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

[amirm]

Is a single bit error, or all zeros with a 1 or all/somebits being non-zero also illegal?

It absolutely is. Nyquist theorem mandates that the highest frequency in the file be half the sample rate. What you created above has far higher bandwidth meaning it cannot be correctly reconstructed. It will cause ringing, etc. that is not going to be there with a band limited signal.

 

[amirm]

It seems like a bit error in a digital file, or a pop or click on an LP would also then be illegal… but as they seem to happen, they then must be a different law than the laws of physics and engineering.

The two are different animals but still suffer from the same misunderstanding about an impulse signal. The click from an LP would be bandwidth limited and shaped. It will not remotely represent a true impulse. Nor would its playback require any kind of fidelity, sans not clipping it.

 

[pma]

It is a matter of terminology, when we go nit-picking. The Fourier transform of a transient signal, which is a signal that exists only for a finite duration, results in a spectrum that extends to all frequencies, meaning it has infinite bandwidth.

 

[amirm]

Impulse response of a linear system is an inverse Fourier Transform of Frequency Response.

Those are two sides of the same coin so your argument makes no sense. To present an ideal impulse, you would need infinite frequency sweep. Typical frequency sweeps to 20 or 30 kHz is very, very far from that. As a result, the computed impulse response would be very imperfect and nothing like the poster was assuming.

REW or Arta calculate impulse response from frequency response.

That is one implementation method (log sine sweep/CHIRP). There are other schemes including an actual gun shots and such used in the olden days. Regardless, as I explained above, how we measure a system has no bearing on the poster aspiring to have perfect timing where laws of nature does not allow it.

 

[pma]

Of course signal theory is a signal theory and auditory limits are a different discipline.

 

[JaapD]

Regarding Impulse response and FFT:

Do take both amplitude and phase response into account here. The amplitude response only tells half the story => complex Fourier transformation.

The usual frequency response plot never describes inrush or settling aspects, it is missing any dynamic behavior so to speak. This is simply because the applied sinewave sweep is a relatively ‘easy’ signal, not containing any dynamic related contents (just a constant voltage [correction: amplitude] and a slowly changing frequency).

Thought experiment: think for instance a tweeter with a relatively heavy cone and another one with a very small mass, both capable of delivering 20 KHz in the frequency domain. The time domain and their impulse response may be very different though (relatively long for the heavy cone, short for the other one).

Similar example: put a 20 KHz tone burst on both tweeters. After the tone burst stops the one with the heavy mass will slowly play quieter (exaggerating a bit, to clarify things) while the other one immediately stops playing.

 

[pma]

This is obvious. Frequency response = amplitude response and phase response. Impulse response contains both. In a physically realized system amplitude and phase response are related via Hilbert transform. Frequency response may also by described by real and imaginary parts, again related via Hilbert transform.

 

[MaxwellsEq]

This is simply because the applied sinewave sweep is a relatively ‘easy’ signal, not containing any dynamic related contents (just a constant voltage and a slowly changing frequency).

That doesn't accurately characterize a sine wave.

Consider a 1kHz sine wave at the output of a power amplifier: sometimes (at the zero crossing point) the voltage is exactly zero, but the rate of voltage change is high, so it sweeps towards the positive rail. It's now a high, positive voltage, but it's soon heading back towards zero. Then it moves quickly towards the negative rail, before diving back to zero. There's no "constant voltage" in a sine wave.

 

[Holmz]

It absolutely is. Nyquist theorem mandates that the highest frequency in the file be half the sample rate. What you created above has far higher bandwidth meaning it cannot be correctly reconstructed. It will cause ringing, etc. that is not going to be there with a band limited signal.

How can a bit error be illegal?
It is either happening or not happening.

Secondly, what relevance does an FFT have to do with anything whatsoever ever?
There is a time-domain stream of samples at some sample rate, and those are getting converted into by a DAC into voltages.

