Does Phase Distortion/Shift Matter in Audio? (no*)

[markus]

Ah, nevermind... But I get what you mean about the desirability of having all-pass filter(s). Just so you know, I do not use any in my current implementation and am perfectly fine -- as far as I am aware, there are no cancellations between any or all channels.

As long as there's a room there are cancellations Some can be cured by shaping the direct signal, some can't.

 

[ernestcarl]

As long as there's a room there are cancellations Some can be cured by shaping the direct signal, some can't.

I meant *significant cancellations in the summed responses around the xo zone of interaction. Yep, there are very small ones here and there due to the room inevitably mangling the phase, of course. Why no all pass works for me is because the LFE channel shares the same 120Hz LPF as the bass managed section. Phase response between all channels are complementary/match after time alignment even though I'm using an overlapping xo with 80Hz HPF for all satellites.

 

[tomeh]

You still seem to think I didn't do such experiments*? Not sure why.

Anyhow, I'm all for the scientific method and I don't question the results, I'm just sceptical about the theory explaining the perception. I think the reason is much simpler, namely better interchannel tracking. But don't take my word for it. A proper experiment could shed some light on this. Unfortunately the people interested and willing to do proper testing are scattered all over this planet and can't work on things in an effective manner.

---
*One of the informal experiments I did many moons ago involved comparing a 2-way speaker and a 3" full range driver in an open baffle very similar to what you described above. Listening was fairly near. I'd say less than 6' with the walls as far away as possible.

The perception was exactly as you described it. Much better imaging with the full ranger.

Then I equalized (with Acourate) both speakers so they had a very similar frequency response. This involved limiting the low fequency response of the 2-way speaker. At the same time I optimized the phase response within each pair so it was better matched. The phase response between 2-way and full range was NOT matched.

Surprising result: both speaker types now imaged the very same.

This drives my personal theory up to this day – frequency response (magnitude AND phase) of stereo speakers should track each other as close as possible. Overall phase response is largely irrelevant as long as it matches between stereo pairs.

Then I also compared imaging of the 2-way speaker with just the low frequency limit engaged. Imaging improved as well. Second take away for me: overall bandwidth has a significant effect on imaging. Here the room dominates what is heard, so great care needs to be taken at the low end (room treatments and EQ).

What was the low frequency limit engaged on the 2 way?

Thanks

 

[OK1]

Ok if phase is not something we should bother about, and I accept that this may be so, is there then any value in aiming to time align the arrival of bass and high frequencies, in the DRC (Digital Room Correction) process, using whatever methods are relevant?

Thinking that the difference in distance from the listening position, of tweeter and woofer, as well as any timing differences from phase shift introduced by crossovers, may still need to be compensated for, if they are material, to the listening experience.

Still trying to figure out if there is a difference in the timing delay of different frequencies, due to phase shift, and the kind of timing delay that is caused by difference in arrival time of bass and high frequencies, due to measurable difference between the distances from the listener to the tweeter and woofer, for example in a 2 way speaker.

Of the top of my head, my thinking is that the maximum timing delay due to phase shift, will be the time span of a single cycle. i.e. the maximum phase shift will be in degrees 360 degrees. Do let me know if EQ changes such as the kind in minimum phase cross overs (i.e. normal eq either analog or minimum phase digital) can produce phase shifts, that exceeds 360 degrees. If my assertion is true then the maximum timing change will depend on frequency.

Taking a 40hz signal, maximum phase shift here would be 1 second divided by 40, i.e. 25 milliseconds delay. And at the upper end, frequencies like 12k, would have a maximum timing delay of 1 second divided by 12,000 = approx. 0.000083 seconds or 8 hundreth of a millisecond.

 

[Newman]

Ok if phase is not something we should bother about, and I accept that this may be so, is there then any value in aiming to time align the arrival of bass and high frequencies, in the DRC (Digital Room Correction) process, using whatever methods are relevant?

I may have mentioned this before, but frequency response changes are audible, so a phase change that causes an audible change in FR is to be avoided, even though the phase itself is not audible.

 

[Holmz]

I may have mentioned this before, but frequency response changes are audible, so a phase change that causes an audible change in FR is to be avoided, even though the phase itself is not audible.

I am not sure what you mean by phase is not audible?

How many degrees or cycles can we rotate things before that are audible?

 

[Newman]

A lot, as long as it is all-pass phase shift.

I thought I had a reference handy, but I can't find it.

I do recall Toole's words, "...It turns out that, within very generous tolerances, humans are insensitive to phase shifts. Under carefully contrived circumstances, special signals auditioned in anechoic conditions, or through headphones, people have heard slight differences. However, even these limited results have failed to provide clear evidence of a 'preference' for a lack of phase shift. When auditioned in real rooms, these differences disappear..."

Like I said earlier, what is audible, is changes to the frequency response that can result from phase shift, eg delaying one driver compared to another that is playing some of the same content. People tend to conflate the two things, but shouldn't.

 

[René - Acculution.com]

I will soon be doing a third video with Erin's Audio Corner, and the topic this time is phase. I will go through how phase comes from complex numbers, phasors, phase delay, group delay, latency, phase when summing/crossover phase, allpass vs time delay, phase vs polarity vs delay, how viewing systems from their steady state responses only will lead to the confusion about how phase and time relate and why the transient response also must be considered, and total phase of a loudspeaker. If you have questions that you would like me to tackle, let me know.

 

[j_j]

I thought I already said this, but phase shift that is exactly like phase = 2 * pi * f * t, where f and t are frequency and time respectively, is nothing but a pure delay.

