Audible Jitter/amirm vs Ethan Winer

[Phelonious Ponk]

As to equipment today being competently designed, and jitter not being an issue, you can only say this if you have seen their measurements and not manufacturer's claims of jitter numbers.

Or I could listen to them under my "normal" listening conditions.

I'm no anti-measurement subjectivist. I believe measurement, particularly frequency response, can predict sonic character. But measurement can also differentiate the performance of components that are sonically identical, and I believe that jitter has been used (abused?) to great effect this way. Some manufacturers have continued to reduce measurable jitter well below what matters to human perception because they can, and because they can use those measurements to differentiate and promote their products relative to their competitors. There is nothing unethical about this unless they specifically claim that their products sound better as a result (some do), but there is no advantage to the user either. Unless one is willing to invest...sometimes thousands...in no more than a feeling of confidence, the best piece of gear is the one that delivers SOTA performance most efficiently, leaving money on the table to seek the best performance in another part of the system, or another endeavor altogether. MHO. YMMV.

P

 

[amirm]

Or I could listen to them under my "normal" listening conditions.

How each one of you use the information in this thread to determine what you buy is obviously totally up to you. My points is then to make sure that point of view is not generalized to everyone without merit. In this case, I have not conceded to generic term "normal listening conditions" to be meaningful enough to everyone. I don't for example don't know what you call me doing if I am turning up the volume 30db to get to 0db actual playback level:

Note that the above is an actual capture of real content.

I like to encourage people to try to hear digital artifacts and make themselves familiar with what it sounds like. It is entirely possible that they are hearing it under "normal circumstances" and they were blaming it on other things. Most people have multiple digital chains they can create to hear differentiation. For example, try using Toslink optical and coax S/PDIF and switch between them and see if you can spot differences. You really owe it to yourself to run such a simple and low cost test to learn a bit about this space if you consider this a serious hobby for you.

I'm no anti-measurement subjectivist. I believe measurement, particularly frequency response, can predict sonic character. But measurement can also differentiate the performance of components that are sonically identical, and I believe that jitter has been used (abused?) to great effect this way.

Let me clarify that graphs like I posted go beyond jitter. They show the total performance of the system which includes linearities which can be just as important as jitter. Clarifying, there is no guarantee that a DAC resolves all the bits in a 16-bit value even if jitter is zero. It certainly can't resolve it if it is 24-bits. We have good measurements that show us this. The graphics that I post showed that in the case of that box, the DAC is indeed deficient. I put no positive value in a DAC not resolving what is given to it. There is no good distortion there or sonic character. And no way to defend a DAC which outputs things incorrectly.

Now, if two boxes resolve beyond the sample resolution then by all means, use your ears to decide which one to buy.

I think most people ignore the graphs like the one I posted because they don't know what they mean. They understand frequency response graphs which are pretty boring with digital products since they are pretty flat. So the attention is put on the wrong thing.

Some manufacturers have continued to reduce measurable jitter well below what matters to human perception because they can, and because they can use those measurements to differentiate and promote their products relative to their competitors.

Such effort is also indicative of good engineering and design know-how. Whether the panels on my car are aligned or not doesn't impact how it drives but it sure tells me of the level of quality that has gone into the car. I find no excuse in buying digital products that can't resolve the specs they claim they are able to resolve especially if we are talking about resolving the lowest fidelity format we have: CD.

There is nothing unethical about this unless they specifically claim that their products sound better as a result (some do), but there is no advantage to the user either.

Now you are speaking for everyone and I must object! I can assure you that you are not speaking on my behalf. How many other people fall in the category that you mention is impossible to tell and hence, we can't possibly say what you just said. People need to educate themselves by reading threads like this, then go and conduct a few experiments over a weekend. You all spend countless hours auditioning equipment. Why not spend some time learning to hear the artifacts here? Is becoming an educated listener a bad thing?

You are going to tell me I am sure that you may not want to learn what jitter sounds like. That would be your choice but not my recommendation to audiophiles at large. How many would suggest that you should not try to learn the difference between speaker designs? If this is not your hobby then sure, you should not care. But if it is, and you are spending so much time posting here, use some of that time instead to run simple experiments.

If at the end you don't hear the difference then great. But please, please, don't assume everyone is situated the same as you. Having tested hundreds of people in how well they can hear distortion, I can tell you that people differ. I may have to turn up the volume 30 db but someone else may not need to do that to hear the same thing. In the height of my listening training, I thought I could hear a fly move its wings a mile away . Then we had a listening test with one of our partner companies. A guy showed up who was one of their program managers. He heard a clip and kept complaining about it having high frequency distortion. We all looked at him as if he was crazy. We couldn't tell what he was talking about. Went back and checked the code and wouldn't you know it, he was right. We fixed it and he was happy. We were happy too because we hired him to be on our listening panel . I recall him being instrumental in helping improving our audio fidelity beyond that point.

