Jitter - need some clarification please

[glitch]

Okay, now I understand (hopefully correctly) that there are two kinds of jitter - converter (difficult to measure), and data (measurable in a DAW).


Now, I'm seeing a LOT of numbers thrown around.

Black Lion claims their mods produce 7-10 picoseconds of jitter on the clock.
The Apogee PSX-100 is 5 nanoseconds - that's 500 picoseconds.
The Mackie Onyx800R is measured in a review at 7 nanoseconds.
RME's steady clock (which is, arguably, comparable to the Big Ben) is approximately 2 nanoseconds.

What's going on? Why are the Black Lion specs about 50x as good as the Apogee?

Are they measuring something different than what we're seeing posted elsewhere?


Help?

[Nerdyrocker]

Unfortunately, jitter is alive and well in every system. On paper, lower jitter sounds better. But then again, that is on paper. Some manufacturers use jitter to create "the sound" of their converters.

[tINY]

You have to define jitter better to make any comparison.

I would guess that the BL spec is cycle-to-cycle jitter which is the difference in length of one period to the next. So, if one cycle (from one sample to the next) is 20.833010 uS and the next cycle is 20.833000 uS, then the cycle to cycle jitter is 10pS.

Accumulated jitter is measured from a theoretically perfect clock. So, if you add all the cycle to cycle jitter for a few thousand periods, you will get accumulated jitter. If you keep a running tab on this, you will see that it's sometimes the clock is fast and sometimes it is slow.

Sometimes this accumulated jitter is measured peak to peak and sometimes it is mesured as RMS. Usually specs will be RMS. There is no way that the Black Lion mod results in 7-10 pS RMS. If they can do that, they really should be building sattelite radio relays - they'd make a lot more money.




-tINY


BTW - there is 1000 pS in 1 nS

[Ethan Winer]

> now I understand (hopefully correctly) that there are two kinds of jitter - converter (difficult to measure), and data (measurable in a DAW). <

No, there's only one kind of jitter - the kind that is inaudible. Really, it's a big ho-hum who-gives-a-crap non-issue. If jitter actually mattered you'd see it expressed as dB instead of picoseconds, and the levels would be within an audible range. Like hiss or distortion that's 90 dB down or whatever. Jitter these days is typically 120 dB below the music, and present only while the music plays. So who cares? You can't hear it. Nobody can hear it.

Sorry for the rant. I just hate seeing people spend money and waste time on stuff that doesn't matter even a little.

--Ethan

[Chris Parsons]

Ethan 1....Gearslutz 0

***DING***

[tINY]

Of course Ethan is right.

You would have to have RMS jitter on the order of 100 uS before it would be much of an issue. Audio is almost DC....

I work with high speed busses where 100pS is too much. But, error detection is usually built into digital communications, so it just gets resent.

The digital audio interfaces are so slow that errors are non-existant on properly set-up systems.



-tINY

[Snatchman]

OK...Now that we understand jitter is moot, what's the determining factor(s) in a "good clock" design..?..

[kats]

I read somewhere that error correction is important.

But I dunno, I just plug em in....

[bob katz]

Jitter numbers are like CPU benchmarks. Everyone has a different method of measuring it, and very few people own the expensive analyser with sufficient resolution to do it justice. There are weighting figures and average and peak figures and peak to peak figures and it's a mess.

So you can't compare one manufacturer's jitter spec against another unless you find out EXACTLY how they measured it.

Most unfortunate. The method I use for determining jitter susceptibility of a DAC is Julian Dunn's jitter test signal and a spectrum analyser looking at the analog output. For A/D, an 11.025 or 12 kHz signal for 44.1/48 kHz, respectively will do, and look at a spectrum analyser. An excellent paper showing these measurements, which can be viewed visually, is at TC Electronic. Instead of looking at numbers in pico seconds, just look at the sidebands of the test tone on an FFT. If they're not right down near the noise floor, and the skirts aren't very narrow, then the jitter is, frankly, unacceptable. I reach that conclusion because I can hear the jitter in any situation I've ever encountered where the spikes measure more than a few dB above the 24 bit noise floor.

You can measure whether a converter is sensitive to jitter on an external input, measure WC versus AES sync versus internal. You can learn all you need to know with Spectrafoo, a clean test tone oscillator, and your brain.

And in fact, if every manufacturer were more honest, they would publish jitter tests with the same approach. Benchmark's A/Ds and D/As use this spectral analysis approach and it's very revealing. Pico seconds, shmeeko seconds.

BK

[glitch]

Hmmm... thanks for all the info!

[tINY]

That's an awfully sane aproach to jitter testing. Do you have a link to a white paper or a description of the test signal? (It sounds like you may be describing duty-cycle distortion AKA "bang-bang" jitter).


It must be nice to have equipment that has noise down at the -144dB(FS) range.




-tINY