In single ended systems, shielding is what matters; in balanced systems, twisting matters.
PREAMBLE - the following is based upon my understanding of E/M theory, and the result of a good 10 years in scientific research applications using cable...
Twisted pair operates on the premise that induced noise should be equally induced - in amplitude and phase - on both lines. The goal isn't to avoid induced noise; the goal is to make sure induced noise is equally coupled to both leads.
Take a pair of wires: A+ and A-. Now, put a signal down the wires; let's take the signal (X) and break it into two parts,B and C such that (B - C = X). Now, put x+ on A+, and x- on A-. In effect, X = A+ - A-.
Now, couple in noise, N, to both leads. What do you get at the end? Well, on A+ we have B + N. On A- we have C + N. Now do the subtraction to get back to X:
X = A+ - A- = (B + N) - (C + N) = B + N - C - N = B - C.
Note that the noise cancels out! That is the result of a balanced system, where the signal is the DIFFERENCE between what is on the two conductors. And this is why pro-sound applications use balanced conductors (XLR, which has both A+ and A- conductors, plus a shield/ground reference). We don't care if noise gets on to the line; we want it to get on both lines equally so that when we take the difference the noise gets cancelled out.
Now, let's consider an unbalanced system, like you have with RCAs. You have the same wire, A+ and A-. In this case, the signal X is placed on A+. Ground is on A-. All should be OK, right?
Wrong. The problem is that ground IS ground - it can sink lots of current (relatively speaking). As such, inducing noise on to the ground lead does NOT result in the same amount of induced noise as on the signal lead. You can think of it as the signal on A+ getting N noise, and the ground getting N/2 (or less) noise. Now the difference between signal and ground is NOT the same - the noise is not equally coupled. The result? Noise in the line.
Shielding acts to stop radiated fields. In particular, we're concerned with electric fields, as you get from most car noise sources. The noise is coupled into the shield, and - provided the shielding is sufficient - the noise field is blocked from the conductor. It can't get to the conductor at all. Then you can derive a filtered ground at the receiving end of the signal, and the coupled noise on the shield isn't a problem - you don't even worry about the signal/reference on the shield.
So, in terms of balanced systems, shielding is redundant; we don't care if it's there or not, since we want it to equally couple to both conductors.
In terms of unbalanced systems, shielding is the ONLY way we can eliminate noise. We have to keep the noise from getting on the signal at all, since we can't subtract it back out (we're not balanced).
Now, as far as cables go... I worked in the SONAR research field for a a good 6 years. In that time, I designed some VERY high-end (still state-of-the-art) gear, with full 22 bit dynamic range (132 dB S/N+D levels) capability. Beyond what extreme-high-end audio can do, even.
Anyway, we used to instrument up the fronts of dams, to monitor fish bypasses. Lots of concern of salmon getting sucked through the turbines on power generation dams, rather than using the fish bypasses. So lots of money spent on using SONAR to track the fish, determine how/where they are moving, and use that information to build more effective bypasses.
Now, a typical SONAR element will return signals on the order of 10-15 microvolts from a typical 1 year old salmon, at a range of 5 meters. In other words, we get VERY low signals from the elements. And we have to measure and quantify that signal, to make the determination of how effective the current bypass is.
Problem is, we were installing on dams - power generation dams. As such, we often ran over 1000' (yes, one thousand) feet of cable along the top of a dam. You know, where you have 500 kV (half a million volts) floating around. Not like the easy 40 kV you have in a distributor!
So, how did we make our measurements so noise-free? Balanced systems and good cable. We used LOTS of Belden 8412 - around $0.60 a foot, in 100 foot quantities (around $0.11 a foot when buying a mile). Twisted shielded pair. The reason for twisted was pretty obvious - noise rejection. But the shielding was just as critical! If it wasn't there, the noise induced on the signal wires could be as high as 400-500V, and blow up the preamp inputs! With the shielding, the induced noise was in the range of a mV or two; easy to handle, and since it was equally coupled on both lines balancing cancelled it out.
So, from that point on I use Belden 8412 for line level signals. You can use it balanced or unbalanced, it's very tough (rubber jacket over the conductors, and over the shield), cotton filler (non-conductive), good braided shield, etc. And easy to pull - not too fat at all. Overall, great cable to use, easy to work with, and relatively low cost. Looks nice, too...
Dan Wiggins
Adire Audio