Temporal resolution

[fpitas]

Without really taking sides, my curiousity has been piqued. Most recording studios pass the finished mixes to the mastering engineer in the form of digital files. This is almost universal these days, and has been for several years. Are all those recordings hopelessly muddled up? Where does the true analogue afficianado find unmolested material?

 

[CopperTop]

A simple explanation for non-DSP experts:
In a digital audio system, the audio signal must be bandwidth-limited (low pass filtered if necessary) before being sampled by the ADC, and for all hi fi applications (CD and better) this will be to a frequency that is beyond the range of human hearing, anyway. When reproduced by the DAC the sampled points are fed into a reconstruction filter whose output is truly continuous. Ignoring the amplitude quantisation that has also occurred, this gives out precisely what went into the ADC - there is no reduction of temporal resolution whatsoever. The actual sample rate is not relevant to the making of this statement.

Amplitude quantisation (e.g. to 16 bits) gives rise to errors which could be non-random and possibly audible with quiet signals and high amplifier volume, but they are rendered truly random by the mandatory addition of a sufficiently high level of random dither noise at the input to the ADC. So the output of the system contains truly random noise in addition to the reproduced signal. As with any random noise (including the much, much higher levels found in analogue storage systems), this can be viewed as instantaneous errors in time or amplitude, but produces no average error. Again, overall temporal resolution is not affected in any way.

 

[SY]

There's a nice experimental demonstration of the timing issue here:

Xiph.Org Video Presentations: Digital Show & Tell

Start in at about 20:45

 

[Barleywater]

FWIW, my digital gear is all 24/96 capable. Some of it will get up to 24/192 on a good day. Unfortunately almost all the music I like happens to be CD quality, the material available at high bit rates is all audiophile waffle.


Some changes are audible, others aren't. There have been some studies done. Also we can look at the known limits of human hearing and apply time-frequency duality to get an idea of what the limits look like in the frequency domain. The ear drops off sharply at 20kHz and has a simultaneous dynamic range of about 30dB, so we know its timing resolution can't be all that great.

We trust that the changes made to the waveform by the digital sampling and reconstruction process are the kind that aren't audible. One example is the "pre-ringing" of linear phase digital filters. It freaks out old-school analog people but I don't know of any evidence that it is audibly bothersome. It actually leads to a tighter group delay spec than an equivalent analog ("minimum-phase") filter.

I've purchased several high resolution format downloads, very nice recordings of wonderful performances. Then I've down sampled them from 24bit to 44.1kHz 16bit. Zero perceptual differences.

Human hearing doesn't sharply drop off at 20kHz, it starts rolling off from a few kilohertz, just as most microphones do that are used for recording. Mass, size, damping. And ears have active damping, for indeed dynamic range of basilar membrane is very restricted. Damping function and other non linearity of ears is source of much that is perceived as poor sound system behavior.

Loss of damping accompanies much of classic hearing loss. Old person say "speak up, I can't hear you", so you do, and next thing the old person says is, "Stop shouting, I'm not deaf."

 

[Barleywater]

I do not agree because you are assuming from the start that a REAL system follows the theorem perfectly. I don't think so. So your conclusion is arbitrary.

Your beliefs are arbitrary.

Real systems follow theorems so well, because people who questioned beliefs and nature of the universe found symbolic systems with exceptionally high correlation coefficients regarding the symbolic system's predictive powers of the real system.

You posses the mystical and magical perceptions of a young child, and the manners of a troll trying to control the bridge to knowledge.

 

[mohammed li]

What do you want to "proove" with that ?

That there are proper scientific test results where people are unable to detect the insertion of a ad/da chain in a LP playback system.
This is a test done in 1984!

First it shows that under the conditions of ABX testing no differences where
detected by this (very small) listening panel, deducting from this that "a consumer level ad/da chain has no audible influence on a "relatively high end" LP playback system"
is highly unscientific. Actually this was not done by those conducting this test. It appears that this was merely a test if one person (Ivor Tiefenbrunn) could identify reliably the presence of ad/da chain when playing
back LPs.

1 I would never generalize a single ABX test and I never have.
2 you should read the article better as there were more tests involved.

Second, this has nothing to to with differences between 16 and 24 bit resolution and higher sample rates.

I have tests for those to, all with negative results, do you want them?

I would appreciate if you would not attribute sentences to me I did not say.

You said: "Yes "proper" and "scientific" is the key. Unfortunately such listening tests are
quite rare." I presented some.

I do not agree because you are assuming from the start that a REAL system follows the theorem perfectly. I don't think so. So your conclusion is arbitrary.

