groundloop correction

[Haible Juergen]

Remember the GND loop discussion a few weeks ago ?
Seems I was wrong.
I had emphasized the influence of hum induction into the loops
that are formed by patchcords. I'm still sure there is *some*
induction here, but I think the people who pointed out that this
is hardly the main reason were right.

My point of view was that only with a very stupid beginner's design
error one could get hum into the modules because of *galvanic*
coupling. Which is still true: When I opened my JH-3 on the
weekend (it got a wooden enclosure now, finally), I looked
at the power supply, and there *was* a very stupid design error.

So for those who are interested, a brief summary of my PSU structure:
Mains transformer, bridge rectifier, two large eloctrolytic's.
Then 3 pairs of 7818 / 7918's for three 1A +/-18V branches. Each of
these branches feeding 7 Modules with star connection for +18V, GND,
and -18V to each module. Onboard +/-15V regulation on each module.
Then a quite loose GND connection across the front (aluminium) panels
and non-isolated 6.3mm jacks, and a GND connection between
the Modules when shielded patch cables are used.

Everything was fine when I started with a few modules. But with the
finished system, I got a lot of hum at almost random level, depending
on which modules were connected.

Analyzing the problem, I found that large hum occured every time when 
I connected modules that were supplied from different +/-18V branches.
(This was masked under normal operation conditions, because the more
connections I made between such modules, the more was the hum
reduced ! - quite obvious in retrospect ...)

Now to make a long story short, the problem was that the main GND star 
point, where all the 18V branches and also the 7818 / 7918 GND connections
met, was located at the large electrolytic's GND point. Which is in general
not a bad way to to it, if you just make sure that the regulator's GND
connection lies between the electrolytics' GND point and the GND output,
and if the electrolytic's GND point lies between the xformer's center tap
and the regulator's GND. (Hope I described this fairly right - I mean just
the general PSU design rules.)
BUT with my 3 independent branches, this caused a problem.
If I was patching two modules from different PSU branches together, the
internal
GND connection included a loop over two (ever so slightly) different
locations
on the GND trace between the electrolytics and the regulators. Just a few
millimeters, but enough to cause severe problems.
The cure was easy: I shifted the GND star point from it's original location
to the GND distribution bar of one of the 3 supply branches. This might
be slightly suboptimal for the other two branches, but who cares - this
is the 18V pre-regulated voltage only! The important thing is that the
GND return path between two modules does not go *into* the PSU
anymore, but is closed a few millimeters before it reaches the triple
PSU structure.
This solved my problems completely.
Sorry if I frightened you too much with the hum induction stuff. 

JH.