwise expect.

for the hassle of two 100nF, I would put them down

Yes styx, I agree about the noise of the power supply. Good point. In one situation I think about, the supply is quiet and has 100nF bypass on it. So what about local decoupling on OA? Also I agree with you, whats a couple of 100nF small caps going to do? Throw em in! Yes well, I normally do but sometimes I think for no good reason but it doesn't hurt! ----- until you have 25+ opamps all on a low noise design and no board space for what would be at least 50 decouple caps. so I wrestle with whether or not to bother!

 

[Nigel Goodwin]

Yes, always fit them, they can't do any harm, and may well be essential under particular circumstances - one important thing they do is help to prevent the circuit oscillating due to positive feedback via the supply.

Think of it as crossing the road with your eyes open!, you may well be able to cross the road quite happily with your eyes closed! - then one day you meet a bus :lol:

 

[Styx]

a controller-board some1 design for a project at work has alot of opamps on it, what he did was basically "shared" decoupling between a few chip's (those chips already very close together - not talking abt a 100nF nx to one chip an another chip an inch away)

IF one section was especially susceptible to noise then a quick mod usually fixe's it

 

[Optikon]

Yes, always fit them, they can't do any harm, and may well be essential under particular circumstances - one important thing they do is help to prevent the circuit oscillating due to positive feedback via the supply.

Think of it as crossing the road with your eyes open!, you may well be able to cross the road quite happily with your eyes closed! - then one day you meet a bus :lol:

I know they cant do any harm.. I am asking about trying to justify putting them in when it seems like they are ineffective.

The case I'm interested in:
1) power supply is linear and filtered the @#$ out of it. It's as quiet as it will ever practically get. Not a switcher in sight.

2) All opamp circuitry is low noise, low bandwidth & precision.

Could you elaborate on the oscillation due to +ve feedback from the supply? Do you mean via PSRR spec? Output supply noise feeding back? Wouldn't this have to be a pretty low frequency feedback & oscillation (remember supply has no HF noise & Rejection will be good)
To snuff this out with decoupling would require a very large capacitace which, is not usually what local decoupling types are used for.


They cant do harm, unless your out of PCB real-estate.

Thank you NIgel! Anyone else have any ideas about the perils of not using them in this application?

 

[Styx]

The other point of putting the de-coupling down is to "de-couple" the track inductance

if it is a big board then there will be signficant track inductance and if you try to take a blat of current from that rail, it will locally sag (equally you stop taking higher current it will rise).

the cap will attempt to cancel out the indcutive nature of the tracks

 

[Optikon]

The other point of putting the de-coupling down is to "de-couple" the track inductance

if it is a big board then there will be signficant track inductance and if you try to take a blat of current from that rail, it will locally sag (equally you stop taking higher current it will rise).

the cap will attempt to cancel out the indcutive nature of the tracks

This track inductance is only a concern when there are sufficiently high frequency currents traveling in it or large steps (edges) of current draw. Digital switching circuits are the best example of this. Low BW opamps, aren't performing any abrupt changes on the power supply.

I like your idea of sharing decoupling (forgot about that move). I think I'll still put down a few (though, I believe they are doing zilch)

All the points from you guys are a good reminder why to put them there.
I just dont see this case fitting into any of those points. Yet, I still dont feel right about not putting something in... crossing road eyes closed as Nigel put it!...

I'll comprimise and share it amongst a few evenly located.

Thanks guys!

 

[Nigel Goodwin]

The point of them being close to the chip is to decouple the supply connections to it, as suggested by Styx (although it's as much the resistance as the inductance).

If you are 100% sure that their lack will NEVER cause a problem, under ANY circumstances whatsoever, then leave them out - but if you've got room on the board, I'd personally always fit them.

I've seen far too many modifications over the years, due to parts been left out for cost savings - then they find problems occurring, and have to modify all the units.

 

[Dean Huster]

High-gain audio amplifiers with several stages absolutely must be decoupled regardless of the frequency of operation or you'll end up with an oscillator. It was once said that it's far easier to make an oscillator than it is to make an amplifier, and poor decoupling is one of the reasons for that. Poor layout is another, as is too much gain-per-stage.

Caps are so darned cheap that it's silly to not use them. An ounce of prevention .....

Dean

 

[Roff]

High-gain audio amplifiers with several stages absolutely must be decoupled regardless of the frequency of operation or you'll end up with an oscillator. It was once said that it's far easier to make an oscillator than it is to make an amplifier, and poor decoupling is one of the reasons for that. Poor layout is another, as is too much gain-per-stage.

Caps are so darned cheap that it's silly to not use them. An ounce of prevention .....

Dean

Another saying:
Amplifiers oscillate, oscillators don't.
Take it from an old analog engineer. Nigel and Dean and the others speak the truth.