Mixing: when to correct phase?

[GibsonGM]

Here's one that's been bugging me for a while...if you split/blend 2 signals, at what point is it NECESSARY to correct for different phases?  Is it only a 'by the ear' thing, or is there some easy DIY universal 'adjustable phase corrector' out there that i don't know about? I mean, you shift a bit every time you go thru a cap; it's inevitable that there will be SOME shift if you're doing parallel stuff.

Example:  I have split and blended the output from a basic fuzz and a Guvnor.  One of them (forget which, it was the other day) is a little out of phase WRT the input and other distortion's signal.  Maybe 20 degrees or so as viewed on the scope.

I don't hear anything odd because of this, so I am assuming it's fine to just leave it.  (?)    What say the experts who've done a lot of parallel effects work??  My only concern would be that at some distortion level, or blend level, or playing some random combination of notes I haven't hit yet with some random effect before the parallel combination, the thing could go into some really strange funk like a ring mod, ha ha....no reason to think this would happen, I just don't know what's 'standard procedure'.

The 'adjusters' I've seen look like they are about phase inversion rather than any mid-point adjustment, so I am again gonna assume you don't need to do anything unless it sounds like junk, and then you try inverting one of the signals?   

[R O Tiree]

As you said, there is some shift every time you meet a cap in the signal path, but it's frequency-dependent, obviously. A 20-degree shift is irrelevant... Cos(20) is 0.94, so mixing a 100% signal with a 94% signal will probably not be detectable. Even a 45-degree shift is going to get you 71% or so wrt input, which is probably fine.

[blackieNYC]

So you've seen RG Keen's inverter.  A 1:1 transformer can work too.  (Easy to have sitting beside your breadboard)
It is by ear, and not worth addressing if you are not talking about 180 degrees.   Or thereabouts - the smaller phase differences would take a lot of corrective effort, which I'm sure someone would be glad to share with us.
The sum of the two signals reaches a maximum of twice the voltage and a minimum of zero, when your blend adjustment is just right.   No fear of anything wacky, or a Tacoma Narrows Bridge effect. Phase relationships in between 180 and 0 will have slight cancellation and this reduce in overall level, low end loss is the first to go theoretically.  But it can also sound interesting, like putting two mics on a guitar.
Another way , other than by ear, would be to look at the schematics of your pedals, and see if you can count up the number of inverting op amp/transistor stages in the signal path ( not the LFO or something truly ancillary).   Odd number of phase inversions is a total phase reversal.

[R O Tiree]

I just did some fiddling with a spreadsheet... My previous reply was a little simplistic, as I only considered the resulting amplitude when one signal hits its peak... and in any case, I didn't complete my sums, because 100% averaged with 94% (mixed, in other words) works out at 97%. Continue the analysis to look at where the resultant waveform peaks is where the 2 signals cross over - one just after it has peaked and the other just before it has peaked. The resultant is about 98%. You're definitely not going to notice. And a 45-degree shift gets a peak at about 93%, so you're still not going to notice. Going to a 90-degree shift, we get a resultant of 71%, which will be noticeable.

[R.G.]

This is a good point to talk about phase.

In my mind, I make a distinction between fixed and variable phase. This is not a real-world distinction, since fixed phase is a special case of phase in general.

Fixed phase is an abstraction used in transformers and some audio work. The idea here is that a signal is either in phase (i.e. zero degrees of shift) or out of phase (180 degrees); one of the two, no funny in-betweens. We also consider opamps to have inverting or non-inverting qualities.

As I said, this is an abstraction. In general, fixed-phase only applies in circuits at low frequencies (to the circuit). As noted, any time a signal goes through a reactive component - a cap or an inductor - its phase is changed. Well, its voltage is changed in phase compared to the current through it, and vice versa. And the phase shift varies with frequency when it's compared to voltage/current across/through resistors or the other-sex reactive components associated with it in the circuit.

In practice, you can only have fixed-phase (true/inverted) when the signal frequency is well away from any reactive rolloffs. Generally, this means 10x higher than any low frequency pole or 10X lower than any high frequency pole. Opamps and other circuits can make this true over a wide range of frequencies if you design for it. Transformers have a "mid-band" where it's true for them, and the criteria for goodness of an output transformer generally acknowledges this by being measured as the ratio of the primary inductance to the leakage inductance; the bigger the better, and this translates to a bigger mid-range with no phase or amplitude funnies.