In order to determine the actual "phase shift" you have to remove the linear component (i.e. the delay). Now, if that phase shift approaches 20 degrees or so inside of one ERB (cochlear filter bandwidth) phase shift is probably audible.

If a slower phase shift that delays one part of an impulse substantially compared to the rest is added (always talking about allpass filters here, so there is no change in frequency response whatsoever) then you will get a "hollow" or "ringy" kind of sound.

If, however, there is substantial phase shift with two tones, say 60 Hz and 7kHz, it doesn't matter. At no point are they both detected at the same place on the cochlea. So it's completely insensitive.

This is much more sensitive interaurally, surprisingly small amounts of differential delay (between the ears) will do interesting things to the stereo perception in headphones.

In loudspeakers, the room interferes so very much in almost all cases that it jsut doesn't matter.

 

[j_j]

Like I said earlier, what is audible, is changes to the frequency response that can result from phase shift, eg delaying one driver compared to another that is playing some of the same content. People tend to conflate the two things, but shouldn't.

This is untrue, and trivially demonstrated using a schroeder section with about a millisecond of delay and loop gain of .5. Put anything impulsive through it, single channel, and you'll hear it.

100% allpass.

 

[Newman]

Take it up with Floyd, jj!

 

[j_j]

Take it up with Floyd, jj!

I think you rather misrepresent Floyd's position vs. mine. We disagree on magnitude. Your assertion that it's 100% due to frequency response anomalies is just plain wrong, and Floyd doesn't claim that, either.

Try the experiment I just suggested. That's how science works. Do the experiment, and then you will know the answer. Seek, and ye shall find.

 

[restorer-john]

Much was made of the phase shift (interchannel delay) in the early days of compact disc players, where the left and right channels were decoded with a multiplexed single D/A converter, shared between channels.

I didn't know anyone who could demonstrate they could hear a difference at the time. When using difference processors (early dolby surround) some effects could be heard.

This is the actual measured interchannel difference, ~11uS @20kHz for one of my Sony CDP-101s from 1983:

 

[Newman]

I think you rather misrepresent Floyd's position vs. mine. We disagree on magnitude. Your assertion that it's 100% due to frequency response anomalies is just plain wrong, and Floyd doesn't claim that, either.

Try the experiment I just suggested. That's how science works. Do the experiment, and then you will know the answer. Seek, and ye shall find.

If you don’t disagree with Floyd’s assertion that it is unimportant for music playback, then I’m happy as is.

 

[j_j]

Much was made of the phase shift (interchannel delay) in the early days of compact disc players, where the left and right channels were decoded with a multiplexed single D/A converter, shared between channels.

I didn't know anyone who could demonstrate they could hear a difference at the time. When using difference processors (early dolby surround) some effects could be heard.

This is the actual measured interchannel difference, ~11uS @20kHz for one of my Sony CDP-101s from 1983:

This is right on edge of threshold, and it's constant, which is to say that it's a very, very small, fixed shift in image. While it is probably 'audible' if you can compare it in an ABX with the same device not having the half-sample delay (good luck with that, unless you the test at 88 and simply shift one channel one sample vs. not shifted), it's very likely not to matter.

Where it mattered to some or so they objected vociferously in the AES standards meeting, is in summing to mono, where you do get a bit of a lowpass filter. That *IS* audible if you're 12 years old and never listened to a fireworks display.

(seriously, it is detectable by ear if you can hear 20K, still). Does it matter? Well, some people insisted it was the end of the world. I had other things to worry about.

 

[j_j]

If you don’t disagree with Floyd’s assertion that it is unimportant for music playback, then I’m happy as is.

As it happens, yes, I do disagree that it's inaudible. Does it matter, that depends on who you ask. Enough all-pass annoys the crap out of me.

 

[Holmz]

I will soon be doing a third video with Erin's Audio Corner, and the topic this time is phase. I will go through how phase comes from complex numbers, phasors, phase delay, group delay, latency, phase when summing/crossover phase, allpass vs time delay, phase vs polarity vs delay, how viewing systems from their steady state responses only will lead to the confusion about how phase and time relate and why the transient response also must be considered, and total phase of a loudspeaker. If you have questions that you would like me to tackle, let me know.

Can you let us know when?

(I always seem to miss EAC live streams, and I am late to the party in general.)

 

[René - Acculution.com]

Can you let us know when?

(I always seem to miss EAC live streams, and I am late to the party in general.)

Sure, but I am not sure anymore if it is okay to make a post that links to another site like that (as I understand it, Erin was not allowed to). Also,I doubt it will be live as it will probably be a two hour lecture.

 

[dasdoing]

 

[René - Acculution.com]

You do that test the other way round. Take a system with proper "full-blown DRC" and resulting smooth FR and flat phase response from each speaker. Then introduce a analytically calculated phase distortion which mimics for example the phase rotations from a typical 3-way passive speaker. This can then be A/B tested (or ABX) and then most people will find that phase distortion is audible and can be detected (once you know what to listen for and have the right music tracks that highlight the effects) but the difference is very subtle, the proverbial icing on the cake. Channel matching of FR and phase is much more important, exactly as you say.
Phase contribution of electronics... forget about it, except for very pathological cases.

I am on board with this. Also, it is interesting that people will argue about audible differences between cables, where there is pretty much no differences to measure, and at the same time have no issue with nonlinear phase, where the temporal distortion is easy to illustrate. Audibility is of course up for discussion, but it would be easier to argue for audibility of phase than of cables.