Phase and frequency question

[Keith_W]

LABELS, NORK!!!! 

[frednork]

Geez, reminds me of Hogans Heroes

[davewantsmoore]

6 hours ago, Keith_W said:

I spoke poorly. I have corrected myself. 

I wasn't meaning to say that (ie. it's not "your fault" anyone got confused)

 

All the terms and names which get thrown around are not what people should be focussed on learning... as it doesn't give them a true understanding of what is happening.

[davewantsmoore]

3 hours ago, sir sanders zingmore said:

So once more I'm not sure what the answer is?

You can apply a correction to a signal that alters the amplitude and phase independently.

 

The phase itself is not (within certain, quite high in terms of speaker, tolerances)  audible .... only the amplitude is audible.

 

Phase of signals which are being added together .... will affect the resulting amplitude  (which is how changes in phase can end up being audible).

[davewantsmoore]

2 hours ago, frednork said:

Thanks Dave, a not unexpected reply.  I would consider this still audible but do realise I am a lone voice here so will need to provide a bit more proof to back up my casual observation. 

 

Thanks for indulging me on these Dave and everyone.  This is the last one and it is of interest as it is similar to the last in that there is little difference below 100Hz but a much bigger difference above . As with all of these they track on amplitude at the same sort of accuracy I showed previously.  So is this audible or not you think.

 

There is an assumption I've made in my answers to you previous... and that is that all the drivers in the multiway speaker have the same crossover slopes and frequencies between each thing you're listening to.

 

For example what I think you're doing

 

3way speaker

24dB high pass and lowpasses at 400Hz and 4kHz

A and B have identical amplitude

A:  Linear Phase. B:  Minimum phase

 

.... rather than something like

 

3way speaker

A:  some set of crossover slopes and frequencies

B:  some set of different crossover slopes and frequencies

A = linear phase, B = minimum phase

A and B are filtered so they have an identical measured amplitude.

 

 

The reason is that in the second one ... with the different XO slopes and frequencies .... then A and B might have the same amplitude at your measurement axis (because you constructed a specific correction for force them to be that way) ...... but at other measurements axis' ... they won't  have the same response because of the different driver blending.

 

Hope that makes sense.

 

 

Anyways.  That's quite a lot of phase rotation (steep XOs I suppose).... but no, not audible.

 

 

Another (roundabout) way of saying/asking the above is to ask.... is the ONLY difference between all the filters for result A and result B the phase correction?     ie. the amplitude corrections are identical AND the measured amplitudes are identical.

 

 

[frednork]

Thanks Dave, appreciate what you are saying. What you are seeing is a passive 3 way being manipulated by FIR ,  The speaker is not changing, just the overall filter.

 

And again in this example I think it is audible, more easily than the previous but for different reasons.  I will see if I can easily verify any of this in an objective way.

 

Just for info here are the corresponding Amplitude, phase and step measurements for the 4 filters used in my examples for left channel only 

 

They are as you could see a minimum  phase correction, a linear phase correction ,  as well as a mixed phase correction and an amplitude only correction.

 

Amplitude

 

Phase

 

Step response

 

 

[sir sanders zingmore]

@davewantsmoore do you mind indulging me a bit while I see if I'm getting what you are saying? (although I must say the idea of doing this over a few wines (whiskies perhaps better) with a whiteboard sounds much more fun)

 

I understand this bit. Even that you might choose for whatever reason, to make a change to the phase of a signal but not make a change to the amplitude  

8 hours ago, davewantsmoore said:

You can apply a correction to a signal that alters the amplitude and phase independently.

 

This bit is fine. It confirms what I'd thought I'd learnt, but as mentioned, it's not my specific question

8 hours ago, davewantsmoore said:

The phase itself is not (within certain, quite high in terms of speaker, tolerances)  audible .... only the amplitude is audible.

 

This is the crux of the thing for me: As I understand the below statement, is it correct to say that unless you're emitting a single frequency tone, all signals will be added together. Even, as per @tripitaka's comment above, if they are all coming from a single source (single driver)? And therefore the resulting amplitude will be affected?
If that understanding is correct, that means the only way to change phase without affecting amplitude is when it is done for a pure single frequency tone.