Information theory applies to ALL channels transmitting information. It has been proven to be right and it is simply how nature works.

You are wrong again because it's not defending analog vs digital but just wishing for a better digital source that is EFFECTIVELY better than the vinyl. It's quite different concept! The CD is not better. It is worse.
You might consider that the typical stereo systems people have can make a lot a systematic errors in reproduction, starting from the speakers and their "integration" in the room. If there are such big errors one might not find a difference or even the contrary but is INconclusive because one still doesn't know how much is missing respect to best possible available.

I did not say that digital or vinyl is "better", I did not even use the word "better".
All I said is that digital comes closer to what the mastering studio outputs. This is a purely objective statement witch can very easily be verified with objective measurements.

To make a very basic example, if you just change a bit of the initial slope of a single instrumental note and you listen to it the sonic sensation will be different. It will have a different expression! If you then change a bit a more complex signal (i.e. music) you cannot assume that the result is the same or better. In principle is worse because a listener associates a more complex meaning to music than simple notes. Otherwise would not be music! You cannot measure this.

1 You should learn about masking in audio.
2 And you can certainly measure it with a proper subjective measurement.

There's a nice experimental demonstration of the timing issue here:

Xiph.Org Video Presentations: Digital Show & Tell

Start in at about 20:45

I already posted that video in post 12.
But it is good to repeat this. It shows a very clear demonstration of the timing resolution of digital audio.

 

[45]

Your beliefs are arbitrary.

You posses the mystical and magical perceptions of a young child, and the manners of a troll trying to control the bridge to knowledge.

I think it's rather the contrary. However you might be right if you explain in practical, scientific terms (not mystical masturbations) why the CD format is so perfect that there are better formats as masters that perform exactly so.
Explain why all audio industries produce maters (now available for us as files) at higher sampling rates and lower sampling rates are offered just because the market is made of stuff at different quality levels and not everyone is interested in hifi. They must be stupid.....??

Real systems follow theorems so well, because people who questioned beliefs and nature of the universe found symbolic systems with exceptionally high correlation coefficients regarding the symbolic system's predictive powers of the real system.

So well, so bad, symbolic system's predictive powers.... this is certainly arbitrary nonsense!

 

[Barleywater]

CD format is perfectly suitable as end user product.

24bit formats for recording make is easy to set levels without worry of clipping or loss to noise floor of media. This makes mixing and mastering a lot easier.

Hi sample rates are primarily response to quell the quacking from skeptics of basic information theory.

Catering to skeptics, and faster internet, and cheaper smaller faster storage medium made for lots of money.

You sure use "I think" a lot, lots of self gratifying mental masturbation that doesn't get you, or anybody out of circle jerk that you turn threads into.

 

[45]

A simple explanation for non-DSP experts:
In a digital audio system, the audio signal must be bandwidth-limited (low pass filtered if necessary) before being sampled by the ADC, and for all hi fi applications (CD and better) this will be to a frequency that is beyond the range of human hearing, anyway. When reproduced by the DAC the sampled points are fed into a reconstruction filter whose output is truly continuous. Ignoring the amplitude quantisation that has also occurred, this gives out precisely what went into the ADC - there is no reduction of temporal resolution whatsoever. The actual sample rate is not relevant to the making of this statement.

Amplitude quantisation (e.g. to 16 bits) gives rise to errors which could be non-random and possibly audible with quiet signals and high amplifier volume, but they are rendered truly random by the mandatory addition of a sufficiently high level of random dither noise at the input to the ADC. So the output of the system contains truly random noise in addition to the reproduced signal. As with any random noise (including the much, much higher levels found in analogue storage systems), this can be viewed as instantaneous errors in time or amplitude, but produces no average error. Again, overall temporal resolution is not affected in any way.

The experiment by Khunchur investigates exactly: "the temporal resolution of human hearing through its discrimination of the time constant of low-pass filtering applied to a periodic signal. ". So it is band limited and finds that time and frequency are not equivalent in general for a human being. Also an upper limit is only found for temporal resolution in a specific case, not a definitive value because this is limited by the equipment used. A rather simple signal in comparison to music, I would say. I don't make some many assumptions on the consequences of all the required manipulations to make it work simply because in the real world I don't find this to be true for the CD format. They might become really irrelevant for better formats. I still have to try this in more detail. A DVD audio is better than its CD counterpart, for now.

 

[45]

CD format is perfectly suitable as end user product.

24bit formats for recording make is easy to set levels without worry of clipping or loss to noise floor of media. This makes mixing and mastering a lot easier.

