Artifact Audibility Comparisons

[Ethan Winer]

Folks,

Debate rages in both audiophile and professional audio circles about the importance of low-level artifacts such as distortion, jitter, quantization noise, and summing errors in DAW software. We touched on that a while ago in the now-locked thread about dither. This article addresses the audibility of very soft artifacts, and includes Wave files you can download to discover for yourself at what volume level these artifacts can be heard.

http://www.ethanwiner.com/audibility.html

--Ethan

 

[NL5]

I'm not really sure how adding that noise equates to jitter, etc, but that test was very revealing. Thanks.

 

[Ethan Winer]

I'm not really sure how adding that noise equates to jitter, etc, but that test was very revealing. Thanks.

As explained in the article, jitter manifests as FM sidebands added to the signal. You can't hear the jitter itself because it's in the GHz range! But the sidebands fall within the audible range.

--Ethan

 

[Massive Master]

Cool stuff, Ethan - Hardly notice it on the MAW1-40 until around 11 seconds where it comes out and smacks you in the head.

 

[NL5]

As explained in the article, jitter manifests as FM sidebands added to the signal. You can't hear the jitter itself because it's in the GHz range! But the sidebands fall within the audible range.

--Ethan

So that noise is FM sideband noise? Wow. That is some seriously nasty stuff.

 

[SouthSIDE Glen]

Ethan, what I personally really would like to see you write and put on your website is a treatise on the technical methodology of analog waveform reconstruction from digital information.

Then follow that up with a piece that takes a technical definition of whatever kind of digital artifact you want to talk about - one at a time, please - and explain on the basic but detailed level how that definition affects the reconstruction process and it's result.

Then, for the final blow, relate how the effects on reconstruction described in part two cause/explain the audible - or inaudible, if you will - analog results you are using as your test data in your crusading articles.

Then you might actually have a really fantastic breakthrough article.

EDIT: Just to try and be a bit clearer about what I mean and why I mean it, let me jump start it this way:

If I am reading you piece right, what you are presenting and talking about are analog samples of the introduction of a harmonic distortion signal at <=0.1%THD of the original source.

First of, the question of the audibility of THD is nothing new. I first was exposed to it 25-30 years ago when this was the hottest of hot issues in consumer audio electronics magazines such as Stereo Review and High Fidelity. The topic arose there and then because of the debate over the necessity for the increasingly ridiculously low THD measurements in audio amplifiers, and as to whether it really mattered if one amp had a THD spec of 0.001% and another was rated at 0.0001% The general consensus pretty much agreed that a) when levels of THD start getting much below 0.1%, the difference really doesn't much matter, and that b) exact numbers aside, that low-level THD itself was, at best, a somewhat dubious spec to consider that important to begin with because it is a type of distortion to which the human ear is just not all that sensitive. For example, some argued that intermodulation (IM) distortion - which at the time was not a standard spec to be published, and still isn't, really - is far more important on the audibility scale than THD.

In this way, yep, experts were agreeing with you even 30 years ago that very low level THD is a relative non-issue and pretty much inaudible. This is not news, and your test simply confirms that. However your test is not testing jitter, it is testing the periodic injection of THD to the signal. You admit that yourself, when you say that you cannot e-create actual jitter in a test condition, but rather you are taking an analog signal and assuming that THD is a satisfactory proxy for the end result of jitter. Which bring us to the second point.

The second point, and the one you need to address, IMHO, is just how low-level THD is in fact the resulting artifact of analog reconstruction of a jittered signal, and that it is the sole analog artifact of jitter. You make no connection between one and the other.

In the past you have used the reasoning with dither, for example, that because it happens only in the LSBs of the digital samples that this directly correlates to something that only manifests itself on the analog side at extremely low volumes. Please describe the process of reconstruction of the analog wave from the digital information that explains/describes that correlation.

Now with jitter, you are basically saying that because a particular timing offset in a sample (that is the jitter) may only be a variation in the sample timing of a nanosecond, that that variation in digital timing directly correlates to an analog frequency in the GHz range. That is, you are directly equating the time shift in digital jitter to an analog frequency. Please explain this mechanism; please explain how shifts in the timing of a series of digital samples actually correspond to an analog signal whose frequency is the equivalent of that time shift.

(And for bonus points, please explain how the sidebands of a microwave-band signal are wide enough to fall into the audible range. Are you actually saying that the lower sideband of a GHz signal actually extends with any significant energy whatsoever some 99.9% of the way down to DC?)