 

8 hours ago, davewantsmoore said:

Phase of signals which are being added together .... will affect the resulting amplitude  (which is how changes in phase can end up being audible).

 

So my original question perhaps could be reworded to make my intended meaning clearer: 

"Is it possible to change the phase (but not the amplitude) of some frequencies in a full-range signal without affecting the amplitude of that full range signal"  

To which the answer (if I've understood this), is unequivocally "no it is not possible".

 

Or do we still need the whiteboard

 

[davewantsmoore]

8 hours ago, frednork said:

Thanks Dave, appreciate what you are saying. What you are seeing is a passive 3 way being manipulated by FIR ,  The speaker is not changing, just the overall filter.

That's what I thought.  (y)

 

8 hours ago, frednork said:

And again in this example I think it is audible, more easily than the previous but for different reasons.  I will see if I can easily verify any of this in an objective way.

Switch back and forth between the same amplitude filter, with different phases.

 

Without knowing which is which.

Arrange the test so that you "think it has changed", when in reality it might not have changed (ie. try to confuse yourself....ie. try to uncover the "just guessing" result).

Don't try and pick which is which to begin with.... just try to see I you can reliably tell when it has changed (to the other one), vs there was a short gap (like the "tell" from "the same are changing now), but it's the same one again... hope that makes sense.

 

[davewantsmoore]

21 minutes ago, sir sanders zingmore said:

This is the crux of the thing for me: As I understand the below statement, is it correct to say that unless you're emitting a single frequency tone, all signals will be added together. Even, as per @tripitaka's comment above, if they are all coming from a single source (single driver)? And therefore the resulting amplitude will be affected?
If that understanding is correct, that means the only way to change phase without affecting amplitude is when it is done for a pure single frequency tone.

 

Yes.

 

When you view the multi-tone waveform, it will have a different shape ... but the overall "frequency response" of the speaker will be the same (well, almost identical).

 

It's not audible.... within the bounds of phase distortion that you get in a typical speaker crossover/filter.

 

[Grizaudio]

9 hours ago, frednork said:

And again in this example I think it is audible, more easily than the previous but for different reasons.  I will see if I can easily verify any of this in an objective way.

 

What differences do you hear between the filters..? Is it transient differences, leading edges? dynamics? etc 

Edited May 26, 2024 by Grizaudio

[tripitaka]

3 hours ago, davewantsmoore said:

 

Yes.

 

When you view the multi-tone waveform, it will have a different shape ... but the overall "frequency response" of the speaker will be the same (well, almost identical).

 

 

This had previously been a paradox to me, but now seems ok. A waveform might be different because it contains different frequency or amplitude components or, crucially, it might only differ in the phase of one or more components.

In the latter case, it seems that the act of measuring (or hearing) will eliminate that difference, since the measurement of any given frequency will occur over a period of time. 

 

I suppose in our own ears, the nerve firings in response to stimulation at a given frequency must have some time band within which the brain is unable to register any delay.  Still, all bets are off when the phase mismatch becomes large enough, which accords with what @davewantsmoore has been telling us.

 

I dunno what we would register if the noise was steady state but with some frequencies phase-shifted, but then, who listens to white noise for pleasure?

 

 

Edited May 27, 2024 by tripitaka

[frednork]

11 minutes ago, Grizaudio said:

 

What differences do you hear between the filters..? 

It really depends on which you are comparing and it should be said all sound much better than unfiltered as we are minimising the impact of the room on bass frequencies  most obviously but there is more going on as well.

 

More  generally the IIR is reduced in dynamics  and transients are smeared compared with FIR . The minimum phase is excellent in impact and clarity and transients, The linear phase changes the character of the sound and makes it a bit smoother/softer sounding overall and a bit more  depth and 3d'ness but  less impactful in the bass and some evidence of preringing.  The mixed phase seems to bring the best of both minimum and linear phase options.

 

I wouldnt say the differences between all these are that large and if applying the 80:20 rule you would say that you have easily reached the majority of benefit with a good IIR (ie amplitude only) filter and after that improvements are much smaller. But for people that have spent significantly on their system or perhaps even for those who have not, the dollar spend on getting this level of improvement (which is highly significant)  is ludicrously low compared with pretty much all other upgrades. Should also mention that acoustic treatment in combination with this kind of dsp is the way to get the most benefit.

