[time-nuts] Characterising frequency standards

Poul-Henning Kamp phk at phk.freebsd.dk
Tue Apr 7 16:22:06 UTC 2009

In message <1231b6a80904070509m5b8bb638gbc088500444254c3 at mail.gmail.com>, Steve
 Rooke writes:

>So why am I saying this, well from what I have read on this group and
>on the web, I have been left with a feeling that it was vital to
>capture every event over a samplig period to ensure an accurate

It is vital only to simplify the calculation of the uncertainties
on the result more than the result itself.

If you skip every other time interval, you have no information about
noise of the obvious 1/p frequency, just like Nyquist says.

Dividing a 10MHz signal to 1PPS, and measuring the adev on that,
therefore gives us no right to talk about what happens on the
fast side of tau=1sec.

Aperiodic sampling can be an incredible powerfull tool to use
instead:  Comparing the two 10MHz signals by measuring the
difference i duration between ramdomly chosen sequences of
thousand samples, gives very detailed information, as long as
you know the exact relative placement of your 1k sample
runs relative to each other.

The mathmatical handling is nasty though.

Poul-Henning Kamp       | UNIX since Zilog Zeus 3.20
phk at FreeBSD.ORG         | TCP/IP since RFC 956
FreeBSD committer       | BSD since 4.3-tahoe    
Never attribute to malice what can adequately be explained by incompetence.

More information about the time-nuts mailing list