Hi sample rates are primarily response to quell the quacking from skeptics of basic information theory.

Catering to skeptics, and faster internet, and cheaper smaller faster storage medium made for lots of money.

This is not an explanation!

 

[gk7]

That there are proper scientific test results where people are unable to detect the insertion of a ad/da chain in a LP playback system.
This is a test done in 1984!

If this test was "proper scientific" I hope you are never exposed to
a medical treatment tested by such standards.

1 I would never generalize a single ABX test and I never have.
2 you should read the article better as there were more tests involved.

Those that conducted the test did not draw the same conclusions as you
did, go figure.

I have tests for those to, all with negative results, do you want them?

You are obviously are unable to understand that a "null result" is no proof.

You said: "Yes "proper" and "scientific" is the key. Unfortunately such listening tests are
quite rare." I presented some.

You presented only one (unscientific) test and drew false conclusions from it.

You better get yourself accustomed with the theory behind such tests and
especially how to interpret them. It´s really ridiculous how people use the
term "scientific" and then come up with such "tests" or some internet video.

 

[mohammed li]

I think it's rather the contrary. However you might be right if you explain in practical, scientific terms (not mystical masturbations) why the CD format is so perfect that there are better formats as masters that perform exactly so.
Explain why all audio industries produce masters (now available for us as files) at higher sampling rates and lower sampling rates are offered just because the market is made of stuff at different quality levels and not everyone is interested in hifi. They must be stupid.....??


So well, so bad, symbolic system's predictive powers.... this is certainly arbitrary nonsense!

-You must think Force equals mass times acceleration is arbitrary nonsense to.
-Higher sample rate files are now available so gullible people buy their record collection again. Its a scam to make more money.

If this test was "proper scientific" I hope you are never exposed to
a medical treatment tested by such standards.

It was an ABX test done by a very highly regarded audio professional who has made hundreds of publications published in peer reviewed audio magazines.

Those that conducted the test did not draw the same conclusions as you
did, go figure.

And what conclusion would that be?

You are obviously are unable to understand that a "null result" is no proof.

I have different test methodology results to.
Do you want them?

You presented only one (unscientific) test and drew false conclusions from it.

1 Do you think Stanley Lipshitz does not know what he's doing?
Edit: http://www.torontoaes.org/bios/lipshitz.html
2 What false conclusion did I draw?

You better get yourself accustomed with the theory behind such tests and
especially how to interpret them. It´s really ridiculous how people use the
term "scientific" and then come up with such "tests" or some internet video.

I have said it here before: ABX test can't be generalized.
I used scientific to indicate a specific method was used.

 

[jneutron]

CD format is perfectly suitable as end user product.

24bit formats for recording make is easy to set levels without worry of clipping or loss to noise floor of media. This makes mixing and mastering a lot easier.

Hi sample rates are primarily response to quell the quacking from skeptics of basic information theory.

Actually, it's a tad more complex.

How close the reconstruction gets to the true analog signal is dependent on the depth of the reconstruction math. One bit depth consistent with a dac-s/h stage is not very accurate temporally.

"cut....with the human auditory system being capable of detecting changes down to 6 microseconds."
Comment: This is only the actual upper limit that has been possible to measure until now. It is not a definitive value!

Nordmark demonstrated 1.2 uSec human lateralization capability back in 1972with a dithered signal, and about 5 uSec undithered.

"To also accurately reproduce changes in a signal’s frequency spectrum with a temporal resolution down to 6 microseconds,the sampling rate of a digital audio system must operate at a minimum of the reciprocal of 6 microseconds = 166 kHz.

1.2 uSec inverted is a tad over 500 Khz. But that's not the system bandwidth requirement, just the temporal accuracy. It's certainly not 1 bit depth, and not 100, but something in between.

The application of information theory to analog/digital/analog reconstruction is many times misunderstood..and with good reason, it ain't easy.

It's simple enough to reconstruct the analog well below 1 uSec using 44Khz sampling, but it requires up front high depth digital math as well as high depth math in the d/a phase.

jn

ps..sy your mailbox is full..

 

[SY]

Only PM (which I hate with a burning passion). Drop me an email.

 

[45]

1.2 uSec inverted is a tad over 500 Khz. But that's not the system bandwidth requirement, just the temporal accuracy.

Yes, in fact 5 uS was limited by the equipment but this not the main point for me. It demonstrates that even for a simple meaningless signal frequency and time are not conjugate variables for the human being. As music is not just a meaningless signal the less errors (and so less manipulation) in the system the better.

