Query about the "jitterstream"

[dmagnus1]

Been a long time since I have posted here - but still frequently peruse the topics and learn a great deal. In the past, I have often found answers to simple questions that seem to be hard to find otherwise, and have always appreciated the help. This time I am here with a question about "jitter"...

 

I think I understand (superficially) the importance of digital signal timing and the faithful reconstruction of an analog waveform. I am using a Naim SuperNait with integrated DAC that only accepts SPDIF digital inputs, so I feed the USB output from my Sonore mRendu to the Naim SN through an iFi USB/SPDIF converter (which claims to reclock the signal to reduce jitter). Each of these devices has different jitter rejection specs, and I am wondering if the level of signal jitter that affects the DAC's performance is determined by the "best" or the "worst" clock in the system. More specifically, I think the Naim SN reclocks incoming signals with jitter rejection to about 100 pSec - which sounds a bit high compared to more contemporary products. My mRendu was upgrade to the uRendu "femto-clock", so I assume its output is timed to something like 100 fSec. I think the iFi, which also "eliminates" jitter probably reclocks the signal to specs somewhere in between.

 

So - if the built-in jitter rejection and reclocking mechanism at distal end of the chain is less precise than that of the incoming signal, is all that upstream jitter rejection technology for naught?

Also - what constitutes an acceptable level of "jitter rejection" these days? Seems every digital product I read about boasts femto-second specs. Is that for real? What is the audible threshold?

 

I'm horrified to think of how many conceptual errors I have just exposed, but I hope someone can shed some light on this for me. Thx in advance!

[JohnSwenson]

Hi dmagnus1,

you are inquiring into one of the most complicated and controversial subjects in audio, so worry about not understanding it well.

 

First off the "number" for jitter, this is like giving a single "performance number" for a car it is almost meaningless without defining exactly how it is come up with and what it means. As far as jitter goes there are a whole bunch of different measurements of jitter that all give a single time number (ps, ns, fs etc) yet these measurements are quite different. For example I have a particular oscillator that when measured one ways gives 11ps and another 97fs, kind of a big difference.

 

The result is that using a "ps" number in comparisons is only valid when the same person, using the exact same test equipment in exactly the same way is doing the measurements. Comparing numbers coming from different companies is essentially meaningless. Even if they specify exactly WHAT the measurement is, using different test equipment, or even the same equipment in different ways can give different results for the SAME test.

 

And all this is assuming that what you are measuring in time units (ie ps) is even correlated to sound quality. My experience seems to to be showing that using a spectral measurement of jitter (ie phase noise) seems to have a greater correlation to sound quality. But GOOD phase noise measurement equipment is VERY expensive. Unfortunately understanding a phase noise plot is not easy and again comparing two is fraught with peril.

 

Then there is what you are measuring. Are you measuring a clock right at the oscillator, the DAC chip pins or somewhere else? You can get radically different results depending on where in a circuit you measure it. Then there is the interesting part that the inside of a DAC chip can be much worse than the jitter at the package pin. What is going on inside the chip can make it worse. Unfortunately there is essentially NO way to measure THIS directly.

 

The upshot of all this is don't even try to select equipment based on jitter numbers, it is meaningless. This doesn't mean that jitter doesn't matter, or that everything is the same, far from it, it just means that the commonly bantered numbers are meaningless for use in comparing the jitter from one device to another in any way that has any correlation with sound quality.

 

John S.

[CG]

Truer words have seldom been written!

 

In communications systems, it's generally more valuable to measure the spectrum of the phase noise, as John describes.  Not easy or cheap, by a long shot.

 

For audio reproduction, you'd think that there'd be a way to measure the phase noise sidebands of a synthesized tone at the output of a DAC (DAC as in a complete digital audio converter, not just a DAC chip).  It wouldn't be easy or cheap, but possible.  That might give a clue as to the effect of phase noise on the audio output.  Of course, it would also be combined with other effects, but those affect the sound, too.