Where is the law being violated there?
What law izzit?
Nyquist says that the 44.1k samples/sec can only carry info to the Nyquist frequency or 22 kHz or less.
Ok - fine.

What do any of these strawman arguments have to do with a step function response going out to ~10 ms?
And the math to get to an impulse response is also something that I think that the Klippel does.
Are they violating the law too?

But I thought that we were sort of discussing speaker phase.

 

[JaapD]

That doesn't accurately characterize a sine wave.

Consider a 1kHz sine wave at the output of a power amplifier: sometimes (at the zero crossing point) the voltage is exactly zero, but the rate of voltage change is high, so it sweeps towards the positive rail. It's now a high, positive voltage, but it's soon heading back towards zero. Then it moves quickly towards the negative rail, before diving back to zero. There's no "constant voltage" in a sine wave.

Correction: constant amplitude.

 

[gnarly]

As such, it is an "illegal" signal, certainly in the context of bandwidth limited digital music. It also has infinite rise time, again an illegal real life signal.

An impulse is a conceptual signal that is useful in engineering analysis but you don't want to confuse yourself into thinking your system needs to reproduce it with high fidelity. Such an attempt will lead to designs that have very wide bandwidth but otherwise, have poor response (former doesn't come for free).

It sounds to me like you are talking about a theoretical infinite bandwidth Dirac pulse. And using it as a stimulus signal.
That is not a system impulse response.

Every audio speaker, every driver, has a measurable impulse response that reflects its bandwidth.

This is measurement 101 stuff....I'm at a bit of a loss trying to understand why you are calling an impulse response a conceptual signal, with no value for speaker development.
Unless you truly see no value in time and phase at all???

 

[amirm]

How can a bit error be illegal?

The term "illegal" is a formal term used in audio and video technology. As an example of latter, broadcast/professional video defines luminance values from 16 to 255. Whereas for computers, we have 0 to 255. A value of say, 8 is valid for computer RGB signal but "illegal" for video. You obviously can create video samples that are below 16 but you can't expect the system to behave any predictable way if you attempt to play it. You could get clipping, or other artifacts.

By the same token, a digital sample going from dead zero to max amplitude is "illegal." It was not prefiltered as Nyquist theorem demands. As a result you get artifacts that would not be of interest with real content.

 

[Holmz]

I feel compelled to reply here, in the hope of setting a stage for better understandings.

I didn't say NFS can't measure phase, I made no such claim. I know full well NFS can. I've watched Erin's interview with Bellman several times all the way through, and have read/downloaded all of NTK's NFS posts. Raving at me for making false claims and not understanding NFS is simply off the mark.
I commented on spins not showing phase, like the spins you provide.

A quick search of my files shows I've saved over 6200 dual-channel transfer functions via Smaart, over the last decade, building and tuning DIY speakers.
I probably take at least 5 transfer functions for every one saved, because Smaart makes real-time measurements, allowing real-time alterations of prototypes and fixing missteps.
I have saved 320 REW mdat files which probably have an average of 15 measurement each. REW is used on finished builds, to check distortion, multitone, and CEA2010 on subs. Hundreds of saved REW scope images, microphone captures of wavelet transients.
Over 50 saved Crosslite+ project files, an advanced measurement & processing simulation program. Each CL+ project usually has around fifty to one hundred impulse & transfer functions in them.
All are acoustic measurements of speakers, for making the best quasi anechoically tuned speakers I can, at home.

And am certainly not running on intuition alone. I've had 8 full days of formal training in acoustic measurements, and regularly attend a weekly online training meeting.
Honestly, if you knew more about me and the work I've done, you'd see that i fit in the top of this highly informed thread.

I'm sorry my views challenge your status quo views. I feel mine are in sync with the research you posted, with the I highlights I added.
Like those authors, I believe more work needs to be done to ascertain the limits of phase audibility. Especially given that speaker technology has improved to where speakers can be made linear-phase, for a truly valid comparison.