In actual practice, if you've included a midband with filtering in it, true/inverted phase is no longer a useful concept. The phase should be considered variable.

And on the original question, whether phase shifts NOT zero or 180 should be compensated, the better question is CAN they be compensated. They can, but it requires a huge amount of work to do this over any wide frequency band. So my advice is to ignore it until you absolutely know it's giving you grief.

Audio delays give another set of issues with phase changes. Once the time delay exceeds the period of a signal, what exactly "phase" means starts to get really odd. Flangers use the phase shifted cancellations to make comb filters - many spaced notches in amplitude response when mixed with the dry signal, and very odd resulting phase shifts on either side of the notch after mixing. Once you get to reverb, echo, and recirculating delays, what phase means exactly at the end of the process is truly muddy.

[ashcat_lt]

RG at the end there started to lean toward this, but it's worth noting that "phase problems" only really happen when the two signals would otherwise correlate rather closely.  To simplify - if you take two signals and invert one (so that all frequencies appear to be 180 degrees out of phase) and mix them, they will cancel completely when mixed in exactly equal amounts.  Turn either one up or down and they won't cancel completely.  Yeah, cause 1 + (-1) = 0, while 1 + (-0.9), 0.9 + (-1), etc do not = 0.  Right?  And this is true of frequency specific phase shifts as well - if the frequency in question is much louder in one signal than another, you won't hear much interference (constructive or destructive) in that range.

The reason that slap-back delay type thing doesn't cause as much comb filtering as the shorter delays in a flanger is that the delay is long enough to where the original signal has usually "moved on" from what the repeat is doing.  The actual signal content is different enough that they just can't really line up enough to interfere noticeably.

And I'm thinking that if you're distorting one signal and not another, and you mix them together, the signals are going generally to be different enough - especially in the higher frequencies - that you won't be able to tell anyway.  Even mixing two different distortions, each gives what is essentially a random shotgun blast of harmonics, and it would be completely accidental if the many of the pellets from the one blast went through the holes made by the other.  It's a good idea to check for absolute polarity/phase smearing in the lower ranges as it's maybe a bit more likely that the fundamentals will be correlated closely enough to cancel.  That is, you might have some "thinning" from the fundamental and/or first couple harmonics canceling, but one might not expect a lot of that comb filtered "phasey" sound.

All that said, "accidents" like I described can and do happen in ways that are sometimes difficult to predict.  To extent, that's exactly what a mix does.  It is worth trying flipping absolute polarity on one of the signals just to check and see if you like it any better.  That, I think is best done by ear.

There are boxes sold as "phase correction" tools for studio use in instances where one might be mixing say a DI and a miked amp, or a couple of drum mics and getting comb filtering from the relative delay between them.  These are generally just all-pass filters with variable center frequency - just like the stages in our phaser pedals.  The idea being that the frequency dependent phase shift just might accidentally undo whatever problems you're hearing, or at least move them to a frequency range that is less critical to the application.  All-pass filters are all kinds of fun!

[GibsonGM]

Thank you VERY much, everyone, for the awesome thoughts!  There are many educational nuggets of goodness in all these replies!   In reading, I've seen yet another use for a phenomenon like phase cancellation, ha ha.   

I did think it was something you ignored unless it was a problem, and here it's confirmed.  And also that you can't count on a very 'static' condition, but that your phase issues are liable to be variable as signals are constantly changing.     I think most of us have wired a pickup out of phase; the case is different when mixing signals from parallel 'normal' circuits, though.

Didn't know about the cosine thing, either ;o) Thanks, makes perfect sense once you consider it.

[MrStab]

the better question is CAN they be compensated

There are boxes sold as "phase correction" tools for studio use

i picture some kinda key-in type device with microcontrollers being needed for precise phase matching. i wonder what such exact compensation would sound like - probably quite unnatural if we're all used to the current history of electric guitar & audio engineering, no?

thanks for the explanations, RG & Ashcat. the whole phase thing has eluded my understanding for years but i think i'm finally getting it.