You say you want your articles to extend scientific understanding of the topics you are tilting at. Good. Then all I ask is that you fill in the serious gaps in that understanding and in your constructions where you make the given leaps you do between the digital values and the analog result without explanation of how one gets from A to B (or maybe more correctly, from D to A ). If you can correctly scientifically explain how the reconstruction process turns jitter into low level THD, that would greatly help your cause.

G.

 

[The Green Hornet]

Pretty Cool Stuff Ethan.


"...And the music comes out here...."

Green Hornet.

 

[Ethan Winer]

Ethan, what I personally really would like to see you write and put on your website is a treatise on the technical methodology of analog waveform reconstruction from digital information.

Don't hold your breath.

I'm an empirical kinda guy, not a math whiz. I understand the basics in practical terms, but don't expect proof from me in the form of math.

To be clear, my test does not add THD to the music, and I explained that in the text. What I did was create a nasty sounding treble weighted noise figuring if folks can't hear that under the music, they won't hear anything else at the same low levels.

There are many types of artifacts, and it's really tough to emulate them all. The point here is it doesn't really matter what nature the artifacts are or what caused them. Once they're too soft to hear, they're too soft to somehow affect the audio in a "subliminal" way either.

--Ethan

 

[SouthSIDE Glen]

Don't hold your breath.

My face is already blue, Ethan. But it beats a red face, which is what you just may be facing again. before replaying to that, please read on.

I'm an empirical kinda guy, not a math whiz. I understand the basics in practical terms, but don't expect proof from me in the form of math.

I didn't ask for math, though I admit that at least a little may be hard to aviod in such a description, just like we need a little math to explain things like % of THD and low dB levels and such.

You are claiming a two step proposition: step 1 that specific digital conditions cause specific analog results (e.g. jitter causes low-level THD or low-level white-ish noise), and step 2 that that those specific analog results are all but inaudible. The problem is all you ever have argued - since I've been in the audience, anyway - is step 2. You make zero case for how we get from step one to step 2. This is a REAL BIG problem in the fact that the issues you are actually dismissing are at the beginning of step 1, i.e. the digital artifacts themselves.

In this case you are taking harmonic noise - I am basing that on your FFT analysis, which looks an awful lot like a white-weighted noisy harmonic series to me, and the fact that when you do add low-level noise to a signal, whatever it's character, the main result *IS* harmonic distortion - and injecting that into a guinea pig source signal. The problem is, Ethan, is that you are passing that off as an audibility test for jitter, without demonstrating whatsoever any description of the actual effects of digital jitter on analog wave reconstruction. You simply say that you don't know how to recreate jitter in the lab, so let's just pretend that this noise I'm injecting is jitter. You're not even talking about jitter at all, but pretending that you are.

And with all that snake oil showing fancy FFT graphs and talking about THD levels and GHz (more on that fallacy in a minute), and making the claim of wishing scientific understanding, people are actually going to believe you, even though the fact is that you have no scientific understanding at all of the most key part of the process; how the digital information gets transformed into the resulting analog wave. Explain that (you don't need to get too heavy into the math) process and explain how jitter actually results in low-level analog noise, and then maybe, maybe you have a case.

Without, that, though, your paper is just a house of reasoning built on a foundation of factual quicksand, and it's falling apart before our eyes.

And the whole idea that because jitter happens in the nanosecond range of sift in sample timing, that that defines a 1:1 correlation with the frequency of the analog result, demonstrates no basis in actual D/A reconstruction process either. You see a number in the digital domain and just try to assign an analog relevance to it by looking at analog scales. You had the same kind of bias in the dither conversation; it's like trying to explain quantum events via classical mechanical models, it just doesn't work that way, Ethan. Once again, demonstrate how such digital shifts actually DO result in GHz analog events, instead of just making the groundless claim, and you'll have real science, and not pseudoscience.

And finally, the idea that the GHz events have sidebands with high enough energy to extend all the way down to 10kHz and below, let alone with the type of energy distribution you show in your FFT analysis, is a major leap of naked faith that, like the rest, you present in scientific clothing, even though there is zero science behind the claim. Present that science, explain how analog artifacting in the GHz range is caused by digital jitter, how it is of sufficient energy to have a sideband that can bleed all the way down to the audible frequency range, and how you determined that the spectral profile you picked of a white-weighted noisy harmonic series is indeed actually part of that sideband profile, and then you just might have something on which to stand.

Without it there is absolutely nothing scientific about that paper whatsoever, and the samples you provide have nothing to do with the phenomenon of jitter whatsoever.

