PCB Differential Pairs for Audio

The primary thing I would recommend is to not put any "in band" signal traces between your differential pairs. Keep the differential pairs together on the circuit board.The current loop induced voltage will look just like desired signals. Since you're putting this on a PCB it will also be good to have a ground plane on the opposite side if only using two layers. With a multi-layer board ($) having ground and power as your inner layers and signals on the outer layer works very well. If you have to cross your low level and high traces, try to cross them only at right angles.

Found this...

http://www.ieee.li/pdf/essay/differential_signals.pdf

Not much to worry about at the drive levels you cite. Here are some basics:

Keep the traces short and together.

Don't fret about the difference is trace lengths. That's not important at these frequencies.

Signal traces should cross at right angles and avoid parallel traces (ground traces are okay)

Don't mix digital and analog sections. try to keep each sequestered to their own area.

Split ground planes are not a de facto science between analog and digital sections. There are studies that show benefits and there are equally valid studies that say they are not important. Splitting a ground plane is tricky and can lead to issues. If you feel you want to go that route you must do your homework.

Sound power supply design and filtering are important.

Exactly. GA. (Don't let the Monster Cable mentality infect your thinking. People overthink the whole audio thing. Now if we're talking about LVDS at gigahertz frequencies, that's a whole other ball of wax. Bob Pease loved to tear apart various claims.)

http://electronicdesign.com/analog/what-s-all-splicing-stuff-anyhow-part-ii

http://electronicdesign.com/lighting/whats-all-hoax-stuff-anyhow

Audio black magic exists because people don't understand the underlying physics of audio. People then get pulled along with group think with the rest of the uneducated crowd.

The response is to "be safe" and throw money at their ignorance, buying things that make little or no difference at all. The deeper their pockets, the more esoteric they become, proud of their folly.

Perception is reality

Probably nowhere more true than with regard to audio

Thank you all for your replies. However, they were not helpful at all.

I'm well aware of all these "tips" that you have mentioned. I'm very experienced at signal integrity and I have designed a great number of complex and high-frequency (analog and digital) schematics and PCBs in my professional career.

In my present project I have to route some differential pairs of audio signals and I wondered if there is an "optimal way" to design these pairs (I mean the "optimal" geometry of each pair). I.e. how close can the (+) and (-) signals of the pair can be to each other? Or how much the traces' width should be? So, I'm talking about the width and gap of the (+) and (-) traces of each differential pair.

At audio frequencies you probably cannot physically construct any PCB features that would make any difference.

RF is a totally different critter.

Go with what's easiest.

http://www.icd.com.au/articles/Differential_Pair_Routing_PCB-Oct2011.pdf

That was interesting (although not helpful in my case).

It was interesting that the "loose coupling" is better than the "tight coupling", in the case where the two signals of the pair have to be further separated for a while (in order to come around an obstacle), as their impedance doesn't change much because of this.

However, I always prefer the "close coupling" (taking care not to change at all the gap between the traces throughout their entire length) as this approach provides a higher noise immunity (and lower radiation).

Thanks for being clearer.

The best thing is to just keep the lengths short and away from sources of stray noise. Yes, we said that already.

Differential audio, if it is the typical 600Ω balanced line, is too low an impedance and too high a signal drive to be concerned with the details you are worrying about.

How do you Brits put it? Don't over-egg the pudding!

If your traces are only a few cm long, 15 mil wide traces with a minimum 0f 15 mil spacing should be safe.

Yes, AH, I also believe that 15 mil width and 15 mil gap is OK in this case. This gap keeps the two traces close enough to each other (tight coupled) for high noise immunity.

Thanks.

Theoretically, magnetic pick-up is minimal if wires of pair are very close together to minimise loop area and electrostatic if wires are very small. However, you mention some pairs, so it's best not to put pairs close together (a strong + signal to earth on an adjacent wire is not differential even if it has a - wire beyond it) . It might be that your major noise source is magnetic from your power transfo - in that case traces on opposite sides of board, rather than side by side may present the least loop area to the field direction. Telephone cables use twisted pairs and transpose the pairs along the cable so that same pairs are not adjacent all through the run - but within the size of a PCB I doubt any benefit.