 

One gotcha:  Almost all measurements in the audio world are made using various averaging or smoothing techniques.  This is done with the idea that noise is random, while the desired tones you are trying to measure are not.  So, if you take a bunch of repeat measurements, the noise will average to a much lower level because the noise is assumed to be completely random, while the desired tones will remain constant.  This lets you detect lower level signals within the noise.

 

That is true, but probably misses a key point.  

 

Many "phase noise events" are transitory by nature.  So, you'd only see them once in a while.  Once in a while, as in every 1000 sample sweeps.  There you might witness an awful spectrum due to a bunch of factors.  The next sample sweep could be fine.  In a communications system, that symbol is lost and is a loss of actual bits.  But, one lost symbol out of a thousand gives the misleading impression that everything is ok when you apply averaging.  Not so if you count bit errors.  (If you don't believe me, try measuring MER versus BER some time...)

 

But, is one bad audio sample out of a thousand actually audible?  I'd guess so, but I don't know of any research that concludes anything either way.  (If anybody knows of any, please post a link!)

 

The point is, you probably really need to use a different approach to measuring the effects of phase noise in an audio system as I described.  (You could argue this is necessary for plain old distortion, too, but that's another topic for another forum.)  Perhaps using a peak or max hold function would be more valuable.  You'd just retain the highest value in every FFT bin for several thousands sweeps and display that.  This isn't that difficult to do in software.  (Here's one simple example - http://www.w7ay.net/site/Applications/Amici/index.html)

 

In addition, phase noise on a clock effectively modulates each and every converted tone in the audio spectrum with that noise content.  If the frequency content of the phase noise is low enough, that's pretty much the same thing as rocking your loudspeakers back and forth at whatever rate the phase noise modulation is.  Maybe that's bad; maybe that's even good!  Dunno.

 

I apologize to the OP for going off into the esoteric here, but the topic probably is worth investigating.

[Superdad]

37 minutes ago, CG said:

Truer words have seldom been written!

 

Actually they have. By you, just now! 

Warms my heart to read somebody who understands the limitations of the ubiquitous FFT.

[CG]

What you're feeling is heartburn...

 

Actually, it isn't the limitation of the FFT, but how it's used.  Like anything else, FFT's have limitations, but how we typically apply and interpret them may not be in our own best interest.  It's too bad, really.

[dmagnus1]

Thanks for the replies - I may need to go back to school...😉

 

So I guess it still comes down to actually listening!

 

And thx to JS - I love my sonicTransporter i5 and mRendu. I agree with Chris that its nice to be able to access 50M tracks on Roon/Tidal with outstanding sound quality and reliability and do it for less $ than a cigarette boat...

[pkane2001]

7 hours ago, CG said:

Truer words have seldom been written!

 

In communications systems, it's generally more valuable to measure the spectrum of the phase noise, as John describes.  Not easy or cheap, by a long shot.

 

For audio reproduction, you'd think that there'd be a way to measure the phase noise sidebands of a synthesized tone at the output of a DAC (DAC as in a complete digital audio converter, not just a DAC chip).  It wouldn't be easy or cheap, but possible.  That might give a clue as to the effect of phase noise on the audio output.  Of course, it would also be combined with other effects, but those affect the sound, too.

 

One gotcha:  Almost all measurements in the audio world are made using various averaging or smoothing techniques.  This is done with the idea that noise is random, while the desired tones you are trying to measure are not.  So, if you take a bunch of repeat measurements, the noise will average to a much lower level because the noise is assumed to be completely random, while the desired tones will remain constant.  This lets you detect lower level signals within the noise.

 

That is true, but probably misses a key point.  