Unless one is willing to invest...sometimes thousands...in no more than a feeling of confidence, the best piece of gear is the one that delivers SOTA performance most efficiently, leaving money on the table to seek the best performance in another part of the system, or another endeavor altogether. MHO. YMMV.

I am asking you to invest in your own time to learn, not money. Only when you are educated on the science and auditory effects, are you able to judge whether something is overpriced for what it is or not. Until then, it is all justification after the fact. I think I can't hear the artifacts so I am going to say there is no use in buying something better. We created this forum to do away with blind buying like that. .

 

[Ethan Winer]

I would like to see this discussion to evolve in the direction of detecting jitter while listening through loudspeakers. Amir stated he has only detected what he has identified as jitter while listening through cans.

Believe it or not, I'll side with Amir here because listening through headphones is perfectly valid. Some people listen through phones because it's the only way they can play music loud enough to fully enjoy. But others listen to phones precisely because it lets them hear more detail.

how come there has been no discussion in this debate about jitter, or a jitter equivalent, in vinyl playback, which in a best possible case has only about 12 bits of dynamic range?

No sh.., er, no kidding

(I can't believe the "s" word is censored here!)

--Ethan

 

[Ethan Winer]

Random jitter, like random noise, is not "distortion" from an engineering perspective, btw.

I called it more like distortion because it's not steady like tape hiss. It comes and goes with the music. But I guess technically that makes random jitter more like tape modulation noise, which is noise that comes and goes with the signal on tape.

--Ethan

 

[Ethan Winer]

I just did a quick web search on your name and jitter and the second Google link said this: http://www.skeptic.com/eskeptic/10-01-06/
...

I sleep in peace if I have gotten you to move from "never" to "normal circumstances."

Not to be a jerk now, after the fact, but I might have been too generous in conceding that jitter could be audible if you crank the volume on soft passages. That could be true for dither, but jitter level depends on the signal volume level. So a very soft passage will have correspondingly soft jitter, yes? In that case, even cranking the volume will not raise the jitter to an audible level because it's still the same 80+ dB below the music. Unless I'm missing something about the relation between jitter level and signal level?

--Ethan

 

[Ron Party]

Believe it or not, I'll side with Amir here because listening through headphones is perfectly valid. Some people listen through phones because it's the only way they can play music loud enough to fully enjoy. But others listen to phones precisely because it lets them hear more detail.

E, I do not discount in any way the validity of listening through headphones. To the contrary, I suspect listening through headphones will yield more reliable results. Instead, I question whether the results of the AB test Amir performed can be extrapolated to a real world situation of listening through loudspeakers. I'm strongly dubious that even a trained listener would pass an ABX test while listening through loudspeakers except in a worst case scenario. I am glad to see some recent discussion on the subject.

 

[Phelonious Ponk]

Amir, your objections are noted, but I wasn't speaking for you and I didn't say that there aren't people out there who can hear things I do not hear. What I said is that reducing a distortion further when it is already below audibility (not my personal hearing threshold, audibility) accomplishes nothing. We can argue over what is audible, we can obviously disagree, but if a distortion is truly inaudible, I don't know what you can possibly be objecting to. Is it indicative of good engineering and know-how to further reduce that which is already inaudible? It shows ability, I suppose, but not good sense. If it is audible, on the other hand, it is worth addressing. And perhaps it is. Maybe even I can hear it. Tell me what to listen for. Given a good enough recording, I've been able to identify the brand and model of an acoustic guitar, I should be able to pick up on an obvious distortion.

P

PS - i have the ability to switch instantly between optical and coax. Tell me what to listen for and I'll give this a try.

 

[amirm]

Not to be a jerk now, after the fact, but I might have been too generous in conceding that jitter could be audible if you crank the volume on soft passages. That could be true for dither, but jitter level depends on the signal volume level. So a very soft passage will have correspondingly soft jitter, yes? In that case, even cranking the volume will not raise the jitter to an audible level because it's still the same 80+ dB below the music. Unless I'm missing something about the relation between jitter level and signal level?

--Ethan

Let's review an observation you had before. Please confirm if you are still standing by these statements before I comment further:

I've heard 14 bit audio and it sounds fine to me. At least for music recorded at sensible levels. I'd never use less than 16 bits for a live classical music recording, and that's the one place where using 24 bits makes sense. Even 12 bits is pretty good for pop music when the music is already normalized. But that's all beside the point.

