AW: [sdiy] Seperate Grounds

[harry]

Well Ingo... thanks. Now you've taken us to about level six... ;^P

Of course you are completely right.  Especially on the "EMC" points.
I'll add just a little.

Ingo Debus wrote:

> So, to summarize, we want three things:
> 1. Star ground for the power supply, so the ground connection from a
> module to the power supply doesn't share supply currents of other
> modules. This is according to Harry's drain pipe analogy.

ok.

>
> 2. If a signal is transmitted from one module to another, a direct
> ground connection between the two modules that does not share any other
> current. Otherwise the voltage drop caused by these "other currents"
> will add to the signal.

Maybe yes..maybe no. You pointed out that this means there are now two
ground paths...  If both grounds are at the same potential, this would be good...
If not... this would be bad.  If these are patch cords (with shields) it might be
best
to connect the shield at only one end... and let the signals use the other ground
as a reference...

>
> 3. No ground loops.
> (I'm using the term "modules" here, these can be modules of a modular
> synth or anything else)

A ground "loop" means that there are two grounds at different potentials...
ground loops are always bad.  But sometimes increasing the number of grounds
lowers the overall impedance of the ground system... and can reduce problems.
For instance... the "ground plane" used for power distribution in a multilayer PCB

usually reduces the grounding problems.  Its like making a local "ocean" for all
the
grounds to dump into.

Martin Czech pointed out in his post we need to include the parasitic elements of
the circuit... the resistance, capacitance, and inductance of the ground traces.
If these
were not present... EVERY ground would be the same no matter what.

> Obviously, we can't fulfil all three demands at the same time. If
> there's a direct ground connection (this can be the shield of a patch
> cord, for instance) between two modules and a star grounding scheme, we
> get a ground loop. If ground loops aren't allowed, and the modules are
> star grounded, the only ground connection between two modules shares the
> supply currents of these modules. If one of the modules involved has
> nasty supply current spikes, this can be a killer.
> Sometimes this dilemma can be solved by introducing two grounds: a power
> supply ground and a signal reference ground. But this often isn't
> possible, many ICs have only one ground pin which serves both for power
> supply and signal reference.

Agreed. Usually the best solution is to make two separate grounds... and then
connect them at exactly one and only one place.... a place that you choose to let
this happen. That's why I'm using insulated 1/4" jacks these days... I can say
with
certainty they are not connected through the chassis... ONLY through the wire and
I
say where that happens !

> I think the only thing one can do is find a compromise. There's no
> one-and-only solution that fits best for all cases. Sometimes ground
> loops aren't that bad.

No... Ground Loops are always bad.  If they are not at different porentials...
then they are no longer a loop, they are the same point.  Tying the same point
together
with multiple wires would reduce the inductance and resistance for sure... THAT
might be a better thing.

> Sometimes bus grounding might work better than
> star grounding, because the ground connections between modules become shorter.

Agreed... because of the parasitic elements of the ground.  But I'd make the
statement

"STAR grounds are almost ALWAYS better than Bus grounds... its just that a lot of
times they are not practical.  If the bus ground is performing well there is no
reason to
'gild the lily' and try to fix what is not broken"

face it... star grounds are not real practical. Usually a dual-bus ground (clean /
dirty)
is the best trade.

One random, haphazard ground that daisy-chains everywhere without any forethought
is probably the worst case....  ;^P

> There's another demand, by the EMC people: the ground of a jack has to
> be connected directly to the chassis where the jack is, no "long" (i.e.
> few inches) wire to a ground reference point. Now things really get
> complicated ;-)

Bingo... that's parasitics again.  I spent a week trying to get RF out of a
commercial
mixer for this reason... the RF was riding the 'ground' shield inside the
chassis... through about 3" of cable before it touched the real ground. The
chassis was a poor ground and the signal got inside the case... where it could
radiate to the sensitive preamps.

Imagine my surprise when I plugged a shorted audio cable into the mixer and the
noise got worse. How can that be ??? Easy (once you understand)... the shield of
the cable was not really "ground" so the whole thing was adding an antenna to the
radio receiver.

The fix involved making the 3" wires inside the box very unattractive to RF...
Ferrite
beads on all.  Now why didn't the manufacturer do this ??? Either they did not
notice the problem (not near a transmitter...) or they didn't care. It might have
been good enough for them.

OTOH this is why a lot of us have day jobs... I hope that ground problems don't go
away anytime soon...  (they won't)

H^) harry

>