slightly ot: DAT clock deviation

Magnus Danielson cfmd at swipnet.se
Mon Aug 7 21:52:29 CEST 2000


From: Martin Czech <czech at Micronas.Com>
Subject: slightly ot: DAT clock deviation
Date: Mon, 7 Aug 2000 09:21:05 +0200 (MET DST)

Hi Martin,

> Does anybody know about usuall DAT clock jitter and drift specs?
> I mean the sampling frequency should be 48.00000 kHz, but
> there will be offset, drift (temperature) and long term drift (aging).
> I'm especially interested in short term drift (1s-100s).

Howha! Now you are moving into unchartered areas I fear... I have never seen
anything like it and I even doubt that it would be easy to find any
measurements at all. I would go the actual spec in hope of a requirement or
recommendation there.

Even in a rather detailed description of the Compact Disc system I can find
even maximum deviations of sampling frequency, this figure is often mistaken
for describing the system's jitter. So, I guess you are out of luck.

I guess that since the signal now is locked to a crystal oscillator enougth
damping is there for it to be "sufficient for audio applications".

The first thing which can break for you is the reference oscillator and it's
temperature sensitivity and speed of aging. In the range of your interest
(1s to 100s) I would estimate that it is the temperature drift which is the
number one succer for you. I wouldn't be supprised if it is just a standard
"computer grade" crystal oscillator sitting there... your wrist-watchs
oscillator is better than that! The type of oscillators sitting all over the
place in a computer is usually something like +/- 100 ppm and by no means
have much stabilization tricks been taken on. I know from my own experience
that it is easy for them to drift away quickly for small changes in
temperature, and this QUICKLY!

There are ways to destroy the properties of the oscillator, but most of these
relate to jitter (phase deviations with frequencies above 10 Hz) and not
wander (phase deviations with frequencies below 10 Hz). For instance will
temperature dependencies of components within a PLL loop translate into
phase deviations.

Naturally, if you want to do high preformance measurements you can not solely
rely on crystal oscillators even thougth their short-term stability is really
supperiour to most anythings you could lay your hands on.

> I want to record the same thing severall times and add
> the recordings up, in a matched manner. 

Now, how much is really 100 ppm of difference?

Well, let's say you sample along in 48 kHz, then 100 ppm of that would be

48000 * 0,000 100 = 48 * 0,1 = 4.8 Hz

Thus, if you experience a full 100 ppm of frequency difference between each
sampling opertunity you would drift 4.8 samples per second.

Now, I don't think you would normally see a full 100 ppm, you should expect to
see less than this. A +/- 100 ppm oscillator specifications really says that
3 sigma (99.7%) of all oscillators made will be within the +/- 100 ppm
boundary. A deviation of one or a few ppms is much more realistic.

So, now my key question is, just _how_ on the sample do you have to be?

If you like I could set up my portable AIWA DAT with a tone and then measure
the frequency (and deviations) as it plays, I can do that with a fair
precission and my longterm stability is acceptable for this measurement ;)

Naturally, there are measures you could be making in order to reduce your
temperature instability, but I guess you are able to dream these up yourself ;)

> I could compensate for initial delay, but not for arbitrary drift.

Certainly...

So, how accurate are you required to be?

Cheers,
Magnus



More information about the Synth-diy mailing list