[ntp:questions] Change reference clock soon after DCF-signalislost

Rob nomail at example.com
Thu Jul 15 16:21:13 UTC 2010

Matuschka, Sebastian <Sebastian.Matuschka at gcd-solutions.de> wrote:
>> I think this isn't a good design.
>> In my experience, DCF-77 reception is simply not stable enough to
>> directly
>> use the pulses from the receiver as clock ticks.
>> When there are thunderstorms, local interference, and sometimes
>> propagation
>> problems, there can be spurious extra pulses that you do not want to
>> count.  Or pulses can be missing.
>> A good way of using DCF-77 is to collect the 59 pulses that make up a
>> minute, average the offset between the pulse start and the local clock
>> tick,
>> decode the time from the pulse lengths, and then at the end of the
>> minute
>> decide if all this information is valid and should control the clock
>> (adjusting the clock offset/frequency), or should be discarded as a
>> whole.
>> This is also what the DCF-77 drivers in ntpd do.
> Sorry, i forgot to mention an important thing:
> The DCF-77 Signal is generated by a precise clock that is connected via
> fiber optics to the device the ntpd runs at.
> The distance isn't very long, so i can expect that the signal is very
> precisely and there is no jitter nor any (relevant) offset.
> So, it has not really has anything to do with a normal DCF-77 receiver.
> I hope my questions make more sense now :)

I think you mean that you have a 1PPS source that is synchronized to
DCF-77 but does not send the actual received DCF-77 signal.
In that case it should be OK to use it to increment a clock.

However, when you have expectations that you can control what source
is used to sync your clock, and how quickly that switches, you cannot
use ntpd.  It does not offer that control.

More information about the questions mailing list