 

Many "phase noise events" are transitory by nature.  So, you'd only see them once in a while.  Once in a while, as in every 1000 sample sweeps.  There you might witness an awful spectrum due to a bunch of factors.  The next sample sweep could be fine.  In a communications system, that symbol is lost and is a loss of actual bits.  But, one lost symbol out of a thousand gives the misleading impression that everything is ok when you apply averaging.  Not so if you count bit errors.  (If you don't believe me, try measuring MER versus BER some time...)

 

But, is one bad audio sample out of a thousand actually audible?  I'd guess so, but I don't know of any research that concludes anything either way.  (If anybody knows of any, please post a link!)

 

The point is, you probably really need to use a different approach to measuring the effects of phase noise in an audio system as I described.  (You could argue this is necessary for plain old distortion, too, but that's another topic for another forum.)  Perhaps using a peak or max hold function would be more valuable.  You'd just retain the highest value in every FFT bin for several thousands sweeps and display that.  This isn't that difficult to do in software.  (Here's one simple example - http://www.w7ay.net/site/Applications/Amici/index.html)

 

In addition, phase noise on a clock effectively modulates each and every converted tone in the audio spectrum with that noise content.  If the frequency content of the phase noise is low enough, that's pretty much the same thing as rocking your loudspeakers back and forth at whatever rate the phase noise modulation is.  Maybe that's bad; maybe that's even good!  Dunno.

 

I apologize to the OP for going off into the esoteric here, but the topic probably is worth investigating.

 

Here's a way to try to go about determining audibility of various types of jitter: https://distortaudio.org/

 

DISTORT is still in development (I call it a continuous beta ), but already has a few useful features.

 

One, you can apply any number, frequency and level of sine-modulated clock jitter to an existing recording and then listen. You can add random jitter. Or correlated jitter. Or 1/f noise-modulated (close-in) jitter. Or any combination of the above. The recording can be as simple as a sine wave, or as complex as a symphony -- your choice. Apply jitter, save it, then play it back on your 'perfect' system and listen to see if you can hear the effect. Or even listen directly on your computer with DISTORT playing the audio file. You can then switch nearly instantaneously between the original and the distorted tracks using the Bypass button.

 

Here's some correlated, 1/f and sine-modulated jitter applied to a simple 1kHz sine wave:

 

Oh,  and best of all -- no FFTs are used to generate jitter or harmonic distortion, everything is done in the time domain.

 

 

[CG]

11 hours ago, pkane2001 said:

 

Here's a way to try to go about determining audibility of various types of jitter: https://distortaudio.org/

 

 

Very clever!  Two thumbs up.  (If I had a Windows computer with the necessary horsepower, I'd be right on this.)

 

This still leaves open the question of just what the jitter slash phase noise profile might be in a real system, but it looks to be a great start.  

 

Somewhat OT: I wandered through some of the commentary on the support forum and very much appreciated the distortion analysis you provided of what out of band signals do to the audio spectrum.  Wonder why power line filters and even interconnect and power cables might affect the sound?  Or, even, rectifier switching in a power supply?  There's your answer.  With all the stuff conducted through and coupled from the power lines around the house due to switching power supplies and various digital control circuits, it's no wonder everything seems to matter.  It's actual physics based phenomena, not voodoo.

 

BTW, how did you determine the out-of-band linearity performance of the amplifiers in question?  Is it extrapolation from the published in-band performance?  I missed that.

 

http://www.audiodesignguide.com/Ibridone/Sen_Semi_Diode_Apps-quik108.pdf

[pkane2001]

9 minutes ago, CG said:

BTW, how did you determine the out-of-band linearity performance of the amplifiers in question?  Is it extrapolation from the published in-band performance?  I missed that.

 

http://www.audiodesignguide.com/Ibridone/Sen_Semi_Diode_Apps-quik108.pdf

 

The out of band analysis you saw is caused by the non-linearity of the system causing inter-modulation distortion (IMD) and aliasing from the energy in frequencies above the audio range to reflect back into the audible range. While aliasing of out-of-band frequencies is unlikely to happen in a DAC (much more likely in an ADC), IMD can happen in both.