24/192 Downloads ... and why they make no sense?

[Miska]

Where did I call for a Dirac? We were discussing impulses shorter than 22 us. And yes, the shorter the impulse the more insignificant its sub-22kHz spectrum becomes.

 

I was talking about transients, not impulses. Your example was Dirac delta function as we all know.

 

Not an impulse, not a transient. Something periodic. It has a fundamental. The fundamental exceeds 44.1kHz.

 

Well, it is closer to real world transients in low cycle counts. But if you look back in the thread, I also presented single-step response that does have all the components up from DC. You can make your pick.

 

What do you know about the audibility of the (pre-)ringing? There have been studies. In the midrange it is very very obvious. Towards higher frequencies it gets inaudible. Where is the crossover?

 

I've been studying the topic since 2008 and optimizing my oversampling filters to be as perfect (short) in terms of ringing as well as frequency domain behavior (maximum attenuation at stop-band).

 

You should also study different kinds of ringing, IOW, different oversampling filters.

 

When combined with transients it is especially audible in terms of pre-echo or pre-ringing. In addition to blurring the transient in time domain.

 

Here are two of my different apodizing(* oversampling filters exposed to Dirac delta from RB, I prefer the sound of the latter one (both have identical frequency response):

 

*) These replace ringing of the original ADC down-sampling filter.

 

Comparable to the plots published by Stereophile, like this one:

http://www.stereophile.com/content/dcs-debussy-da-processor-measurements

 

 

[PeterSt]

Well, I don't see any distortions on wideband spectrum analyzer plot. Above 150 kHz there's only noise floor...

 

I doesn't sound "softer" at all when there's real high frequency components present. This kind of square definitely doesn't sound soft.

 

Too easy in my view;

Let's keep it simple :

First we say something like "150KHz is perfectly processed (DSD), but let's make 100KHz of that because you probably imply that the roll off ended at 150KHz and b. it should be clear that we desire the 100KHz to be there at full scale (not to perceive that high frequency, but to process transients well (jangling keys etc.).

 

Now, you are not going to tell me that your tweeter will process the 100KHz sine wave (that's what it is) with the implication of it being a square wave, while in the mean time all other nice sines won't show distortion ...

Or ?

 

And besides that you just said yourself that it wouldn't be nice to tweeters (there implying even the 150KHz at full scale).

 

Now make a choice ...

I can't, because both don't go along. Either the tweeter distorts and causes something which is at least not real, or the 100KHz is not there at all.

 

In either case there's no argument in processing the transients like we are talking about them now.

Ehm, with the data available now ... could include a few mistakes.

 

 

[Julf]

"I can hear the difference between low and high res (same recording, both downsampled if needed) in two seconds."

 

I would love to have you try the hi-res listening test, just for fun. How about it?

 

[Miska]

Now, you are not going to tell me that your tweeter will process the 100KHz sine wave (that's what it is) with the implication of it being a square wave

 

I don't know where the square comes to this picture. 100 kHz sine would be and should be 100 kHz sine. But no, it won't produce it properly. My tweeters have frequency response flat to only 45 kHz. But I'm planning to buy pair of Tannoy ST50's with frequency response "To 54kHz, usable output (-18dB) to 100kHz".

 

B&W diamond tweeters go to 70 kHz.

 

Anybody knows more precisely how high these go?

http://www.acapella.de/en/hornspeakers/ionic_tweeter.php

 

 

[PeterSt]

"I can hear the difference between low and high res (same recording, both downsampled if needed) in two seconds."

 

I'd merely love to see how those two tracks would be composed. And how they would differ from eachother.

I'm taking the words of the quote quite literal ...

 

[Fokus]

Why ? because you don't want to do it yourself ?

 

Wanting is one thing, having the time for it another.

 

But mainly to avoid any potential for misunderstanding. There is already enough fog in this place.

 

 

And better cut out the humans stuff

 

[srcsm]I honestly did not know we were in the business of recording music for not listening to it.[/srcsm]

 

 

I was talking about transients, not impulses.

 

Impulses are transients. Or do you want to redefine all words?

 

Your example was Dirac delta function as we all know.

 

A Dirac function is an impulse with zero duration, unit area, and therefore infinite amplitude. My example emphatically was not.

 

You should also study different kinds of ringing, IOW, different oversampling filters.

 

Have been doing so since 2000 or so.

 

When combined with transients it is especially audible ...

Here are two of my different ... filters exposed to Dirac delta from RB, I prefer the sound of the latter one

 

Could you provide sound samples that clearly indicate the audible differences?

 

Input file is enough. I can generate just about any oversampling filter myself.

 

 

 

 

 

 

[Miska]

A Dirac function is an impulse with zero duration, unit area, and therefore infinite amplitude.

 

In digital domain it is single sample of maximum amplitude as we know.

 

Could you provide sound samples that clearly indicate the audible differences?

