Really controversial idea...

[amcluesent]

>this thread was too big an invitation.<

Interesting. So I guess some form of ring buffer to decouple the input word stream from the DAC wouldn't work as even a fractional difference in mean clock frequency between transport and DAC would lead to near instant buffer over/under runs.

So if a PLL is needed to lock the DAC clock to the input stream recovered clock, then that PLL circuit needs a super-stable reference clock itself and some form of damping to minimise tracking jitter.

[opaqueice]

This is an interesting case. They use a high frequency clock (I think its something like 110MHz) to reclock the data going to the DAC. This DOES make the system almost immune to jitter on the input, but DOES effectively create jitter on the output. The clock might be low jitter, but its reclocking the 44.1 (or whatever). What hits the DAC chip is pulled in or pushed out to the nearest clock pulse of the high frequency clock. This produces distortions in the waveform exactly the same as jitter on the 44.1 clock. This is a perfect example of being immune to jitter does not necessarily mean its LOW jitter.

Have you measured that, or do you know a reference?

Another interesting idea is the Lavry way of doing things, where the local crystal has an adjustable frequency, and there's something that monitors the state of the buffer and adjusts the clock speed as needed. Is that what you were talking about above (VCXO etc.)?

[ezkcdude]

Here, I tested it for you!

http://www.seanadams.com/jitter_fiber_test/

Cool test. I wonder if making the toslink signal brighter (is it an led?) would decrease jitter. I was just reading something today about how researchers have found a way to coat fiber optic cables on the inside so that there is less light leakage going around corners and such.

[Phil Leigh]

Here, I tested it for you!

http://www.seanadams.com/jitter_fiber_test/

Sean - you are (still) my hero and you saved me $70! Atthis rate I can afford my own pony soon.

So this test (which must go down in the annals of audio history by the way) proves one of two things:

A) toslink is so bad to start with that only really violent abuse makes it any worse than it already is
or
B) toslink is actually pretty good and quite resistent to abuse


You didn't do any listening at all did you? it would be interesting to know how the respliced cable sounded compared to the new one...

[seanadams]

Sean - you are (still) my hero and you saved me $70! Atthis rate I can afford my own pony soon.

So this test (which must go down in the annals of audio history by the way) proves one of two things:

A) toslink is so bad to start with that only really violent abuse makes it any worse than it already is
or
B) toslink is actually pretty good and quite resistent to abuse

I'd say they're both true depending on what you want to use it for. If you believe that jitter below 1ns is not audible, or you just aren't concerned about it, then there's nothing wrong with toslink.

You didn't do any listening at all did you? it would be interesting to know how the respliced cable sounded compared to the new one...

No I only spent a couple minutes hooking up each test case and grabbing a screenshot. This was not intended to be any kind of thorough test.

[Phil Leigh]

I'd say they're both true depending on what you want to use it for. If you believe that jitter below 1ns is not audible, or you just aren't concerned about it, then there's nothing wrong with toslink.



No I only spent a couple minutes hooking up each test case and grabbing a screenshot. This was not intended to be any kind of thorough test.

Fair enough!

When we get down to nanoseconds my brain starts to hurt because isn't the interval limit of what a human can discern in the single-digit milliseconds? I can't even imagine what a nanosecond is. I know people go on about 20pS of jitter - and I know you've often rightly questioned what they mean by that. I just don't think jitter lends itself to a measurement-audible quality correlation in the same way as (say) THD or IMD.

Cheers
Phil

[tot]

When we get down to nanoseconds my brain starts to hurt because isn't the interval limit of what a human can discern in the single-digit milliseconds? I can't even imagine what a nanosecond is. I know people go on about 20pS of jitter - and I know you've often rightly questioned what they mean by that. I just don't think jitter lends itself to a measurement-audible quality correlation in the same way as (say) THD or IMD.

20kHz is 50ns for full cycle, or 25ns for half cycle. Some parts of the body are clearly quite fast reacting.

