[sdiy] decoupling caps again

[René Schmitz]

Hi everyone,

Am 03.09.2013 02:53, schrieb Magnus Danielson:
> On 09/02/2013 11:54 PM, Ian Fritz wrote:
>> I agree with the first comment below the article.  I'm sure that most
>> diy-ers have at least one decoupling cap where power enters the
>> board.  That already makes our typical boards different from the
>> author's experiment, where he was comparing to the case where there
>> was no decoupling at all.
>>
>> Just because the author saw some oscillations in one particular
>> circuit/layout doesn't mean you have to put decoupling caps on every
>> chip.  The earlier studies, in fact, showed that a few caps here and
>> there on an analog board are plenty.  This has always worked well for
>> me.  In fact, I've never found a problem of any kind that could be
>> fixed by adding more bypass caps.
> Most op-amps running in linear mode would not benefit from bypass caps,
> but their bandwidth might be a bit too high, so a feedback cap can keep
> things clean.

The OPA343 in the example circuit might be a bit pathologic as an 
example. This is a rail to rail opamp that accomplishes its full input 
common mode by really having two sets of input stages. So at one point 
during a cycle (when you near the rail) one of the stages goes out of 
conduction. So there is probably a cyclic variation of the amplifier 
supply current here. (Would have been interesting if the oscillations 
would stop if he'd chosen an inverting circuit.)

Your garden variety opamp would do this as well (among other nasty 
things like phase reversal), but usually we avoid exceeding the common 
mode range. (And bypass caps are of no help there if you do.)

> My mind comes to comparators or other pushers of fairly steep slopes,
> such as digital gates. Putting caps over them helps to keep the rest of
> the board fairly quiet.

Quad opamps where one amp is used as a comparator for example can 
benefit from a bypass cap. I had this on a circuit where one of the 
other amps in the package was a sine shaper, and there was some 
feedthrough of the pulses that were generated in the neighbor amp.

> Feeding an analog design with caps as you would do for a digital design
> is a bit excessive, agree.

I agree. But people tend to think in generalizations, and safe-side. If 
you don't want to worry you can add some extra caps, instead of 
analysing where they might be necessary.

Cheers,
  René

-- 
uzs159 at uni-bonn.de
http://www.schmitzbits.de/