[LEAPSECS] LEAPSECS Digest, Vol 45, Issue 1

Zefram zefram at fysh.org
Fri Sep 3 08:02:59 EDT 2010

Tony Finch wrote:

>On 3 Sep 2010, at 01:41, msokolov at ivan.Harhan.ORG (Michael Sokolov) wrote:

>> I very soon will, as soon as I get my rubber time generator working.


>Oh, do tell, where will you get your GMT reference from?

If I were doing it, I would take the DUT1 projections from IERS Bulletin
A <ftp://maia.usno.navy.mil/ser7/ser7.dat>, interpolate between them,
and add the resulting DUT1 onto a UTC clock that is synchronised by NTP.
At the millisecond level this should provide a satisfactorily smooth
approximation of UT1. Of course, it will not reflect any diurnal
influences on DUT1, nor any unpredictable rapid changes such as those
resulting from earthquakes. There would also necessarily be some
discontinuity when switching from one set of projections to the next
later-produced set of projections; optionally one could paper over that
by slewing gradually between sets of projections, though this would
degrade the accuracy of the simulated UT1 frequency.

I note that in this endeavour leap seconds in UTC are a slight
inconvenience, because they result in discontinuities in DUT1. The clock
logic would have to treat these discontinuities specially. It would be
slightly easier if Bulletin A supplied UT1-TAI instead of UT1-UTC, and the
underlying NTP-synchronised clock supplied TAI instead of UTC. Actually,
probably the easiest way to handle the leap second discontinuities
is to transform Bulletin A and the NTP clock into this TAI-based form.
So it's not really much more inconvenience than the difficulty of getting
a clock to tick UTC through a leap second in the first place.


More information about the LEAPSECS mailing list