 

There is a bit of a learning curve however, so more difficult for those that are time poor or not interested/low aptitude in the technical side of things.

[tripitaka]

3 minutes ago, frednork said:

It really depends on which you are comparing and it should be said all sound much better than unfiltered as we are minimising the impact of the room on bass frequencies  most obviously but there is more going on as well.

 

More  generally the IIR is reduced in dynamics  and transients are smeared compared with FIR . The minimum phase is excellent in impact and clarity and transients, The linear phase changes the character of the sound and makes it a bit smoother/softer sounding overall and a bit more  depth and 3d'ness but  less impactful in the bass and some evidence of preringing.  The mixed phase seems to bring the best of both minimum and linear phase options.

 

I wouldnt say the differences between all these are that large and if applying the 80:20 rule you would say that you have easily reached the majority of benefit with a good IIR (ie amplitude only) filter and after that improvements are much smaller. But for people that have spent significantly on their system or perhaps even for those who have not, the dollar spend on getting this level of improvement (which is highly significant)  is ludicrously low compared with pretty much all other upgrades. Should also mention that acoustic treatment in combination with this kind of dsp is the way to get the most benefit.

 

There is a bit of a learning curve however, so more difficult for those that are time poor or not interested/low aptitude in the technical side of things.

You really are doing yeoman's work there, Mr Nork.

 

I guess it is nice for DSPers to think there may be more gains to be squeezed from all our FIR tinkering. Just to confirm, are these filters that Dave guessed would not make any audible difference?

[frednork]

1 hour ago, tripitaka said:

You really are doing yeoman's work there, Mr Nork.

 

I guess it is nice for DSPers to think there may be more gains to be squeezed from all our FIR tinkering. Just to confirm, are these filters that Dave guessed would not make any audible difference?

 

Yes mostly, @davewantsmoore has identified that there may be a change with the linear filter but none for the rest and he is absolutely correct in that the current public state of knowledge is that there is no difference with any of these based on the studies done so far. To his credit he has acknowledged there may be some differences even though that work has not been done to my knowledge. Others have identified some of these differences and have called upon the audio research community to do some updated studies to test these effects as earlier studies were not testing some of these specific scenario's, but am not aware of any done thus far as it is expensive difficult work to prove on a large scale as it involves lots of blind testing and its hard to see the ROI. Science is usually not very nimble. (especially if it involves using humans to detect something)

 

Edited May 27, 2024 by frednork

[Grizaudio]

46 minutes ago, frednork said:

Yes mostly, @davewantsmoore has identified that there may be a change with the linear filter but none for the rest and he is absolutely correct in that the current public state of knowledge is that there is no difference with any of these based on the studies done so far. To his credit he has acknowledged there may be some differences even though that work has not been done to my knowledge. Others have identified some of these differences and have called upon the audio research community to do some updated studies to test these effects as earlier studies were not testing some of these specific scenario's, but am not aware of any done thus far as it is expensive difficult work to prove on a large scale as it involves lots of blind testing and its hard to see the ROI. Science is usually not very nimble. (especially if it involves using humans to detect something)

 

Mark, I'm definitely no expert, but have you tried comparisons blind?

Or with a group of participants .... at the very least this would rule out any preconceived filter bias and psychoacoustic elements - which I'm sure would exist.  

 

It does raise the question how much of theory, stacks up in real world listening, using modern compute and FIR tech, brain and hearing science.  

 

Check this study out:

https://www.researchgate.net/publication/286939870_Sensitivity_of_Human_Hearing_to_Changes_in_Phase_Spectrum

 

Edited May 27, 2024 by Grizaudio

[Satanica]

4 hours ago, sir sanders zingmore said:

So my original question perhaps could be reworded to make my intended meaning clearer: 

"Is it possible to change the phase (but not the amplitude) of some frequencies in a full-range signal without affecting the amplitude of that full range signal"  

To which the answer (if I've understood this), is unequivocally "no it is not possible".

 

That is how I understood your original question and I don't think that's been universally understood.

