192kHz and jitter

[Rad]

I'm wondering if anyone here knows whether there is any scientific basis for the claim that at 192k, jitter becomes more of an issue than at lower sample rates, like 44.1?
My standard reference (Aldrich) didn't have anything on that. Yet I know various people who use this as a justification to prefer 44.1 and offer all kinds of unclear theories as to why it's true. So I'm wondering if it is that they misunderstood something that is actually true, or this whole thing is nonsense as I suspect.

[dpd]

Just literally off the top of my head, Rad, I would think that the jitter time spread, as a percentage of the sampling rate, would be a greater percentange of that sampling rate, and cause more sampling errors (sampling the data at an incorrect time). However, one still needs to consider the spectral spread of the jitter as that tends to change the spectral content of the sampled signal, via modulation effects.

1st I have heard of this...

[Dave Burris]

I would think it would be directly dependent on the jitter frequency as a ratio to sample rate and the jitter amplitude. I'm not sure there's a blanket statement about this. It's almost entirely related to the nature of the jitter.

[RawknRoll]

you guys make my head hurt.

any books you could recommend for me to read up on all this technical jargon?

Thanks!

[DAS]

Too many different kinds of and causes of jitter to make a blanket statement like that. However tolerances do get tighter as frequency goes up so I think there is room for things to potentially get slightly worse at higher rates. I know that Lavery goes on about this sometimes, but I'm not aware of anyplace he actually explains it.

Rawk - Nika's book (that Rad refers to above) does do a good job of taking you from square one all the way through to some pretty high level concepts. It's not going to make you a digital designer, but it does explain a lot of it in plain English.

http://www.sweetwater.com/store/detail/NikaBook

[Rad]

Dpd and Dave - both of you make very interesting points. I'll see if I can talk to someone who actually designs convertors to see what they say.

RawkinRoll - yes, I also highly recommend Aldrich's book. This is the best comprehensive intro to digital audio I have seen in many, many years. Recommend it very highly for any level.

[dpd]

BTW, while assembling a dual MOTU 24 channel converter DAW into a piece of test equipment for my company (use it to simulate ocean-borne acoustic signals), I had the opportunity to compare the spectral purity of a 24/96 sine wave generated in Matlab using the MOTU internal clock and Apogee's Big Ben on a high end HP Digital Signal Analyzer (800 pt FFT). Zooming in to a pretty narrow bandwidth (maybe 10 hz?) the only difference I noticed coming out of the MOTU converters was a difference in absolute frequency of a fraction of a percent; I measured no perceivable jitter artifacts with either clock.

I wish I could have spent more time on these experiments; however, what I learned was enough to know that the Big Ben - in this application - demonstrated no quantifiable benefit.

[Rad]

Dpd,
When you say a fraction of a percent, I'm wondering how that translates into absolute time units?

The usual wisdom about jitter is that jitter need not be large by any absolute scale in order to have audible artifacts - usually jitter in the order of picoseconds will suffice to generate distortionary "sidebands". I'm wondering if, say, half percent deviation of 96k is still sizeable if measured in picoseconds^(-1)?

Otherwise, my guess is that the Big Ben offers greater stability for a wider range of temperature and humidity conditions than a regular clock. But in a regulated environment like 70F that difference may be unobservable.

An interesting post, overall. Thanks for sharing the info!

[dpd]

I think I did the test at a pretty low frequency (300 Hz, IIRC), so I'm sure that's not the best way to review jitter.

I measured the output frequency on a high-resolution HP frequency counter and I think I was measuring .05 Hz differences in absolute frequency, or something in that range. BTW, neither clock was perfect: one was off high, the other was off low. My guess is that you could probably tweak the crystal oscillators in each to be more accurate on an absolute basis (the test waveform was a 300.0000 hz sine wave synthesized in Matlab, so the baseline frequency was dead-nuts accurate)

It was just one data point, but at least it was a quantifiable one... If I could score a Big Ben from my friends at Sweetwater for a day (they're 10 minutes down the highway), I'd love to try more tests at various frequencies and test amplitudes.