 

I don't post content to which I don't own copyright. You should anyway use material you listen to anyway.

 

It's also about knowing how and what to listen. My Navy career gave me good abilities on listening small details (passive sonars). And teaching others to do so.

 

Input file is enough. I can generate just about any oversampling filter myself.

 

I'm listening my own ones and several from others. My filters are available in my software, easily switched so comparisons are easy. If you don't hear differences, don't bother buying such products. It's easy...

 

Where can I find number of your oversampling filters that all sound the same?

 

 

[PeterSt]

Only if you view it linearly, as your eye would. That's not how your ear works.

 

But this is not even what I meant. So Julf, this one is for you ...

 

Here is your own graph again :

 

 

 

Imagine a new setup of this ...

 

One graph shall show 16/44.1 and the other 24/88.2. No time shifts are allowed / applicable this time.

 

The difference in height between the green and brown steps imply the perceived 256 times better resolution.

 

Tell me, why don't doesn't the difference between green and brown shift by 1/256 per sample going to the right ? Why doesn't it consume all of the available 256 amplitude steps ?

 

Right, because it first needs an increase of 256 times the sample rate. Not 2. Only then the steps imply the max utilisation of the bit depth.

 

This is where I said "transients are unnecessary high compared to the lowest step possible" (similar) and "DSD will be much better" (but still not enough at 64x and at 24 bits).

 

Only when those 256x seemingly higher amplitude resolution is met by the sample rate (hey, that's 256 times :-) we can speak of normal inherent stepping which doesn't need to be solved by a brickwall filter. Otherwise we will be missing "transients" (or perceive distortion from them).

 

Again : Something like that ?

 

 

PS: So, the 24 bits allow for grasping the accurate amplitude alright, at going from the one sample to the next one. No problems there. But the relative step in amplitude is not accurate at all. At least not as the 24 bits seem to imply.

Only when both factors are in balance, it can be seen as an analogues line, although still "stepping" will be in there. But is is a linear one (forgetting about the frequency and implied varying steepness of sine slopes).

When this has been achieved, it is time to look for a higher bit depth again *and* a higher sample rate.

 

[Fokus]

Where can I find number of your oversampling filters that all sound the same?

 

Where did I claim they all sound the same?

 

 

[PeterSt]

And better cut out the humans stuff

 

[srcsm]I honestly did not know we were in the business of recording music for not listening to it.[/srcsm]

 

Don't take this out of context. I said that yesterday's studies are obsolete today.

Not that they are not valid today when performed today.

 

I could also just pose the question :

Do you think nothing proceeded in audio the, say, last two years ?

 

 

[Julf]

Peter,

 

I am sorry, but I just can't figure out what you are trying to say, at least not in a way that would allow me to respond in any meaningful way. Would it be possible for you to use a picture that actually has the waveforms you are talking about instead of me having to try to imagine how you have imagined the picture changed?

 

[PeterSt]

Could you provide sound samples that clearly indicate the audible differences?

 

Allow me too :

Goto XXHighEnd, use any music recording you like, and compare AI vs AP filtering. You can even use the AI-what-you-want-DAC.

 

The difference is in the transients, and the 100% following of them with AP. It should overrule whatever your DAC does, although this is no guarantee.

 

Nobody says what is better. But you asked for an audible difference.

 

Peter

 

[Fokus]

I am sorry, but I just can't figure out what ...

 

I am afraid I could. It is ... shocking, to put it mildly.

 

 

I have to go back to the very first lines I wrote yesterday.

This place is beyond repair.

 

Centuries from today this may become the subject of a few Ph.D.s in archeo-cognitive sciences.

 

Oh well, have fun you all.

 

 

 

 

[Miska]

Where did I claim they all sound the same?

 

Now, please enlighten us why do you think they sound different and how we can get to listen those?

 

 

[PeterSt]

I am sorry, but I just can't figure out what you are trying to say

 

I am the most sorry Julf, but you are not to blame.

Here's another guy happily stating that he has no time to make graphs. But actually I am not in the position to do to. I mean, wrong office ...

 

I hoped you could do it easily.

 

But ... I think it really is only about this :

 

Apply the 3KHz purple wave exactly as you did before. So, this is 16/44.1

In the same graph, apply another 3KHz, but now of 24/88.2.

Superimpose the stepping of either, similar to what you did before.

No time shifts apply here.

 

Look what you got.

Then the remainder should become clear and otherwise I can exlain further.

 

I hope.

:-)

 

 

PS: No obligations of course. I'm ust trying to shift some time towards yours.

 

 

[Julf]

"In digital domain it is single sample of maximum amplitude as we know."

 

No, it isn't. As "we know", "A Dirac function is an impulse with zero duration, unit area, and therefore infinite amplitude", just as Fokus stated.

