[time-nuts] Re: Does cm accurate GPS via RTK give more accurate time?

[Poul-Henning Kamp]

--------
Geoffrey Baehr writes:


> 3 - I don’t know if I can believe chronyc  saying it’s 10nS off (from what ?) 

I suspect what they're actually saying is, that their PLL never
estimates the phase error to be more than 10 nanoseconds.

That only tells you that they are really good at tracking the input
signal, it says nothing about their resulting stability and it
strongly hints that their PLL is (too) loose.

With modern hardware a loose PLL is not necessarily a bad strategy,
but it does have the downside that during loss of reference, it is
anyones guess what your last frequency estimate was, and therefore
your clock may rapidly wander of target.

I usually explain it like this:

The XTALs used have spectral purity as controlling parameter, because
their output get PLL'ed into the GHz domain.

The XTAL temp-co is typically 1 PPM/K.

Well adjusted air-con swings the temperature +/- 1-2 K over a 20
minute period.

That means that the XTALs, left alone, will swing +/- 1-2 msec over
the same 20 minute period.

A loose PLL will try to out-compensate the 20 minute swing, giving
you timekeeping better than the 1-2 msec, at cost of rapidly
wandering into the weeds during LOS.

A stiff PLL will average out the 20 minute swing, giving you 1-2 msec
timekeeping, also during reasonably long periods of LOS.

Pick your poison...


-- 
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.