Okay, but that still doesn't mean it's ever audible! People often use the example of reverb tails that decay as an example of why 24-bit audio is "better" than 16 bits. I have tested this several times, and was never able to hear the "fizz" people describe as a reverb tail decays unless I cranked the volume much louder than normal during that decay.

 

[amirm]

And one more thing: please tell me if your views above in not hearing jitter in soft passages at any level would also apply to all artifacts in digital conversion combined.

 

[Phelonious Ponk]

And one more thing: please tell me if your views above in not hearing jitter in soft passages at any level would also apply to all artifacts in digital conversion combined.

Not my question but...are these other digital artifacts reduced by lowering jitter?

P

 

[amirm]

Not my question but...are these other digital artifacts reduced by lowering jitter?

P

Don is more of an expert on the internals of DACs but in my view, no. Non-linearities in the DAC are there for a number of other reasons. Jitter simply adds other forms of distortion to it.

 

[DonH50]

Sorry, I dozed off... What specific artifacts? Some forms of distortion are actually reduced by adding noise (that's what dither is, after all), while others are not (or little) changed. In all cases, if you add jitter, SNR is degraded (lowered, reduced, made worse, whatever). Digital artifacts to me are something added between ADC and DAC, in the digital domain, and the impact jitter has depends on the type of error (artifact).

I will say as a general rule Amir is right -- jitter adds another error source to what's already there, and gets RSS'd along with everything else.

 

[Mark (Basspig) Weiss]

Given a good enough recording, I've been able to identify the brand and model of an acoustic guitar, I should be able to pick up on an obvious distortion.

I have this same hangup with pianos. It seems obvious to me, because the tone of a Steinway "D" varies greatly from the tone of a Yamaha C3, or a Bosendorfer 290. I may be biased, since I play the piano.

 

[DonH50]

Pianos and guitar are a bear to record and to reproduce well. However, I suggest that, given time and training, you could learn to identify many forms of distortion you are now hearing but ignoring or neglecting. There is a catch, naturally; sometimes, ignorance really is bliss! It took me years to learn to identify subtle error sources in recordings and various sound systems, and more years to train myself that the music, not the noise, is what matters and to listen through the artifacts.

BTW, in my experience musicians are among the least-discerning listeners, not because they hear poorly, but because they listen for different things. (I am a musician so hopefully I can get away with that statement.) Critiquing a recording, a musician is more likely to say (e.g.) "Did you hear that chord? The brass didn't flat the third!" than "Did you hear the jitter noise in that chord?"

While I am rambling, I will reiterate my position that many audiophiles would faint if they actually spent time in a studio and saw the amount of processing that goes on behind the scenes in most recordings...

Anyway, my 0.000001 cents - Don

 

[amirm]

I always like to look at Wikipedia articles to see if they can save me some typing and they do indeed in this case! http://en.wikipedia.org/wiki/Digital-to-analog_converter

DAC figures of merit:

Static performance:

• Differential nonlinearity (DNL) shows how much two adjacent code analog values deviate from the ideal 1LSB step [1]
• Integral nonlinearity (INL) shows how much the DAC transfer characteristic deviates from an ideal one. That is, the ideal characteristic is usually a straight line; INL shows how much the actual voltage at a given code value differs from that line, in LSBs (1LSB steps).
• Gain
• Offset
• Noise is ultimately limited by the thermal noise generated by passive components such as resistors. For audio applications and in room temperatures, such noise is usually a little less than 1 ?V (microvolt) of white noise. This limits performance to less than 20~21 bits even in 24-bit DACs.

Frequency domain performance:

• Spurious-free dynamic range (SFDR) indicates in dB the ratio between the powers of the converted main signal and the greatest undesired spur
• Signal to noise and distortion ratio (SNDR) indicates in dB the ratio between the powers of the converted main signal and the sum of the noise and the generated harmonic spurs
• i-th harmonic distortion (HDi) indicates the power of the i-th harmonic of the converted main signal
• Total harmonic distortion (THD) is the sum of the powers of all HDi

If the maximum DNL error is less than 1 LSB, then D/A converter is guaranteed to be monotonic. However, many monotonic converters may have a maximum DNL greater than 1 LSB.

