"gndloops"

[Martin Czech]

We head a thread about "groundloops" a while ago, and I just listened.
I'm sorry that I have to repeat some generall ideas, it may all seem to
be very basic and boring, but please read on.

I want to propose some idea and let me precisely say: it is about ONE
(big) modular system. I'm not talking about different separate machines
tied together, just ONE (in the case of a large distributed system,
with different power supplys etc etc. you won't get away without
balanced connections). The case is connected to power-gnd, but the jacks 
are isolated, power-gnd may be connected only at the star point to signal gnd.

Ok, it was proposed to go balanced inside the modular (a inexpensive
version though), for control voltages and for audio signals. This would
the the best solution, no question. But this causes a lot of headake, too,
what about mixing multiple inputs, "stereo" jacks and cables, offset
and noise due to cheap balanced to unbalanced converters, no good
common mode suppression for cheap solutions and so on.

So I thought: is this really neccessary? What is really going on?  And
I felt I had no clue. And so I drew a simplified model of the ground
net of two modules. I.e. you have a massive groundbar (star point) and
then, say 1m of 2,5mm^2 gnd cable for each module A and B (7mOhm). You
have contact resistances (I found 15mOhm in a catalogue), lead trace
resistance (5mOhm), and then you have the module output/input jacks
(again 15 mOhm), and last but not least the patchcord (2m, 56mOhm
Signal, 14mOhm Shield). Module A is a generator, it has an ideal
voltage source (which is modelled just as a ideal short, we're not
interested in the signal, but in the error signal) tied to it's local
ground with added Rg of 100 Ohm, this is one side of the signal wire,
and module B is a reciever, with 10kOhm input impedance to it's own
local ground, that's the other side of the signal wire.
The local ground of module B was taken as reference, or 0.  

Ok, a very simple dc model.

And now ? Nothing happens. No ground currents flow, thus the input of
module B sees no error voltage. The reason for "ground loop" problems
is not that there is a mystical loop, but that there are currents in
the module reference leads. Inside our modular system there are no
ground currents from outside, say from our washing machine or whatever
(see restrictions above).  So where should gnd currents come from? 

If you have a look at the typicall circuits we use, we very often find
gnd as a reference point (say for opamps), but that means no current,
or very little current (uA). Only a few situations draw/dump really current
into gnd:  LED drivers, timing cap discharge, filters with caps to gnd
etc.  Ok, there we have the current.

Now I introduced such a current (note :  dc, this is all about dc) into
the local ground of module A or module B. Ok, then we have an erroneous
"input voltage" at the reciever module B, simply because module A has
been lifted or lowered from the reference potential by the currents
through the various lead impedances.  

With the said element values and a current of only 10mA I found -40dB
error at the input (0 dB assumed to be 1V). If the current would be the
discharge current of an vco, or lfo, this would be very very noticeable
(somebody reported such a feedthrough in his system).

And what's more: this kind of error can't be cured by removing the loop
(disconnecting the shield).  This will make the problem even worse!
Because removing the loop means more resistance and thus more potential
difference between local grounds A and B. The only way to get rid of
the error is to remove the signal wire (you may leave the shield), but
this is of course ridiculous. Or to put it the other way: Even if you
have a star gnd system and one gnd "wire" in each module, you will end
up with severe crosstalk (using unbalanced signal processing).  You
just can't make the total gnd wiring impedance any lower than ~10mOhm.

TO SAY IT LOUD AND CLEAR: I THINK THE GND LOOP BY CABLE SHIELDS
CONNECTED AT BOTH ENDS IS NOT HARMFULL IN THE SPECIAL APPLICATION I'M
TALKING ABOUT.  IT IS USEFULL. That's why the hyphenation in the
subject field. (Ok, if there is a strong magnetic ac field, it will
induce a current, but this should be avoided anyway).

My proposal to avoid such problems is old and simple: there must be
different types of gnd.  One for reference purposes, one for dirty LED
currents, one for audio frequent currents from filters etc. etc. All
these gnd go directly to the star point.  Inside the module the
different gnds are connected by , say 5 Ohm carbon type resistors.
This will prevent too large potential difference on the gnd nets in
case of failure.  This way, even a 100mA current will cause only little
error (~ -100dB).

So, far, so good. All wellknow to the engineers. Nothing new. So what?

Now comes the interesting part.

This was more or less dc. If we also want to understand ac
behaviour, supply decoupling becomes an issue. Because the decoupling
caps go directly to gnd.  The question is: Is it possible to play the
same trick again, i.e. keeping the decoupling current away from our
reference via extra leads to the star point? This would mean to use
decoupling caps with "very long legs", the 1m of gnd cable to the star
point. This may well be 1uH of lead inductance and may disable the
whole decoupling. But maybe there is a way to "half connect" local
reference and decoupling gnd (e.g. the 5 Ohm carbon), in order to get
an overall aceptable behaviour? I don't know. This is not so easy to
simulate, because you have to know the complete ac characteristics from
0-100MHz of all components (caps, opamps, regulator, 100MHz is just the
border to HF, the wavelenght is still about 3m).  I think I should
instead build a hardware model, i.e. two pseudo modules, where I can
easily change things and see, what happens.

By the way: It was show by measurements here at Micronas-Intermetall,
that low loss caps are not always recommended for decoupling. Low loss
means high q for the resonance of the cap, this could result in hf
"ringing". Better use a cap with low inductance but some resistance,
i.e. loss. This will damp hf oscillations.

Any comments ? Experiences ?


m.c.

------------------------------------------------------------------------
m.c. has made it finally:  3 CDs out now; 72 min. minimum; "1"
(1994-1995),"2" (95-96),"three" (96-97); experimental stuff; mostly
Eimert/Stockhausen style; but also modern popular style