The tired of the obfuscation on Digital Audio article.

[xiphmont]

While I think digital has room for refinement, and I think 24 or 26 bit is probably about optimum for home playback

Based on what do you think this? The best DACs in the world today top out at 21 bits; the extra three bits is pure noise. Even very good professional equipment often stalls out around the 18-19 bit mark.

26 bits is rougly equal to the thermal noise introduced by a 10 ohm resistor in a line-level feed.

and I have no problem with higher sampling rates just because we can

I don't either actually. It's justifying it with ignorance, mysticism, and 'bigger is always better' that I object to. If your argument is 'f*** you, that's why', I can at least respect that.

Why not go a bit over the top just so we can move on?

Every year, last year's 'over the top' is no longer 'over the top enough'. There are seriously people on forums now talking about how 384kHz won't be enough-- they want 768kHz.

Interestingly, no one on this forum has a real power amplifier for real world speakers that can even get to 20 bit resolution, and no speakers even get close. Actually, analog is the bottleneck now in some cases.

Yup.

 

[xiphmont]

[edit: response post deleted by mutual agreement with microstrip. I'd misunderstood his joke post as serious, and we want to bury the hatchet on it]

 

[xiphmont]

I'll try to analogize, hopefully not go into the weeds...

a question; how does MP3 compare to Redbook 16/44 from a math perspective?

PCM (what redbook is) is a rather wasteful representation from a mathematical perspective. It is very much a 'keep the baby, the bathwater, all the towels and most of the rest of the house too' sort of thing. What it lacks in efficiency, it makes up for in convenience. It is, by far, the simplest representation to implement and manipulate.

MP3 and other lossy compressions start with PCM and try to pare things down a bit to keeping just the baby without the entire subdivision the baby lives in. But it's still fundamentally starting and finishing with PCM.

does the math or theory side of things say MP3 is also without steps? that it's smooth and continuous?

Yes.

that it has 'all' the information from the source.

Definitely not. However, there's a difference between the information you care about and the information you don't.

as the time gaps between samples gets longer and longer i suppose that the approximations of what should be represented between the samples gets worse and worse.

It doesn't remove any samples. It is a means of approximating every sample in a more compact form with less frequency-domain accuracy (removing what you can't hear anyway). You can't really think about what mp3 does in the time domain at all, it's a predominantly frequency-domain process. Mostly--- it reduces the precision with which _noise_ is rendered.

 

[amirm]

Please don't let your emotions get the better of you. Stay focused purely on the technical points please.

 

[xiphmont]

Please don't let your emotions get the better of you. Stay focused purely on the technical points please.

No double standards please, Amir.
[edit: deleted the rest, since I have no beef with microstrip]

 

[xiphmont]

this dither noise effectively 'fills in' the steps in the transfer curve so that a straight line can be drawn through the noisy staircase

I think a slightly clearer analogy is that it 'stirs' the input waveform along the steps in the transfer curve to keep the steps from 'sticking'. This is a more accurate technical analogy as well, because dither is in a real sense adding heat energy to the system, much like a form of annealing.

That said, what you quoted is not wrong, it's just a limited analogy like mine.

 

[Orb]

Based on what do you think this? The best DACs in the world today top out at 21 bits; the extra three bits is pure noise. Even very good professional equipment often stalls out around the 18-19 bit mark.

26 bits is rougly equal to the thermal noise introduced by a 10 ohm resistor in a line-level feed.


I don't either actually. It's justifying it with ignorance, mysticism, and 'bigger is always better' that I object to. If your argument is 'f*** you, that's why', I can at least respect that.


Every year, last year's 'over the top' is no longer 'over the top enough'. There are seriously people on forums now talking about how 384kHz won't be enough-- they want 768kHz.


Yup.

Xiph, I followed the 24-bit is it random noise in lsb at HA with great interest and was a great discussion as it developed.
However if the extra 3 bits is pure noise how does a triangular probability density function (TPDF) show this not background noise?
A different measurement (more relating to high rez music) also done by Keith Howard shows a difference in dynamic pattern-behaviour between those with actual high resolution bits and that of noise (only a very few showing actual music behaviour down to -115dBFS).
So if it was random or background noise why again is the dynamic behaviour-pattern that of music?
Now this is not for all recordings they have reviewed-measured, it is not even common for the high rez to reach limits of 16-bit, but it seems it is possible to record and playback down to -110dBFS to -115dBFS.

Edit:
NVM misread you were talking about last 3 bits up to 24-bits but thought you meant 18-21bits so apologies, although interestingly some recordings are reaching down to -115dBFS.
I need to dig out the Keith Howard TPDF analysis of high resolution (was some time ago so forgot when it was done bah) as it was something that also interested him, and I think he shows it is possible for sure to -120dBFS but again within what you say.
Cheers
Orb

 

[Orb]

That is not correct dither - the correct dither is TPDF which does not give rise to noise modulation and is two LSB in amplitude (not one as claimed in this article).