Time domain performance:

• Glitch energy
• Response uncertainty
• Time nonlinearity (TNL)

 

[Ethan Winer]

Let's review an observation you had before. Please confirm if you are still standing by these statements before I comment further:

Actually, I withdraw my comment from yesterday because of something you explained previously that I forgot. Earlier you said that jitter affects low-order bits rather than being a fixed number of dB below the music. So if that's true, then jitter is not some level below the music but at a fixed level. Maybe you or other experts here can clarify.

please tell me if your views above in not hearing jitter in soft passages at any level would also apply to all artifacts in digital conversion combined.

I hesitate to speak about "all artifacts" here because it's so broad. This thread is about jitter only.

--Ethan

 

[amirm]

Actually, I withdraw my comment from yesterday because of something you explained previously that I forgot. Earlier you said that jitter affects low-order bits rather than being a fixed number of dB below the music. So if that's true, then jitter is not some level below the music but at a fixed level. Maybe you or other experts here can clarify.

Here is a simplified way to look at it. Jitter creates unwanted harmonics. Once those harmonics rise up to the same level as what a single bit of audio sample would represent, then you would lose that bit of accuracy.

Let's say you have a 16 bit system with 96 db of signal to noise ratio. The last bit of that sample is responsible for that last 6 db of performance (-90 to -96 db). Now if jitter causes a sideband which has 6 db of magnitude, then it obliterates the signal that last bit would represents, making your effective performance now 15 bits. If jitter creates 12 db of distortion products, then you lose 2 bits. So jitter eats your resolution bit by bit from the right hand side (least significant bits first).

If during A-to-D conversion, higher fidelity system was used to create it than you are using to play it due to jitter, then the distortion is not below music. But rather, it is taking what used to be represented by music data and replacing it with distortion products.

Note that I am not saying you can't hear the music below the distortion products. You can. But that the DAC performance is compromised to the level explained above.

I hesitate to speak about "all artifacts" here because it's so broad. This thread is about jitter only.

--Ethan

But how we look at it is no different. Either one has the position that all the bits in a DAC matter or not. You can't on one hand say that 14 bits is good enough for some music but not for other and in the next breadth say that if jitter robs you of the same 2 bits, it doesn't matter. Or the fact that you heard the difference between 16 and 24 bits but somehow, can't hear the difference between jitter-less 16-bit audio and one that has been degraded to 14 or 15 bits due to jitter. Quantization noise is harmonic distortion and if you can hear that, then you should have no problem hearing jitter effects. And quantization noise when going from 16 to 24 bits is much lower than the effect of jitter on 16-bit samples.

 

[DonH50]

There are some mistakes and points not well explained in the Wikipedia article, but the parts quoted by Amir are good enough imo. I keep thinking I should contribute, in my spare time (ha!)

Regarding DNL, DNL > 0 implies a "long" code, so DNL of +2 at some transition is OK -- the output will still be monotonic (i.e. codes going 0, 1, 2, 3, 4 as desired even though 2 may last a long time...) DNL < 0 implies a short code, and if less than -1 you can end up with codes going e.g. 0, 1, 3, 4 -- there ain't no two! That is nonmonotonic. Some applications are more sensitive to nonmonotonic codes than others. And, you may see an N-bit DAC (or ADC) specified as M-bit monotonic, particularly a very high-resolution device.

INL can be (and usually is) derived from DNL, and in fact the shape of the INL curve often provides insight in to a converter's dynamic performance. A bow contributes even-order distortion, a sinusoidal shape odd-order, etc.

To head at least somewhat back on-topic, (time) jitter matters only in the presence of a varying signal. If the sampling time varies when there is no signal, or a d.c. (static) signal, jitter does not matter because it does not change the value of the sample. (Other error sources may, natch.) A constant jitter level may be more noticeable at lower signal levels as the error is a bigger fraction of the signal; otoh, the signal is also using fewer bits so there's a trade. I'm still working on the Jitter 101 thread...

 

[amirm]

..jitter matters only in the presence of a varying signal.

This is worth repeating and something that has been implicit in all of my explanations. Jitter modulates other signals. By itself, it does nothing. So unlike noise which you can measure even with nothing playing, jitter requires the signal itself to create distortion. And since there are infinite varieties of signals, and infinite profiles of jitter, it is a hard concept to boil down to canned scenarios.

 

[DonH50]

Amir, you posted whilst I was typing. I'd like to offer a couple of quickies:

Random jitter does not create harmonic distortion; it creates spurs in the response but these are somewhat randomly related to the signal. Deterministic jitter will create harmonic distortion terms.

Quantization noise is not harmonic distortion of the signal, but rather more like, well, noise! It sounds worse than thermal noise because it is not truly white noise and there is a relationship to the signal. INL can (and does) create harmonics.

And yeah, thinking of a good set of test cases to show the impact of jitter still has me cogitating....