It's simple enough to reconstruct the analog well below 1 uSec using 44Khz sampling, but it requires up front high depth digital math as well as high depth math in the d/a phase.

You could a genius but the CD's I have don't appear to be yours and thus your ability is of no use for me. My vinyl material in my turntable is by far superior. I don't care about of what is possible because my interest is music and this is misunderstood as well. I wonder if some people use the stereo as a mean to listen to music or just to play with it. And so if the higher sampling rate makes things easier, because those who make music I like may not be masters of digital electronics, that could be a solution. If not I think I would not be interested....

 

[jcx]

RedBook CD has plenty of bits for nanosecond resolution

this can be simmed - like I have with the dealyed clik tone bursts before when Kunchur's lack of DSP prowess has come up

LTspice - free, already used by many here


below I have created 2 50 ms raised sine enveloped 1 kHz tone bursts: B1, B3

they are dealyed by td1-td2 = -119 us

added 2 lsb pp TPDF dither (B2, B4), then write out both to a 2 channel .wav file at 16/44100


the next asc reads the .wav, then I run a fft on the burst wavefrom

the cursors are set to the peak at 1 kHz

calculating from displayed phase diff of -42.8401° / 360° * 1 ms = -119.000278 us

thats sub nanosecond resolution folks - encoded, transmited through a CD resolution .wav file

all the moving parts are visible, you can play with the sims yourself

(LTspice group delay calc seems to have a rounding? error - looks like I'll be emailing mike)

just to see it wasn't dumb luck I can change the td2 value to 12 us for a diff of 11 us - which is pretty much halfway between 2 samples @ 44.1k
-3.96005° / 360° * 1 ms = -11.000139 us

again sub nanosecond temporal resolution - just coincidence?


things you can try with the sims - edit the BV for shorter, higher frequency, lower amplitude - say 5 ms of 6 kHz at -20 dB: -23.7746° / 360° / 6 kHz = 11.007 us

that's 7 ns error due to shorter burst and lower amplitude showing limiting with dither noise floor

 

[Barleywater]

Actually, it's a tad more complex.

How close the reconstruction gets to the true analog signal is dependent on the depth of the reconstruction math. One bit depth consistent with a dac-s/h stage is not very accurate temporally.


Nordmark demonstrated 1.2 uSec human lateralization capability back in 1972with a dithered signal, and about 5 uSec undithered.


1.2 uSec inverted is a tad over 500 Khz. But that's not the system bandwidth requirement, just the temporal accuracy. It's certainly not 1 bit depth, and not 100, but something in between.

The application of information theory to analog/digital/analog reconstruction is many times misunderstood..and with good reason, it ain't easy.

It's simple enough to reconstruct the analog well below 1 uSec using 44Khz sampling, but it requires up front high depth digital math as well as high depth math in the d/a phase.

jn

ps..sy your mailbox is full..

And just where is this reconstruction math?

Where is reconstruction math in class D amplifier?

At some point it becomes charge, integrate, and smooth.

Much higher sample rate and bit depth system may be used to analyze output of lower rate, lower bit depth system. It sees lower system's full bandwidth, noise floor, and analog signal content. It may be able to separate out jitter, break that into noise jitter, and periodic jitter, in case of 44.1kHz 16bit, it will find all these well below levels even remotely capable of infringing human temporal perception.

From recording engineer's application needs, the increased dynamic range of 24bit is all that's desired. Additional temporal resolution is not consideration.

Noise of resistors in production electronics is well above noise floor of 24bit capability; this is limiting real performance to less than 24bits; and can serve as built in dither.

 

[jcx]

longer delays show that u(time-td) window in front of envelope is required too - added to .asc below

an LTspice obscurity is that to get different radom numbers for the dither in each run of the sim you need to check the box in the
tools>control panel>hacks> "use the clock to reseed MC generator"

 

[MiiB]

I dont think that in absolute terms any digital format can get more than 12-14 bit resolution, the needed resistor tolerance and current noise simply doesn't allow for that. in relative terms I am not so sure, as what we listen to is differences between what is now and what was then.

Why is this digital thread in the analogue forum..??

 

[45]

Why is this digital thread in the analogue forum..??

Because it was the continuation of another thread and not really meant to talk about digital only but more about humans who are defined a little bit too superficially when it comes to define standards, quality etc.. needed to for best results. I don't find digital (i.e. CD) good enough yet to be my reference source and I am trying to understand why it doesn't work. In the end it could even a simpler reason that sets mostly the quality of the format: the methods used nowadays for music recording (independently of analog or digital support).
Personally I don't care about off topics if they deal with interesting stuff.