Yeah not really happy with the article myself in other ways as well.

MEP, the following URL does a much better job tbh and with actual examples and why, and also loook to what JA shows in his measurements for both a delta-sigma DAC and NOS.
Other technical papers will discuss the various transfer functions (relates to the quantization step) and also even type of DAC reconstruction, such as ZOH, all applicable to said "stair-case" including aliasing artifacts.

Anyway here is a better link IMO detailing dither than the SOS: http://www.users.qwest.net/~volt42/cadenzarecording/DitherExplained.pdf

Cheers
Orb

 

[xiphmont]

Hey Orb,
Ah I see your edit now. Yeah, I wasn't sure what you were talking about, now I see
Also, I have to be honest-- although I would expect those lowest bits to be a gaussian-distribution noise, I suppose it's possible they're not always pure, or that they're polluted by some kind of pattern on some parts. I've not tested this one extensively, opus111 would probably know better.

 

[Orb]

Yes glad that I re-read your post in right context
Cheers
Orb

 

[microstrip]

No double standards please, Amir. There was no personal remark or attack directed at the poster; I told him his remark was flippant, dismissive, willfully ignorant, and made me angry. All true. Then I said why. I violated none of the Forums terms or rules.

You could also have simply said "Hey dude, a bit over the top" and I'd probably have edited it myself.

Xiphmont,

Any one being familiar with WBF and knowing what means in the header of a post would understand that the whole post was a joke. It was so wrong that I assumed no one would take it seriously. Happily I am not the first one to discover that some people are not prepared to an humorist vision on audio matters.

It should have made you laugh, sorry it made you cry with anger.

 

[xiphmont]

Xiphmont,
Any one being familiar with WBF and knowing what means in the header of a post would understand that the whole post was a joke. It was so wrong that I assumed no one would take it seriously. Happily I am not the first one to discover that some people are not prepared to an humorist vision on audio matters.
It should have made you laugh, sorry it made you cry with anger.

I believe you that you meant it as a joke; it was very good satire. It was not nearly as outlandish as things I've gotten from supposedly respectable elders in the high fidelity trade (sometimes with a decent amount of abuse as a bonus). They're just kind of depressing to read. Anger is a healthier emotion than depression

So it is with some weariness that I have indeed lost some of my sense of humor on the subject. I do not automatically assume the outlandish is a joke anymore because it usually is not. :-(

 

[Phelonious Ponk]

I believe you that you meant it as a joke; it was very good satire. It was not nearly as outlandish as things I've gotten from supposedly respectable elders in the high fidelity trade (sometimes with a decent amount of abuse as a bonus). They're just kind of depressing to read. Anger is a healthier emotion than depression

So it is with some weariness that I have indeed lost some of my sense of humor on the subject. I do not automatically assume the outlandish is a joke anymore because it usually is not. :-(

Ain't that the truth. Think I'll go play a guitar.

Tim

 

[Soundproof]

Kudos to xiphmont for the considerable effort he is putting into dispelling a mountain's worth of digital myths. They have forced audiophiles to Rube Goldberg themselves into advanced audiophilia nervosa, just to bear listening to the format.
It was sad to watch Computeraudiophile go from an interesting source to an asylum for shell shocked tinkerers in a couple of years. I suspect our kids don't have a similar affliction of the vapours when the subject arises.

 

[Orb]

Thanks Tom,
yeah it is one of the best out there with good accurate info.
Most forget that the improvement to the "stair-case" can be "seen" with either bit resolution or sampling rate increase, but the reason is very different as one relates to quantization (bit resolution) and to aliasing (higher sampling rates pushes the aliasing further out and reduces jagged effect of alias artifacts).
I feel another way to show quantization and dither is more in the context of the actual signal, say a very loud 1khz sinewave to a very quiet 1khz sinewave; the quantization noise is much greater for the low signal 1khz sinewave; but this is made complicated by compounding/logarithmic quantization solutions that are outside of CD.

And of course there is the output of zero-order-hold (ZOH) DACS (again technicallly rectangular or described as "stair-step" or "stair-case"), but worth noting it is technically possible to also have non-return-zero (NRZ) and return-zero, return-one, and slightly different non-rectangular is the Sine DAC, think these are probably main ones and out of these I would say the Zero Order Hold DAC is the default implementation with music products.

Cheers
Orb

 

[xiphmont]

I feel another way to show quantization and dither is more in the context of the actual signal, say a very loud 1khz sinewave to a very quiet 1khz sinewave; the quantization noise is much greater for the low signal 1khz sinewave

Absolute quantization noise level is approximately constant, even without dither.

