Jitter Correlation to Audibility

[castleofargh]

I'm getting more and more interested about how it works, but less and less interested in jitter as an audio problem. ^_^
and I was already thinking it wasn't important when I started reading about it.
 
thanks to all the information dealers of this thread. that was mad dope.

 

[nick_charles]

   
At 20 kHz, 20 ps of sinusoid jitter, even assuming that it is 20 ps RMS (worst case, peak to peak would be 2.828 times better), would create sidebands at -115 dBr level (so it is -112 dB total RMS level). That is quite a bit less than 1 LSB at 16-bit resolution. Did you possibly mean 20-bit ?

 
I recognize the LSB thingy it is a reference (I think) to the mathematical model derived by Julian Dunn and colleagues two decades or so ago, his model is based on a 20Khz signal at (wait for it) 120db above noise threshold (i.e very loud) and in this case his model does work out to the ENOB being less than 16 (he considers that this would theoretically be audible but never empirically tests this) , my math is not good enough to verify this but here http://www.nanophon.com/audio/jitter92.pdf
 
 
is a paper in which this appears...

 

[nick_charles]

   
 
Have you not tried the test for yourself? Why not?
 
Step 1: Hook up DAC with an outboard USB/Spdif converter using a spdif cable.
Step 2: Hook up a different outboard USB/Spdif converter using the same usb and spdif cable, and observe the differences.
 
 
This is indeed a controlled test.
 
 
There has got to be literally thousands of people who will testify in favor of audible jitter. For a good laugh, go and post this same proof request on Computer Audiophile, Audio Circle, or Audiogon, and count the number of skeptics.  There is absolutely no need to get into highly technical arguments. And the only person here who has demonstrated any kind of technical knack on this particular subject is Digitalchkn.

 
I don't quite know where to start with this, but I'll have a go. Anyone can swap out components and perceive a difference, this difference may be real (relating to quantifiable changes)  or imagined. To determine the difference between an imagined and a real difference  requires more rigor than "I know I changed something and now I hear a difference - case proven" - which from an scientific rigor point of view is a load of fetid dingo's kidneys. The bare minimum for your approach to be taken seriously is a set of unbiased controlled tests where neither the listener nor the experimenter knows if a change has been made or not and where the listener can show a reliable ability to detect a difference/no difference condition or make a correct A or B choice over a statistically powerful set of tests.
 
Secondly, thousand of people state that they have seen UFOs, made contact with dead loved ones and bent spoons with the power of their minds, maybe some of them really have - but I'm not prepared to take their word for it without better evidence. To date ALL and I do mean ALL of the controlled tests of jitter audibility have placed actual verifiably audibly detectable jitter thresholds rather north of the claims made by "jitter is a problem" proponents. In fact there are jittered samples out there you can download and try yourself if you feel so inclined. As for the sites you cite these none of them have done any good controlled listening tests on jitter audibility so why mention them, just because some/many of the subscribers believe jitter is a problem does not make it either true or false i.e it adds nothing to the case in either direction.

 

[stv014]

  I recognize the LSB thingy it is a reference (I think) to the mathematical model derived by Julian Dunn and colleagues two decades or so ago, his model is based on a 20Khz signal at (wait for it) 120db above noise threshold (i.e very loud) and in this case his model does work out to the ENOB being less than 16 (he considers that this would theoretically be audible but never empirically tests this) , my math is not good enough to verify this but here http://www.nanophon.com/audio/jitter92.pdf

 
Well, the post I quoted referred to LSB, so I assumed -96 dB for 16-bit, -120 for 20, etc. The noise floor in a specific narrow band (or with weighting) can of course be lower than that, but then it depends on factors like the dithering/noise shaping used and the sample rate. In any case, -121 dBr overall RMS level for jitter products (assuming 20 kHz signal and 20 ps peak to peak jitter) is like less than 0.0001% non-harmonic distortion, about which most people would not worry much.

 

[stv014]

there is a limit caused by the noise introduced by quantization - but you can encode, recover tens of ns differences at moderate amplitude with 16/44: http://www.diyaudio.com/forums/analogue-source/245555-temporal-resolution-6.html#post3697702

 
Any kind of noise - analog noise or dithered quantization - makes it harder to reliably recover phase information from a short sample, because noise is added to the measured phase value as well. But this is only related to the noise density, rather than discrete timing "steps" introduced by the sampling. With correct dithering, and an arbitrarily long sample, any small phase difference can be recovered from 16/44 format digital audio.

 

[stv014]

  What do you guys think about these videos from ESS technology?