[frednork]

18 minutes ago, Grizaudio said:

 

Mark, I'm definitely no expert, but have you tried comparisons blind?

Or with a group of participants .... at the very least this would rule out any preconceived filter bias and psychoacoustic elements - which I'm sure would exist.  

 

It does raise the question how much of theory, stacks up in real world listening, using modern compute and FIR tech. 

 

Check this study out:

https://www.researchgate.net/publication/286939870_Sensitivity_of_Human_Hearing_to_Changes_in_Phase_Spectrum

 

 

Yes, but informally and not particularly rigorously. I am interested to see if I can do something a bit more convincing without inordinate effort.

 

Thanks also for the paper will have a good read. I suppose a lit review would be helpful but it does equal lots of effort and no remuneration.

[Keith_W]

I hope this helps clarify whether phase is audible or not. 

 

You take a pair of speakers and tune them so that they have exactly the same frequency response and the same minimum phase response. Hopefully the loudspeaker manufacturer has done this for you. You then take an all pass filter. All pass filters only rotates phase, and does not change the amplitude response. Adjust the AP filter so that it only affects a limited frequency band, say 200Hz - 2kHz. 

 

1. Slap the all pass filter on both speakers. Is there an audible difference before and after?

The answer is no. This is what Dave has been saying. 

 

2. Slap the all pass filter on ONE speaker only. You have now created a phase asymmetry between left and right speaker. Is there an audible difference before and after?

The answer is YES. Very easily audible. The reason is because it changes the ITD (interaural time delay). An AP filter in both speakers does not change the ITD. The maximum ITD under normal conditions is dependent on the width of your head (and therefore the spacing between your ears). The average width of the head is 15-18cm, or about 0.05ms. So you only need a time discrepancy of > 0.05ms between your ears to hear a difference. Studies show that we are most sensitive to ITD differences between 100Hz up to 2kHz. Humans use ITD to determine direction of sound. We were evolved to have ITD and detect phase differences. 

 

So a 0.05ms time discrepancy between left and right speaker is the same as a 36 degree phase difference between speakers at 200Hz. At 2kHz it is a 360 degree phase difference. Do the math yourself, theta = t (360*f). 

L-R phase asymmetry is not heard as a change in tonality. It's a time related phenomenon, so errors are heard in spatiality, e.g. it might pull the phantom image to one side or another, appear to contract the soundstage, and so on. 

 

If we accept the answer to this question, we move on to the next question. 

 

3. Is the MEASURED phase at the listening position analogous to slapping an all pass filter in one speaker and not both? 

The answer is again YES. This is because the non-minimum phase behaviour of the left and right speaker will be different, especially so if you have asymmetrical placement of the speaker in the room, funny shaped room, furniture on one side of the room and not the other, etc. Phase asymmetry is created in the same way that an all pass filter in one speaker creates phase asymmetry. Your ears don't care if the asymmetry was created by an AP filter or non-minimum phase behaviour, it still engages the ITD mechanism. 

 

OK we now move on to the next question. 

 

4. When you make a loudspeaker correction, you place the mic at the listening position and take a measurement. This will contain the min phase loudspeaker response + the non-minphase response of the room. You extract the minphase response and perform corrections on it. What do you think will happen if your DSP software fails to extract the minphase response properly? Maybe you set the wrong windowing? Maybe you took the measurement improperly? 

Answer: if your DSP software thinks there is phase asymmetry in the minphase response between left and right speakers where there is none, it will try to correct it. This mistake propagates all the way down into your final correction filters. If there wasn't phase asymmetry in your speakers before, it will be there now. 

Edited May 27, 2024 by Keith_W

[tripitaka]

4 minutes ago, Keith_W said:

I hope this helps clarify whether phase is audible or not. 

 

You take a pair of speakers and tune them so that they have exactly the same frequency response and the same minimum phase response. Hopefully the loudspeaker manufacturer has done this for you. You then take an all pass filter. All pass filters only rotates phase, and does not change the frequency phase response. Adjust the AP filter so that it only affects a limited frequency band, say 200Hz - 2kHz. 

 

1. Slap the all pass filter on both speakers. Is there an audible difference before and after?

The answer is no. This is what Dave has been saying. 

