[sdiy] power supply decoupling capacitors

[Czech Martin]

I used to drop caps here and there in the circuit, for each
active component one or two, depending if it has one or
two supplies, and I tied the cap to GND, say filter GND.

This was a habit, because this is the way that people handle
it.

Now I start thinking (which is basically a bad thing, I know).
Is this necessary, or is it good at all?

ok, what's the difference in putting a cap to V- instead to gnd?
GND will be a quite long line, maybe a plane on pcb (most diy circuits
do not have this, they have single layer pcb, wire wrap circuits
have also usually no ground plane), then comes the connector,
then the line to PSU. Say 20cm. Makes 200nH (roughly!).
If the op amp upsets V+ (because the output stage wants to go up)
V- will not be pulled, so why not tie the capacitor to this V-?
If think the result will be the same, the *local* voltages of
V+ and V- are more stable then without cap.
This should also prevent oscillations via feedback through
the power lines (which some fast op amps like the 5534 can show).

The PSRR of op amps is poor, if we go higher then 1MHz or so,
and these are the frequencies where a 100nF cap will get "active"
(X ~ 1.6 Ohm @ 1MHz). Of course, the cap has also some inductance,
if the leads are 5mm long (which is quite short for a board
like I mentioned above) we have > 10nH. So the resonant frequency
will be 5 MHz (1/sqrt(L*C)/2/pi), above that the cap really 
works like an inductance.
Maybe this is ok for the circuit, because it does not see 5MHz
any more. Maybe this is not ok, so a smaller cap  with shorter
leads, maybe 10nF will be good till 50MHz.

So if we talk about decoupling, we're really talking about
1MHz .... 50MHz (for most analog circuits, and some slow CMOS logic).
Lead inductance is very important. A 200nH trace will have
X~1.3 Ohm @ 1MHz, and this will linearly scale with frequency.
Ground cables on my lab bench are 500 mm long...
So we may have to decide between local ground, and far away ground.

Finally I remember that all this RCL-stuff will create resonant
circuits with damping. I think it is not too hard to design an example
where the resonant frequency hits the operating frequency, or some
harmonic and where the damping is so bad, that things get worse
with "decoupling" caps.


???

m.c.


-----Original Message-----
From: Neil Johnson [mailto:nej22 at hermes.cam.ac.uk]
Sent: Donnerstag, 8. Mai 2003 23:24
To: Paul Maddox
Cc: synth-diy
Subject: Re: [sdiy] power supply decoupling capacitors



Paul Maddox wrote:
> As a rule, anything IC has one 100nF cap for each supply for things I do..

A good rule, but not always necessary....as in anything "it depends".
Some logic boards you can get away with 10nF per chip.  If you have a
flooded ground plane you can afford to be even more light on decoupling
due to the much lower ground path inductance.

> > I also wonder why they use two caps go from + to gnd, and gnd to -.
> > Wouldn't one cap between + - be sufficent?
> noooooo....
> the cap passes HF noise to ground, if you connect one across supplies,
> you allow HF on the +ve rail to goto the -ve rail and vicaversa,
> allowing for the noise on +ve and -ve to be present on BOTH supplies...

True, but sometimes that's better than letting it get near the ground.
Remember, a typical op-amp will have a PSRR of around 100dB (for a TL071),
whereas putting that onto the ground could be worse.

But, as always, it depends!

> > Some people say that connecting caps to gnd will introduce unwanted
> > noise in your gnd, which is a bad thing if gnd and signal gnd is the
> > same thing.
>
> yes...
> So you shoud design your grounds carefully...

Amen to that!!

> This PCB I mentin has TWO grounds, one analogue and one digital.

*swoon* Only _two_??  Paul, you're not trying hard enough!

One board I did a few years ago (blimey, was it really that long ago?) had
four grounds---one for the digital section, one for the RF generator, one
for the low-noise amplifier, and one for the receiver amplifier.  We even
put the LNA in its own shieled metal box on the PCB, using its own ground
plane as the bottom plate.  Fun times indeed.

There's a lot goes into designing PCBs.  In many, many cases the layout
can follow standard rules, but when you want low noise, high speed, or any
combination, that's when the "fun" kicks in.  And that's before I mention
RF....

Neil
(currently fighting Catalan numbers and exploding algorithms...)

--
Neil Johnson :: Computer Laboratory :: University of Cambridge ::
http://www.njohnson.co.uk          http://www.cl.cam.ac.uk/~nej22
----  IEE Cambridge Branch: http://www.iee-cambridge.org.uk  ----