Theo wrote:
"Dave Plowman (News)" wrote:
I can see that. Just odd I've never seen it used before.

FWIW it's not wrong to use separate capacitors V+ to 0 and 0 to V-, but
effectively what you have there is half the capacitance from V+ to V- and
double the ESR, so you're adding components to make it worse. It might make
sense where there are single-rail loads running between V+ and 0 - for
example digital logic, which are more the kind of thing decoupling
capacitance is intended for (high frequency switching loads rather than
general audio ripple).

Theo


One thing to be careful of, is this. If you're doing
this, the caps should not be dipped tantalum.

+15 ---+-----
|+
---
---
|
0 ---+-+---
|+
---
---
|
-15 -----+---

The problem with the circuit, is the behavior of
the power source. Any little bit of reverse bias
on a tantalum, sets it up to burst. And the epoxy dipped
ones will leave a PCB in a hurry. I had one ricochet
off a wall like a bullet. This only happened on my
bipolar wired setup (op amp test circuits), like the
one above, not on unipolar circuits.

If you're going to do that, ceramic (unpolarized)
and maybe an electrolytic will deal with a bit
of reverse bias, with less argument.

If you want to do bulk decoupling to reduce ripple
on the outside rails, maybe a tantalum would be
OK for this. But after my experiences, I don't
put tantalum in circuits any more. I still have
a few tantalums left, but they stay in that drawer
in the parts case.

+15 ---+---------+---
|+
---
---
|
0 ---+-+--- |
|
|
|
|
-15 -----+-------+----

The PSRR of the circuit, helps define how clean
your rails need to be. Linear regulators are
surprisingly noisy, and opamp PSRR isn't that
good at 1MHz. When you need absolutely the
lowest noise, supplying power can be a challenge.
A switcher at a fixed frequency, followed by
several stages of filtering circuits, may give lower
overall noise than the usage of linears. That's because,
by concentrating all the noise at the one frequency,
a more effective filter can be designed to "notch"
it out. You don't want variable-frequency switchers,
as the noise moves all over the place.

To start with then, running the project off a couple
nine volt batteries, and some good-sized electrolytics,
might be a way to go. Then there are no linears, and
it's just the noise spectrum of a battery (whatever that
is).

In the top diagram, the separate bypass on each side
of the bipolar supply, I never would have considered
that the rail polarity could reverse slightly at
shutdown. But the tantalums told me what was happening,
in a very effective way. When it goes with a "bang!",
it's like it is saying "you idiot, you reversed me!!!".
With safer capacitor types, you don't have to worry
quite as much.

Paul