AW: AW: To earth, or not to earth...

[Haible Juergen]

	Martin Czech wrote:
> -You could even "upgrade" older systems with a little pcb, and
> changing
> mono sockets to stereo. There is still hope for you J.H. ! A high end
> circuit was proposed in some elektor "audio" collection. I have it
> somewhere.
> 
Yes, there's still hope (;->) - but I don't think I'll make the JH-3
Modular symmetric. Hum is very low, with the exception of one module.
So I don't really have a hum problem there. But if I connect other
instruments (with multiple cables like audio signal, arpeggio clock, 
filter CV input ...) I run into problems. Can solve them with
transformers
or cut cable shields, but going symmetric would just be the better
solution.
So I were going to do it again from the start, I'd go balanced.

      >CON:

The greatest problems I see is 
(1) Input Attenuation and
(2) Output drivers.
... see below.

	>-What would you use as input diff amp? And what about offset
problems
	>at this stage? One solution could be a ssm2017, but I've only
seen ac
      >app notes so far

I am surprised that everybody considers the input stage a problem.
Now, that's easy. While opamp circuits often need some "artificial"
means to symmetrize the circuit, many typical synth circuits don't.
A moog ladder cries for symmetrical input. An OTA-VCA, too.
(spelling: is this "balanced" as well, or "symmetrical"?)
Actually, *most* of the multiplication devices we use for voltage
control
of parameters are symmetrical.
If we don't have an internally-symmetrical module, I suggest building a
simple differencial pair with current source and degeneratre emitter
for your typical AC inputs, and a one-opamp difference amplifier
for typical CV inputs where offset drift is an issue.

	Tim Cockram wrote:
	>If the modules are star eathed i.e. a separate earth lead from
each module to the central 
	>earth point, there should be no increase in crosstalk.  

Yes, that should do it in most cases. An exception might be crosstalk
between several signals to / from the same module. When would this
hurt? Not often, I admit. Maybe a ring modulator might become slightly
unbalanced when one input bleeds to the other one over a shared GND
connection. But I admit it's an extreme example, and wouldn't hurt much
either.

	>The earth "dustbin approach is also quite usefull  ( use 
	>an op-amp unity gain buffer with the non-inverting input
connected to ground,  the output 
	>is a goood solid ground (limited by the outputs compliance).
The "dustbin" uses the high 
	>supply rejection qualities of the op-amp.

This sounds interesting - never heard about this before !
So you use the opamp output for the "dirty" gnd (LEDs etc), and avoid
any bypassing caps to GND around this opamp (neither from its
supply pins nor from its output), so you just get the dirt on the +/-
rails
but not on GND? Great idea !!

	>By symmetrical do you mean balanced Jurgen?

Sure. Guess I speak too much German between writing English emails
(;->).

	Mark wrote:
	>Perhaps a good compromise would be to have unsheilded stackable
banana
	>plugs for control voltages and sheilded  1/4" or 1/8" cables
for audio
	>signals.  Due to typical impedances, for audio it is OK to mult
outputs but
	>not inputs.  Then again, people often like to use audio outputs
as
	>modulation sources with their modulars.

You said it yourself: To many people there is no real difference between
audio and modulation signals.

	>>If I would do it again, I'd use 1/4" *symmetrical*
connections, and
	>>nothing else.
	>
	>Really??  That strikes me as a radical choice -- especially if
you are
	>trying to avoid ground loops.  By symmetrical I assume you are
speaking
	>European and mean TRS balanced connections  :)   Firstly, It
would seem
	>like pure overkill for control voltages. 

Overkill - maybe. But no one says you need to build studio quality
symmetrical microphone inputs for your CVs. Most demanding
problem of symmetrical inputs in general is their CMRR at higher
frequencies. But you wouldn't need this for many CV inputs. Should
just have a decent CMRR at 50Hz. One opamp with four matched resistors
would do here. (This is for typical CV inputs.)

>  Secondly, you would have to
> balance the outputs of each of your modules requiring either
> transformers
> (!!!) or active balancing with additional op-amps.  
> 
Yes, this is really a problem, if not *the* problem.

If you would really use balanced cables exclusively, you could simply
build an additional inverting opamp stage to feed the ring of the
stereo jack. Problems start when you plug in a mono plug, however.
Even with current limiting resistors there is still much current
injected
into the gnd/shield path to cause bad crosstalk.
A truely balanced output stage (like this SSM chip - forgot the number)
is too expensive (and try to build *this* from discrete opamps - no!). 

One solution might be a class A differential output stage. Take the
last section of the minimoog VCA (the buffer stage). Again a degenerate
emitter differential amplifier. But use *both* legs as outputs, not just
one
as in the minimoog. I used this circuit in my JH-4 module, and it makes
a
great symmetrical output stage (to the mixing desk). You can short one
output, and it still works - you can even short several outputs together
and get a mix of the two signals (that's why I chose it for the JH-4
"Four
Voice" concept !) - BUT - you guessed it - output impedance is far from
zero, which is definetly a problem for calibrated V/Oct CVs.

	>Thirdly, how long a
	>cable run would you have in patching a modular -- is CMRR that
much of an
	>issue?? 

This is an important point. We don't have to fight many meters of cable
length. Just want to avoid small (approx. 1 or 2 meters) ground loops.
So, a symmetrical stage with decent or bad (30dB) CMRR is better than 
an unbalanced connection (which has 0dB CMRR). 

My vote is for using *cheap* symmetrical circuits, even with bad CMRR.
They are not much more expensive than unsymmetrical circuits, often
you get them for free (Moog Ladder ...), and they still improove SNR.

Another thought:
Maybe (I' only guessing here) it would even help to use unbalanced
output signals, but balanced input stages. The outputs would drive
the tip with the signal, and connect ring and shield with their *local*
GND.
The inputs jacks would have the shield unconnected, and a differential
input stage would amplify the voltage between the tip (=signal) and
the ring (=local GND of output module). This way you'd loose the
advantage
of radiation immunity (but we have short cables anyway), but you'd still
avoid GND loops (the ususal unbalanced problem) and crosstalk
(the potential banana problem).
I used this method for a critical internal connection in my FS-1
frequency
shifter, btw. It reduced carrier feedthru significantly. (Bode used
transformers for internal connections, which is even better, of course.)

Now, there's one great problem that remains: Input attenuators.
Dual potentiometers often have significan difference between both
pots. You could also try one pot and 2 resistors in a ladder
configuration,
but then it's the limited CMRR that kills you at low potentiometer
settings.

The only cheap way I see for this - and this solves the input
differential
stage problem as well: Use an OTA, and voltage control your input 
attenuation. Noise is an issue here, but you might have standardized
output levels anyway, and the OTA would always see this maximum level,
so you could optimize the divider resistor for full level. 

Long post - hope it was of some interest. 

	JH.