 

[Julf]

OK, so I visualized the picture correctly. So what I don't get is this:

 

"Tell me, why don't doesn't the difference between green and brown shift by 1/256 per sample going to the right ? Why doesn't it consume all of the available 256 amplitude steps ?"

 

[PeterSt]

Oh well, have fun you all.

 

Well, at least you tried !

Appreciated ...

 

Fokus, no matter we are all nonos, zombies actually, the two main "opponents" of you actually do have products. Maybe I am not allowed to say that these products are highly appreciated by that other bunch of undoubtedly more numbos, but we all try our best, and actually are ignorant as hell. This already shows from those two lullos you are mainly addressing. They don't agree on a single thing. Stupids.

 

You come right in between that all, of course being helpful as you can.

But maybe in a not so communicative way. I know, you called names like "unprecicers" and it is not even clear to whom you talk. Read back.

 

Now what shall we do ? Is the world doomed because you bail out, or will the world be doomed because you stay in and press your own ideas without any history shown.

 

Mind you, what you imply is that "we" shout around a little.

What I SAY is that you do.

 

Can we come to some consensus, or must we really go all the wrong direction without you.

 

Sarcasm drips of it. Still I say it for a reason ...

 

Peter

 

[PeterSt]

Why doesn't it consume all of the available 256 amplitude steps ?"

 

I'd almost say : because that is too obvious.

 

Put it this way :

Make the sample rate not 24/88.2 but 24/11289.6 (so, 256 times 44.1). *Now* look at the graphs.

Now one sample sideways will show a maximum increase of the minimum volume step (which is 256 times better for 24 bits vs. 16 bits).

 

Does it start to live now ?

(without picture this is difficult explaining ... ^-))

 

 

 

[Miska]

No, it isn't. As "we know", "A Dirac function is an impulse with zero duration, unit area, and therefore infinite amplitude", just as Fokus stated.

 

So we want to split hairs and be more accurate, so usually we get from

"a generalized function on the real number line that is zero everywhere except at zero":

http://en.wikipedia.org/wiki/Dirac_delta_function

to

http://en.wikipedia.org/wiki/Dirac_comb

to

http://en.wikipedia.org/wiki/Kronecker_delta_function

 

So we call it Kronecker delta function and be happy and masturbate over:

http://en.wikipedia.org/wiki/Kronecker_delta_function#Relationship_to_the_Dirac_delta_function

 

 

[PeterSt]

So we call it Kronecker delta function and be happy and masturbate over

 

With Teresa in the nude around ?

Now wait ...

 

[Julf]

"So we call it Kronecker delta function and be happy"

 

As long as we accept the constraints:

 

"if a Dirac delta impulse occurs exactly at a sampling point and is ideally lowpass-filtered (with cutoff at the critical frequency) per the Nyquist–Shannon sampling theorem, the resulting discrete-time signal will be a Kronecker delta function."

 

So they are the same only if the delta impulse occurs exactly at a sampling point and is ideally lowpass-filtered. And yes, it makes a difference.

 

 

 

[Julf]

"Now one sample sideways will show a maximum increase of the minimum volume step"

 

Uh, why would the maximum increase be constrained to the "minimum volume step"?

And one sample of 88.2 or one sample of 11289.6?

 

"without picture this is difficult explaining ..."

 

Not to mention that it is difficult to understand as well. So unless you have time to produce a picture showing what you mean (or point to something else that helps explain it), I won't continue this thread, as it seems to be in danger of getting stuck in another circle.

 

[PeterSt]

Uh, why would the maximum increase be constrained to the "minimum volume step"?

 

Because my symantics were wrong. :-) I already had a bad feeling about it. How to say it ...

 

A. With your 16/44.1 3KHz sine you will be able to imply a resolution which comes down to 256*256 steps (covering plus and minus).

 

B. With your 24/88.2 3KHz sine you will not be able to imply a resolution which comes down to 256*256*256.

 

C. With a 24/112896 you will be able to imply a resolution which comes down to 256*256*256.

 

Ad B.

You will see that in decimal it will evolve like 1, 9, 20, 28, 37 (numbers are fictive) - depending on the position on the slope (zero crossing point is the steepest).

 

Ad C.

Here you will see 1, 2, 3, 4, 5, 6 nicely filled out, BUT

with possibilities of 1, 1, 2, 2, 3, 3, 3, 4, ...

On the peaks (least steep) this may happen with B just the same, but is is about the relation between the two (B and C).

 

It is 1-0 for you, because of course even in C you can see 1, 3, 5 etc., but now it needs a very high frequency. 3KHz for sure won't do (it will be way beyond the "audio band").

 

 

 

[Miska]

So they are the same only if the delta impulse occurs exactly at a sampling point and is ideally lowpass-filtered. And yes, it makes a difference.

 

Which was of course the case in Fokus' example because it was digitally originated... Single normalized sample.