 

[Orb]

The point being in context of signal level Xiph; specifically low level signal.
That said, the stair-case is noticably different in terms of transfer function and undithered output relating 16-bit to 24-bit so depends the context of approximately constant as well (accept it is approx constant when considered in terms of q=1LSB and error), but my point-context is many misunderstand dither and also quantization/quantization erorr-noise, and the "stair-case" transfer function and the difference in look between the various stages of ADC-digital signal-DAC-filter-analogue signal (which is a smooth sinewave but depends upon usage of filter and separately dither).

Only discussing from a technical standpoint not that music output from a DAC is "stair-case" as I agree it is not (although again technically there are 3 different aspects to stair-case, quantization, ZOH, aliasing), but I am concerned how some are using such points to say "staircase does not exist" when in terms of EE and the various stages it does in specific context, even if not at the final analogue output; also I agree others use it to suggest staircase relates to music quality and again is wrong (although this is complicated by NOS/filter and possibly the actual implementation of dither in the studio).

Cheers
Orb

Edit:
I do feel it would be nice if your site included a presentation and comparison for say a loud and quiet signal (few LSB in amplitude) for the 1khz tone and for both dithered/undithered and went a bit more into detail about quantisation/transfer function and the bit amplitude relating to different level signals, and also 16-bit and 24-bit example and why 24-bit sinewave is better in this context but with dither considered not critical (same applies to low level signal and dither).

 

[Orb]

Hah quite amusing while looking for an article by Jim Lesurf regarding USB and jitter I accidentally found exactly what I was suggesting would make a good presentation regarding low signal and how it can look bad in the early stages of digital but matching the analogue input at the very end.
http://www.audiomisc.co.uk/inadither/Page1.html

Anyway strong recommendation to read and follow for those interested (appreciate you know this Xiph), please note though this is not discussing in detail the ADC quantisation/transfer function which would consider the "stair-step" context more, this article is about how low level signals with dither and little amplitude information does resolve perfectly at the analogue output and relationship to noise.

Cheers
Orb

 

[Phelonious Ponk]

Kudos to xiphmont for the considerable effort he is putting into dispelling a mountain's worth of digital myths. They have forced audiophiles to Rube Goldberg themselves into advanced audiophilia nervosa, just to bear listening to the format.
It was sad to watch Computeraudiophile go from an interesting source to an asylum for shell shocked tinkerers in a couple of years. I suspect our kids don't have a similar affliction of the vapours when the subject arises.

That was sad, wasn't it? The saddest part was that it was deliberate, and even a bit cynical. The monetization of CA was coming from boutique stuff of high price and dubious value, the bulk of the loyal followership was rapidly moving toward participants anxious to eat that stuff up like flies at the edge of a sugar-water pool. If you were a daily participant at the time, you could literally watch it systematically morph from, as you said, a useful resource to yet another...well, there are so many... I haven't even looked in for quite some time, but the last time I did, it struck me as little more than a more polite version of Computer Audio Asylum.

Tim

 

[xiphmont]

The point being in context of signal level Xiph; specifically low level signal.

I knew what you meant, it was just ambiguously worded. There are who think that the absolute level increases a signal level decreases, when it's _proportionately_ greater for a low level signal, just like tape hiss is proportionately greater for a low level signal on tape. (I'm thinking aloud to third parties here, not you Orb

That said, the stair-case is noticably different in terms of transfer function and undithered output relating 16-bit to 24-bit so depends the context of approximately constant as well

Sure. The specific harmonic products are the same, but the amplitudes and phases of the products relative to one another change.

Only discussing from a technical standpoint not that music output from a DAC is "stair-case" as I agree it is not (although again technically there are 3 different aspects to stair-case, quantization, ZOH, aliasing), but I am concerned how some are using such points to say "staircase does not exist" when in terms of EE and the various stages it does in specific context

It exists as an implementation detail in DACs, however even in the vast majority of those DACs it is not a ZOH of the input PCM. I understand what you're saying, but I'd suggest 'a staircase exists at some point in most implementations' is the accurate statement (and an important one), and 'the staircase exists at some point' is inaccurate. It would be like saying that oversampling is an inherent, essential property of PCM when it is not-- it is merely an important engineering concept in practice.
A DSP engineer who thinks of digital waveforms as inherently staircase-like will find himself in mathematical trouble (the kind with real-world implications) rather quickly. But you're absolutely right that any DAC engineer would realize the ZOH is a critical tool.

So, again, no argument with your essential point, which is entirely correct. I'm only debating the wording of the specific nuance.

Edit:
I do feel it would be nice if your site included a presentation and comparison for say a loud and quiet signal (few LSB in amplitude) for the 1khz tone and for both dithered/undithered and went a bit more into detail about quantisation/transfer function and the bit amplitude relating to different level signals, and also 16-bit and 24-bit example and why 24-bit sinewave is better in this context but with dither considered not critical (same applies to low level signal and dither).

I played with the idea of making a little javascript dither playground tool for the website, but it would eb kind of boring. The vid did show a low vs high amplitude sinewave without dither. With dither it's unremarkable... the noise floor just stays static.