I have a permanent outdoor vertical multi-mic with spinorama turntables, set up for main speakers. Subs get ground plane measurements, sometimes including CEA 2010.

DIY projects have spanned traditional speaker builds, PA MTM, PA modular system, floor to ceiling line arrays, CBT line arrays, coaxial design, a ring radiator array, a dodecahedron, nearly a dozen versions of synergy/MEHs, and 7 sealed and ported sub designs... over 50 cabinets in all.

Well - we at the point where we know that the Klippel does indeed do phase measurement.
And we know that not everyone thinks that phase response is worth knowing, and some do not show the Klippel phase.

@j_j mentioned that the phase can be adjusted to give the same magnitude, but when doing so that the system sounds different when the two files with identical FR are played.

And we know that in the time-domain, the only real metric for fidelity to be possibly conjured, would be to say how closely the system’s output SPL match’s the input SPL.
In the Frequency domain that be including phase.

Do you mind putting here some of their names and opinions? Because I am interested in why they say that.

• Researcher A (reason A): "more work needs to be done to ascertain the limits of phase audibility (because it is a significant factor in preference when listening to recordings at home on loudspeakers)".
• Researcher B (reason B): "more work needs to be done to ascertain the limits of phase audibility (because it is my area of research and it intrigues me and because it advances the frontier of knowledge of how humans hear)".

Usually, when a research paper concludes with the standard "more work needs to be done" clause, it is for the sake of knowledge itself, ie reason B.

It is only important to us in terms of setting up our hifi if it is reason A, ie significant to home music playback preference. And if it were so, then Toole would be aware of it and would be saying as much. And then Amir would be, too.

So I think you are out on a limb here. It is cool that you are interested in it, and I surely do admire your efforts to set up your home system with measurements. But your contributions here on the topic look like you are determined to give the impression that it is a significant factor in preference, ie in home audio enjoyment, and that the science just has to catch up. If so, I am left wondering whether your extensive tests have created a bias. If your listening at home to your tweaking of phase has confirmed your beliefs, then it would a classic case of confirmation bias. Not entirely unlike the classic audiophile gear audition method via sighted listening.

So, please show us the context that would impress the neutral observer that this is a significant factor for home listening preference.

cheers

One can point to the paper that has been quoted recently in the thread:

Ok, may I use some research you've just shown....adding my blue and orange highlighting to your yellow.

View attachment 468318
I've added

@Newman I am not sure that it is incumbent upon @gnarly to supply researchers and a synopsis of their studies on phase, mainly because it seems like it has been emerging forever. But everyone seems to stop with the SPL magnitude, distortion and compression.
And my interpretation of the thread conversation, is that it was inviting the perspective of “The Luminaries” to opine.

…
So, please show us the context that would impress the neutral observer that this is a significant factor for home listening preference.

cheers

I think “significant” comes after FR, compression, and distortion.
It seems like it is insignificant to many, and only significant to a minority.
Basically, it is a question of “where and when does one stop peeling an onion.”

 

[Palmer]

The term "illegal" is a formal term used in audio and video technology. As an example of latter, broadcast/professional video defines luminance values from 16 to 255. Whereas for computers, we have 0 to 255. A value of say, 8 is valid for computer RGB signal but "illegal" for video. You obviously can create video samples that are below 16 but you can't expect the system to behave any predictable way if you attempt to play it. You could get clipping, or other artifacts.

By the same token, a digital sample going from dead zero to max amplitude is "illegal." It was not prefiltered as Nyquist theorem demands. As a result you get artifacts that would not be of interest with real content.

Takes me back memory lane, some of us used super black - back in the day...

 

[Holmz]

The term "illegal" is a formal term used in audio and video technology. As an example of latter, broadcast/professional video defines luminance values from 16 to 255. Whereas for computers, we have 0 to 255. A value of say, 8 is valid for computer RGB signal but "illegal" for video. You obviously can create video samples that are below 16 but you can't expect the system to behave any predictable way if you attempt to play it. You could get clipping, or other artifacts.