 
I have not seen the first one, but the second (delta-sigma DACs) seems to claim that very low level (under -100 dBr peaks) noise can be made audible by the presence of much louder music, and backs it up with an analogy that a voice can be heard and understood even in the presence of overall louder broadband noise (like a waterfall). I do not think this is correct, because in the latter case, the reason why the voice is not masked by the noise is that the noise actually likely has a lower level in the narrow band (I think it is about 1/3 octave for masking) around the frequency of the voice, while it can still have a higher RMS level over the entire audio band. Noise density and overall noise level are not the same. On the other hand, for low level noise that is inaudible by itself at normal listening volume to become audible while much louder music is playing, it would require some kind of "inverse masking", which is not proven at all by the video.

 
Also, the competing DAC that is shown to have the problem is not named, nor is there any information provided on how common the issue is (assuming that it is audible in the first place); I suspect they intentionally picked a worst case example. So, it is overall mostly marketing for ESS products (audiophiles are scared off buying the competition, because they might have unmeasured but allegedly audible and disturbing artifacts), rather than useful concrete information.

 

[Digitalchkn]

   
I have not seen the first one, but the second (delta-sigma DACs) seems to claim that very low level (under -100 dBr peaks) noise can be made audible by the presence of much louder music, and backs it up with an analogy that a voice can be heard and understood even in the presence of overall louder broadband noise (like a waterfall).
 

 
I haven't watched the entirety of that video so I presume this might be referring to noise level modulation associated with dithering  noise of delta-sigmas?

 

[bigshot]

I have not seen the first one, but the second (delta-sigma DACs) seems to claim that very low level (under -100 dBr peaks) noise can be made audible by the presence of much louder music, and backs it up with an analogy that a voice can be heard and understood even in the presence of overall louder broadband noise (like a waterfall).

Hogwash. First of all the difference in volume between a voice at close range and a waterfall is a lot less than 100dB. Secondly, a waterfall is a continuous noise. Music is dynamic with lots of different frequencies and rhythms. It's a lot easier to pick out a buried sound under a steady bit of pink noise than from under music that is constantly changing.

Ethan Winer did a video that illustrated what dBs actually sound like. -100dB is ridiculously quiet. You could play noise below music at -50dB and not hear it.

 

[jcx]

can you even ABX the -60 dB added Sousa march? http://www.libinst.com/Audio%20DiffMaker.htm > http://libinst.com/diffmaker_example_files.htm
 
there are scenarios that show you can hear something separated in frequency >80 dB down from other signals played at the same time - at moderate to high SPL masking is strong in the upwards direction - a loud, lower frequency makes it hard to hear the higher ones played at the same time
but masking is less effective when the large signal is much higher in frequency and you listen for a low frequency

 

[ab initio]

  can you even ABX the -60 dB added Sousa march? http://www.libinst.com/Audio%20DiffMaker.htm > http://libinst.com/diffmaker_example_files.htm
 
there are scenarios that show you can hear something separated in frequency >80 dB down from other signals played at the same time - at moderate to high SPL masking is strong in the upwards direction - a loud, lower frequency makes it hard to hear the higher ones played at the same time
but masking is less effective when the large signal is much higher in frequency and you listen for a low frequency

Any test tracks or examples for this? I'd be interested to check it out.
 
Cheers

 

[jcx]

just from masking curve graphs - probably can't do the tests today with college students - the university human experimentation committee probably wouldn't approve of exposing them to >80 dB SPL in a experiment
 
these curves show the threshold for detecting a low frequency, < 1 kHz low level tone is basically the same as the threshold in quiet even with few kHz center frequency critical band wide noise masker >80 dB SPL

 

[ab initio]

  just from masking curve graphs - probably can't do the tests today with college students - the university human experimentation committee probably wouldn't approve of exposing them to >80 dB SPL in a experiment
 
these curves show the threshold for detecting a low frequency, < 1 kHz low level tone is basically the same as the threshold in quiet even with few kHz center frequency critical band wide noise masker >80 dB SPL

So what if I made a sound clip with a 500 Hz square wave at -86 dB super imposed onto 6--10 kHz noise with RMS at -6dB. Is that the situation you are describing?
 
Cheers

 

[bigshot]

That sounds like it would be a lot of fun to listen to.

 

[ab initio]

That sounds like it would be a lot of fun to listen to.

You must  have a bizarre taste in music!

 
Cheers

 

[jcx]

just don't sue me if you have to turn it up to >90 dB SPL to overcome room noise from hiding the low tone