MODIFIED March 18, 2010 (when I found the Thomas Hays 
AES paper preprint).  

Something I once posted to a listserv after totalling my 
business vehicle.

My experience has been that the rules always work, unless there 
is something else somewhere that breaks the rules.  I got my 
shielding and grounding technique from a VERY well-written 
document that Thomas M. Hays, who was with MCI (tape recorders 
and consoles), Fort Lauderdale, Florida published in their heyday.  
In this document, "DIFFERENTIAL TECHNOLOGY IN RECORDING CONSOLES 
AND THE IMPACT OF TRANSFORMERLESS CIRCUITRY ON GROUNDING TECHNIQUE", 

http://www.aes.org/e-lib/browse.cfm?elib=3653

Tom attempted to show every possible interface, analyze the flow 
of the audio, the flow of ground currents, the ground differential 
between source and load, and the phenomenon of electrostatic and 
electromagnetic coupling, and came up with logical reasoning for 
each type of interface.  I've used it religiously since then, 
and have never been let down.  When it hasn't worked, it has 
been because of another broken rule somewhere else, and if that 
rule is obeyed, the problem goes away.

1)  Except for transformer balanced or fully floating sources
(microphones or xfmr output gear), shields are grounded at the 
source end ONLY, preserving the relationship between the signal 
wire(s) and the reference ground at the source (kinda like a 
transmission line).

2)  Shields MUST be continuous, and have no other connection 
to the real world (kinda like a faraday cage).

3)  RFI bypassing can be done at the load end if necessary, 
but the value of the capacitor becomes critical from both a 
resonance standpoint, and from a "sum of all capacitors at 
the end of all lines from all equipment with the same ground 
differential from the load equipment" standpoint, such as all 
the outputs from a 24 track machine to a console.

4)  Unbalanced OUTPUTS are no problem at all.  As long as 
the wiring to the load is balanced and the load is a well- 
balanced differential or transformer input, the 'low' wire 
simply senses the ground differential and cancels it out.  

5)  Unbalanced INPUTS are the bane of our existence, and 
should be avoided at all costs.  There is no way to remove 
the ground differential from the input signal, so it simply 
becomes a part of the input signal.  You're stuck with it.

6)  Long lines require good longitudinal balance, i. e. 
each wire must see the exact same impedance to the ground 
reference at both the source and load ends, so that the 
two wires present the same impedance to ground at all 
points along the run, so that fields will be picked up 
exactly equally in both wires.  It is a voltage divider 
effect, with the electrostatic/electromagnetic fields seen 
as the series element, and the impedance of the wire to 
ground as the shunt element.  As long as both wires in 
the balanced pair see exactly the same impedance at both 
their ends, they'll have the same voltage-divider-shunt 
impedance, and therefore a balanced input will successfully 
cancel out the induced common-mode signal.

While many times you can ignore the rules and things will 
work OK, the environment can change at any time, so 

               >These Rules Mean That:<

A)  Balanced inputs must be VERY balanced.  Differential 
inputs must be (at minimum) a two-opamp inverting input 
design with cross-coupled feedback, or three-opamp designs, 
so that both lines are loaded with the same dynamic impedance.  
Opamps must have phase-linear open-loop transfer throughout 
the audio spectrum for this to work.  An example of an 
amplifier that is a good diff amp is the LM318.  This is 
why the AD518 (similar chip) is used in all analog stereo 
generators from Orban.  TLO7x opamps are lousy diff amps.  
Of course, the LM318 has other audio problems, so I avoid 
them when good diff performance is not required.  

B)  Diff amp inputs must not provide a higher load impedance 
to common mode signals than to differential signals.  Ideally,
the load Z seen by the common mode signal on each wire of the 
input should be nearly half the load Z seen by the diff signal.  
An example of the wrong way to do this is an amplifier with 50K 
ohm input resistors from each of the two wires and the two 
opamps, and a 600 ohm resistor across the wires.  In this 
scenario, even though the common mode signal may be perfectly 
balanced in the wires, and the source and load Z's seen by 
the wires is equal, the induced common mode signal sees 
almost no shunt leg in the virtual voltage divider, so the 
common mode signal can easily overload the common mode input 
range of the amplifier stage.  By simply using two perfectly 
matched load resistors that are half the value, connected 
in series across the wires, with the centertap connected to 
ground, the common mode load Z is balanced, and very low, 
so that the common mode signal is attenuated much more.