By the same token, a digital sample going from dead zero to max amplitude is "illegal." It was not prefiltered as Nyquist theorem demands. As a result you get artifacts that would not be of interest with real content.

I think we are getting mixed up with terminology and semantics, but I’ll try again.

There is nothing illegal about a file with all zeros and a single sample with a non-zero value.
By the Nyquist theorem, it tells us that ^that^ will only represent frequencies between DC and 22kHz (for a CD).
So it is bounded by the sample rate of the file.

What would be illegal, would be the Dirac impulse with finite bandwidth on the input in the time domain, and getting that to be captured as a single non-zero sample.
And also why the input is bandpass, or low pass filtered to keep the incoming information within the range that is defined by as the Nyquist limit.

The only illegality is in terms aliasing, but the file is not illegal, and the output stops at exactly F(s)/2.

Nyquist–Shannon sampling theorem - Wikipedia

en.wikipedia.org

 

[j_j]

There is nothing illegal about a file with all zeros and a single sample with a non-zero value.

Actually, that's not possible at the system output. It implies that you have energy all the way to pi/2, which does raise some interesting issues, but in general, you can't get that INTO the system, because any filter that provides proper antialiasing will have at least 2 samples in its impulse response. Always. Absolutely.

https://pnw.aessections.org/2024/12/30/2022-11-17-what-is-bandwidth/ is offered here for reference.

 

[Holmz]

Actually, that's not possible at the system output. It implies that you have energy all the way to pi/2, which does raise some interesting issues, but in general, you can't get that INTO the system, because any filter that provides proper antialiasing will have at least 2 samples in its impulse response. Always. Absolutely.

https://pnw.aessections.org/2024/12/30/2022-11-17-what-is-bandwidth/ is offered here for reference.

Let’s say we oversample that 44.1kS/Sec by a factor of 8x and run that into a DAC etc. then we can anti alias it at 50 or 100 KHz.

One could certainly run into some issue where there was a bit error that would be 1 bit once in a while.
Conceptually that is somewhat like the odd sample sticking up like a meerkat.

 

[amirm]

There is nothing illegal about a file with all zeros and a single sample with a non-zero value.
By the Nyquist theorem, it tells us that ^that^ will only represent frequencies between DC and 22kHz (for a CD).

As JJ said, "nope." You have an impulse signal there as we have been discussing and hence, has infinite bandwidth.

So it is bounded by the sample rate of the file.

Again, nope. You see this fact in every DAC measurement I perform in the filter test:

As you see, the file, filled with random noise, clearly has spectrum above 22.05 kHz. So just because you can create such a file, it doesn't mean it is bandlimited.

 

[amirm]

One could certainly run into some issue where there was a bit error that would be 1 bit once in a while.

Sure and with it, you get high frequency spray if the previous samples are much lower value.

 

[pma]

There is nothing illegal about a file with all zeros and a single sample with a non-zero value.
By the Nyquist theorem, it tells us that ^that^ will only represent frequencies between DC and 22kHz (for a CD).
So it is bounded by the sample rate of the file.

What would be illegal, would be the Dirac impulse with finite bandwidth on the input in the time domain, and getting that to be captured as a single non-zero sample.
And also why the input is bandpass, or low pass filtered to keep the incoming information within the range that is defined by as the Nyquist limit.

The only illegality is in terms aliasing, but the file is not illegal, and the output stops at exactly F(s)/2.

I agree. And it is not very difficult to make a demonstaration.
/for those who still are confused - it is not any Dirac impulse, it is in fact a 96kHz sampling, reconstruction filter response to a single impulse, thus an impulse response)/

1. 96kHz 16bit 1 impulse file:


2. Time domain response at DAC output (Topping D10s), sampled with Fs = 12.5 MHz:


3. Frequency domain response at DAC output, sampled with Fs = 1.563MHz:


4. The wav file attached below: