[time-nuts] Crystal aging
phk at phk.freebsd.dk
Thu Sep 3 11:57:52 UTC 2009
In message <E28A3329130C445DB0A3FA4C3ED0958B at athlon>, "Ulrich Bangert" writes:
>The true thrill is however one step more subtle: It first considers the
>question WHY trimble choose 100 s as the default loop time constant. Well,
>this one is easy to answer: Just set the time constant of a "cold" TBOLT to
>1000 and watch the TIC value flying to the moon. Expect days before the loop
>locks. Trimble NEEDS to set the deafult time constant that low in order to
>make the loop of a cold TBOLT lock within a reasonable time.
This argument is utttrly bogus: adaptive time-constant PLL's are at
least 30 years old, and rather trivial to implement...
In NTPns I start out with a timeconstant of about 4 seconds in order
to get rapid capture, depending on your timebase, it will increase
to several hours (OCXO) or days (Rb).
The only "hard" thing about adaptive PLL's is knowing when you have
torqued the timeconstant to high, but monitoring frequency of zero-crossing
of the residuals (a cheap approximation to a periodogram) solves that
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