But you can't do that, can you? Because you really don't know how analog is reconstructed from digital. If you did truly understand that, you might realize that many of the assumptions you now make in that regard are not necessarily accurate representations of the truth, and that the very foundations on which you are basing your arguments are faulty.

G.

 

[chazba]

So I guess this means that I better do my tracking at about -12 dbfs , right???, and my final mixes to avg about -6dbfs???. I'm not trying to be a smartass its just that you're talking about stuff that is for the most part beyond my control.

By the way...I loved your video.

chazba

 

[Ethan Winer]

So I guess this means that I better do my tracking at about -12 dbfs , right???, and my final mixes to avg about -6dbfs???

No, you can run any track or mix right up to 0 if you want. However, some CD players start to clip a little below 0 so I never normalize mixes higher than -1 dB.

--Ethan

 

[Ethan Winer]

But it beats a red face, which is what you just may be facing again.

And with all that snake oil showing fancy FFT graphs

making the claim of wishing scientific understanding

you have no scientific understanding at all of the most key part of the process

your paper is just a house of reasoning built on a foundation of factual quicksand, and it's falling apart before our eyes.

instead of just making the groundless claim

you present in scientific clothing, even though there is zero science behind the claim.

there is absolutely nothing scientific about that paper whatsoever

you really don't know how analog is reconstructed from digital.

Since all you have are insults I am now officially putting you on my Ignore list. It's obvious you never even listened to my files or to the noise. I have shown to my own satisfaction that extremely soft artifacts are inaudible. If you believe there are situations where very soft artifacts are audible, the burden of proof is on you to explain why, and to post audio examples proving your point. Otherwise this will go like the other Dither thread where by the end you were begging the OP to lock the thread because you had nothing of substance to refute my points, only insults. I'm here to discuss the science of audio, not be insulted. I leave the schoolyard insults for 5th graders. To refresh your memory:

Oh my GOD, would the OP please lock this thread? ... Please OP: Lock this thread. It has turned completly non-productive.

--Ethan

 

[The Green Hornet]

Doesn't LORA CROFT collect artifacts too?


Ahh, I couldn't resist.

Green Hornet

 

[SouthSIDE Glen]

Since all you have are insults I am now officially putting you on my Ignore list.

And you thought your face wouldn't get red. There's not an insult in the entire bunch of quotes given, Ethan, they are only observations of fact. If you find it insulting that someone is calling you on the facts, you had better toughen up, because I am going easy on you compared to what some real professionals in info theory will do. You are presenting bad science and pseudoscience and representing it as scientific. That's not an insult, that a observational fact.

It's obvious you never even listened to my files or to the noise. I have shown to my own satisfaction that extremely soft artifacts are inaudible.

I didn't listen to all the files, but I did listen to thw two MAW ones. I also read and re-read your article several times before I even hit reply, because I wanted to make sure I didn't misuderstand what you were trying to say.

You have taken great pains to try and show expirimentally that distortion and noise below certain percentages of the signal level are inaudible. Fine. What you have not done is to complete the chain of evidence back to your accusee and show that the kin dof noise and distortion you are referring to are actually what reslts, and all that results, from the digital symptioms your are talking about. This paper claims to be talking about jitter. Yet you don't do a single thing to explain what the analog artifacts of digital jitter actually are, or how, by your reasoningthey actually get there.

What you don't understand is that this is KEY INFORMATION, and that without it, your paper says absolutely nothing about jitter. That's not a personal insult, Ethan, that's a scientific challenge.

And then when you throw in erroneous information such as the size of the digital jitter shift in time is converted to an analog wave with a frequency equivalent to the size of the offset - which is exactly what you're saying when you claim that a nanosecond of jitter equals a GHz of analog frequency - you are throwing in false information. That's not an insult, that's a scientific answer.

Then when you compound that falsehood with a claim that it's the sidebands of that GHz signal that are audible, and that they manifest themselves as low-level, white-weighted noise, my reply to that was that to ask you to provide the science - even on a non-technical, non mathematicla explanatory level - that actually backed up that extreme claim. You found such a request to be "insulting".

From the begining of this thread, Ethan, I have been trying to explain to you what you require to truely present your case and truely do so in the name of good science. I have given you both the leads and the opportunity to follow them, and said that if you were able to follow through, you'd have a case.

The problem is, you can't do so. I know you can't do so because I can see the bad science in your paper. That's not an insult, that's a truth.

The core of your position is that <0.1% harmonic distortion levels are inaudible. I wouldn't put a 100%, every time, sureity on that, but yeah, I'd mostly agree with that as being true. I think just about everybody else except a few crazed golden ear audiophiles would agree to that too. That's not the issue. That's old news that goes back at least 30 years.