C)  If an unbalanced output must be connect to a balanced 
input through a long line which may be subject to electrostatic 
or electromagnetic fields, the wire used to connect to the 
ground of the source equipment (usually the black one) should 
not be connect directly, but should be connected through an RLC 
network that is the equivalent of the output Z of the unbalanced 
output.  Usually, a simple resistor that is the same value as 
the output buildout resistor is fine, but sometimes a more exact 
solution is required.  

D)  Since (C) above can get quite tedious, the simple way to drive 
and receive signals on long lines is to use well-designed 
transformers.  Things that can bite us in the keister here are 
the well known group delay problems, the core saturation problems, 
the ringing and overshoot which always happen when a waveform 
containing energy that is outside the passband of the transformer 
is fed into the transformer, the distortions associated with 
the hysteresis of the core metal, resonance issues of the 
inductive values of the transformer with circuit capacitances,
and so on and so on.
   If you can find an old Western Electric 111-C, it's worth 
more than it's weight in gold.  The telco operating companies 
have forgotten nearly everything they learned in the early days 
about analog audio on long lines.  Try comparing the performance 
of a 111-C to one of those stupid Tellabs 4425 repeat coil cards.  
The 4425 will not successfully pass program audio above about -10 
dBm if it contains energy below around 80 Hz.  The 111-C is good 
down to 10 Hz at better than +10 dBm.

E)  Shields must be carried through punch blocks.

F)  Shields must not be cross connected anywhere, including on 
punch blocks.  I've been in facilities where an attempt to 
reduce wall space has resulted in tying shields of two or 
more circuits together on a single row of punch pins.  This 
has ALWAYS resulted in poor noise performance and higher 
crosstalk.

G)  Patch bays are the enemy.  The industry standard practice 
of tying all shields together on buss bars on the backs of 
patch bays is a bad one.  It is impossible to actually USE a 
typical patch bay without messing up an otherwise good 
grounding and shielding scheme.  Why?  Even if shields are 
carried through on individual normalled circuits, as soon 
as a patch cord is inserted there's this double-connection 
of shields that occurs.  Patch bays CAN be built that prevent 
this, but it requires a form IX jack (having an extra form B
set of contacts) so that the shield connection to the normalled 
circuit will be disconnected if the normalled high and low 
are broken.  Letting that shield stay connected is a big 
lumped Z that destroys the snug relationship between the 
signal wires and the shield that must be maintained to 
optimize shielding performance.

H)  If well-balanced input amps are the rule in a facility, 
poor shield management is worse than no shields at all. 
This is why the Radio Systems CAT-5/Studio Hub approach 
works so well.  Nice, tight twists in the pairs gets us 
closer to a transmission line approach.  I've successfully 
run more than 100 feet of Hi-Z, unbalanced audio from a 
sound board in the middle of an auditorium to the unbalanced 
input of a house sound system at around -10 dBm with no 
noise pickup at all.  We had to set up in a hurry, and 
it worked.  I didn't know until after the show that the 
'star-quad' cable we used had the shield floating 
at both ends.  Chubby Checker had it right.  The twist 
is what it's all about.  

==A paragraph or two on humility:

   Please don't take this as egotistical ravings of a madman.  I 
hesitated to present such a 'complete' presentation of my audio 
religious beliefs to such an outstanding group of peers as 
yourselves, and only decided to do it with the greatest of 
humbleness (is that a word?) and respect.  To paraphrase an 
old ribald saying, 'Opinions are like armpits, everybody's got 
one.'

   I value the opinions of everyone here, love to read them, 
and learn from them all.  If everyone here kept their mouths 
as shut as I usually do, I'd have nothing to read.  The 
other side of that coin is that I shouldn't be so quiet if I've 
got a strong opinion, so I guess stating mine is appropriate.  

==A paragraph or two on my slowly deteriorating mental condition:

   Why did I sit down and write all this?  For those of you who 
read about my accident on Monday, I found out today that the 
armpit who rear-ended me had no liability insurance.  The 
policeman who investigated the accident should have checked to 
see if the vehicle had current insurance.  He didn't.  Had he 
done so, he would have arrested the guy on the spot.  It's too 
late now.  

   My insurance agent said, "This is gonna get ugly."  I've 
done 30K miles a year for the last 15 years with no accidents 
and now this.  Writing is therapy.