Teemu

[Jaco]

My believe is that all these different approaches of slaving the DAC to the clock of the digital source, or by using PLL's or re-clocking/re-sampling are fundamentally flawed. It is the DAC that should be the clock master and not the source. That immediately solves most of the jitter issues if the DAC has a low-jitter master clock.

Fortunately some companies have realised this and started taking this approach. For example, see the Pace-Car by Emperical audio (http://www.empiricalaudio.com/frPace-Car.html).

Another company taking this approach is LessLoss (see http://www.lessloss.com/). By the way, the latter web site is a useful resource of information regarding jitter with some nice animated pictures of what it looks like.

[Phil Leigh]

20kHz is 50ns for full cycle, or 25ns for half cycle. Some parts of the body are clearly quite fast reacting.

Teemu

When I was 18 I applied for a job at the BBC. They did a hearing test. I was 3dB down at 17Khz. (I got the job) I'm probably about 6dB down at 15Khz these days (30 years later).
However, the point is could I reliably hear 1ns of frequency/timing variation? I don't believe I could, but I'm willing to test it.

[Phil Leigh]

My believe is that all these different approaches of slaving the DAC to the clock of the digital source, or by using PLL's or re-clocking/re-sampling are fundamentally flawed. It is the DAC that should be the clock master and not the source. That immediately solves most of the jitter issues if the DAC has a low-jitter master clock.

Fortunately some companies have realised this and started taking this approach. For example, see the Pace-Car by Emperical audio (http://www.empiricalaudio.com/frPace-Car.html).

Another company taking this approach is LessLoss (see http://www.lessloss.com/). By the way, the latter web site is a useful resource of information regarding jitter with some nice animated pictures of what it looks like.

I'm kind of more interested in what it actually sounds like. Valve amps have (typically) higher THD but sound "lovely".

...and yes of course the DAC should be the clock source. But, using a PLL is fine and there are many systems out there that sound fantastic using that method. There are no absolute truths - only things that work and things that work "better".
Enjoy the music
Phil

[Jaco]

Thanks for your reply Phil. Yes, I'm also interested in hearing what different amounts of jitter sound like. That could be an intersting experiment.

A good friend of mine is a recording engineer and we've experimented quite a lot with the "good sounding" harmonic distortion of valve pre-amps. It's an eye-opening experience for most people that have not being exposed to it before!

Being an electronics engineer with a master's degree in digital signal processing I typically prefer the more elegant technical solutions. :-) Unfortunately that is not what I'm seeing with the transport being the clock master - hence my leaning towards the solution where the DAC is the clock master which simplifies things considerably, and hopefully bringing down the price of high quality digital audio.

I agree that existing solutions using the PLL or re-clocking/re-sampling approaches do sound good. However, the really good ones do come at a considerable cost in financial terms My believe is that the DAC as clock master can make even a cheap CD player a great digital source. Now we need to work on the cost of DAC's which can act as clock masters...

That's just the engineer in me striving towards the more elegant and more economical solutions...

[seanadams]

20kHz is 50ns for full cycle, or 25ns for half cycle. Some parts of the body are clearly quite fast reacting.

Teemu

Close - you're only off by a factor of a thousand.

milli = 10^-3
micro = 10^-6
nano = 10^-9
pico = 10^-12

At any rate, the ear doesn't work like that. It picks up frequencies, not samples over time. http://en.wikipedia.org/wiki/Organ_of_Corti

Anatomically speaking, I suppose you could say that jitter would, for each frequency component of the signal, cause additional slight stimulation of the hair cells right next to the intended one for that frequency.

[Patrick Dixon]

I just don't think jitter lends itself to a measurement-audible quality correlation in the same way as (say) THD or IMD.

In many ways there are similarities between jitter and distortion measurements, as they are all commonly expressed as 'a number', although the spectral content behind the headline number is probably just as important.

[Phil Leigh]

In many ways there are similarities between jitter and distortion measurements, as they are all commonly expressed as 'a number', although the spectral content behind the headline number is probably just as important.

Patrick - that's what I meant in my inelegant way...the headlines are not the whole story...

[tot]

Close - you're only off by a factor of a thousand.

I had a gut feeling something was terribly wrong, but went ahead regardless .. :-)