The problem comes in when you extrapolate that basic premise to the dismissal of various types of digital artifacting. You make three assumptive mistakes here: first, that you lump all those types of artifacting together as one common group, second that the only kind of distortions that they introduce on the analog reconstruction are in the form of <0.1% distortions, and third, that without making that connection, that you are advancing the interest of science and scientific understanding - as you yourself say is your intent.

Those assumptions are wrong. You are making leaps of logic based upon a lack of understanding of how the D/A process actually works. Which is why I asked you to try to explain that process as you understood it. You won't, and I have scientific reason to believe i's because you can't, because you are actually in over your head. That's not an insult, Ethan. That is an observation based upon my understanding of the science. If you think I'm wrong, then let's see your science. Lets see you complete the chain of reasoning from digital event to analog artifact.

If you believe there are situations where very soft artifacts are audible, the burden of proof is on you to explain why, and to post audio examples proving your point.

Asked and answered n the last paragraph. That is not the point of contention here.

No, Ethan, you are the one who has put forward the proposition with your paper, the burden is on you to prove it, and that proof requires completing the whole chain of reasoning, not just half of it.

Otherwise this will go like the other Dither thread where by the end you were begging the OP to lock the thread because you had nothing of substance to refute my points, only insults.

I asked for the thread to lock, Ethan, because you were making an ass of yourself, and you just wouldn't stop, even after talking about things like "analog jitter" and other such unicorns, just like you are here. I pleaded with you to stop for your own good.

Believe it or not Ethan, I don't dislike you, at least I didn't at the time. But the fact is, anytime anyone refutes any of your claims, or asks you to explain them further, or provides counter-science, you take it as an insult. That, combined with the obvious unicorns like "analog jitter" and "GHz-band analog artifacts", and having the audacity to label it as "science", it becomes hard for those of us who see the emperors clothes for what they are to just let you be.

G.

 

[Ethan Winer]

Glen,

I don't dislike you either. I really don't. But you still haven't shown an example of very soft (< -90 dB) artifacts being audible. Which is very much the point. If you have such an example, I'd love to see it and be proven wrong.

--Ethan

 

[SouthSIDE Glen]

I don't dislike you either. I really don't. But you still haven't shown an example of very soft (< -90 dB) artifacts being audible. Which is very much the point. If you have such an example, I'd love to see it and be proven wrong.

So much for putting me on ignore .

Though you might as well. I have already answered that many, many times in this thread and the other, and you're you're acting like you haven't read a damn thing.

I'm not going around in circles any more with you Ethan. Quit trying to deflect the issues. You haven't even addressed - let alone answered - a single one of the issues I have brought up with your papers, other than to try to deflect them away as insulting. Let's see the chain of evidence, the full explanation that ties everything together and gives relevance to your tests and weight to your assertations.

Because until you do, every time you come on and make such poorly constructed assertations in the name of "myth busting", people are going to feel obliged to myth bust the myth buster.

G.

 

[Treeline]

.I asked for the thread to lock, Ethan, because you were making an ass of yourself, and you just wouldn't stop, even after talking about things like "analog jitter" and other such unicorns, just like you are here. I pleaded with you to stop for your own good..

Change the name and substitute your own, bro.

This is the kind of discussion you have offline. Even if you are right and Ethan is wrong about his hypothesis, you've chosen not to debate it with him, but merely to condescend to him in public. He's not as an unintentionally entertaining mediocre student of yours to be used to stroke your needy ego.

Do you have any idea how obvious your compensating behavior is? How about sending him a PM if you want to challenge him and then following up here with a simple difference of opinion if you aren't satisfied by his response? Don't hold yourself out to universal ridicule like this.

 

[andyhix]

I don't understand 95% of this technical stuff, but I can recognize tact and appreciate real-world audio examples. Based on those two things, I know who wins this debate in my mind.

 

[NL5]

Cool stuff, Ethan - Hardly notice it on the MAW1-40 until around 11 seconds where it comes out and smacks you in the head.

It almost sounds like another instrument. Like an industrial synth patch of some sort. Actually kind of compliments the music in a weird sort of way.

 

[NL5]

Hey - I have another question. This is only getting mixed in one time - it's bad enough just the one time. How would it apply to having it on EVERY d to a and a to d pass. On an average mix, I must have over 100 passes thru a converter for the various tracks. Obviously it's not 100 times worse, or there would be nothing but that sound. So how does that equate in a real world setting?