 

2. Slap the all pass filter on ONE speaker only. You have now created a phase asymmetry between left and right speaker. Is there an audible difference before and after?

The answer is YES. Very easily audible. The reason is because it changes the ITD (interaural time delay). An AP filter in both speakers does not change the ITD. The maximum ITD under normal conditions is dependent on the width of your head (and therefore the spacing between your ears). The average width of the head is 15-18cm, or about 0.05ms. So you only need a time discrepancy of > 0.05ms between your ears to hear a difference. Studies show that we are most sensitive to ITD differences between 100Hz up to 2kHz. Humans use ITD to determine direction of sound. We were evolved to have ITD and detect phase differences. 

 

So a 0.05ms time discrepancy between left and right speaker is the same as a 36 degree phase difference between speakers at 200Hz. At 2kHz it is a 360 degree phase difference. Do the math yourself, theta = t (360*f). 

L-R phase asymmetry is not heard as a change in tonality. It's a time related phenomenon, so errors are heard in spatiality, e.g. it might pull the phantom image to one side or another, appear to contract the soundstage, and so on. 

 

If we accept the answer to this question, we move on to the next question. 

 

3. Is the MEASURED phase at the listening position analogous to slapping an all pass filter in one speaker and not both? 

The answer is again YES. This is because the non-minimum phase behaviour of the left and right speaker will be different, especially so if you have asymmetrical placement of the speaker in the room, funny shaped room, furniture on one side of the room and not the other, etc. Phase asymmetry is created in the same way that an all pass filter in one speaker creates phase asymmetry. Your ears don't care if the asymmetry was created by an AP filter or non-minimum phase behaviour, it still engages the ITD mechanism. 

 

OK we now move on to the next question. 

 

4. When you make a loudspeaker correction, you place the mic at the listening position and take a measurement. This will contain the min phase loudspeaker response + the non-minphase response of the room. You extract the minphase response and perform corrections on it. What do you think will happen if your DSP software fails to extract the minphase response properly? Maybe you set the wrong windowing? Maybe you took the measurement improperly? 

Answer: if your DSP software thinks there is phase asymmetry in the minphase response between left and right speakers where there is none, it will try to correct it. This mistake propagates all the way down into your final correction filters. If there wasn't phase asymmetry in your speakers before, it will be there now. 

@Keith_W, the matter under close scrutiny for the last coupla pages is buried as a sub-item in your item #1, which you haven't acknowledged.

 

I think everyone is on board with the rest of what you wrote though the inclusion if item#3 is a useful addition to the discussion

[Keith_W]

1 minute ago, tripitaka said:

@Keith_W, the matter under close scrutiny for the last coupla pages is buried as a sub-item in your item #1, which you haven't acknowledged.

 

Item #1 is uncontroversial. Nobody disputes this. The confusion comes from conflating the rest of the points with the first, especially the last point. 

[tripitaka]

1 minute ago, Keith_W said:

 

Item #1 is uncontroversial. Nobody disputes this. The confusion comes from conflating the rest of the points with the first, especially the last point. 

Every mention of the word 'filter' is an opportunity for loss of meaning IMO but, assuming I've  understood your intention, if you phase shift a narrow band by enough it obviously IS audible. 

 

My reading of the discussion is that the discussion has focussed on why it isn't ALWAYS audible.

 

 

[Grizaudio]

The reason I posted the link was for these highlighted conclusions:

Edited, these tests were conducted using, Sennheiser HD650 headphones... 

 

Extracted from: https://www.researchgate.net/publication/286939870_Sensitivity_of_Human_Hearing_to_Changes_in_Phase_Spectrum

 

Edited May 27, 2024 by Grizaudio

[Grizaudio]

Extracted from: https://www.researchgate.net/publication/286939870_Sensitivity_of_Human_Hearing_to_Changes_in_Phase_Spectrum

 

Here: Red brackets 

Edited May 27, 2024 by Grizaudio

[Grizaudio]

The last paragraph is also very interesting. 

[frednork]

22 minutes ago, Keith_W said:

 

Item #1 is uncontroversial. Nobody disputes this. The confusion comes from conflating the rest of the points with the first, especially the last point. 

So , do you think this is why I hear a difference with these filters?