NTPsec leap second experience

[Achim Gratz]

Achim Gratz writes:
> The second one was configured as a stratum1 (via GPS w/ PPS) and
> monitoring the PTB servers for reference.  Since apparently Jessie has
> not been updated with the correct leap second file (it still doesn't
> recognize the leap second as a valid datum), the GPS got marked as a
> falseticker when the leapsecond arrived and the NTP loop opened.  I've
> arrived back home almost two days later to restart NTP.  I keep that
> system at a temperature that has close to zero TC (I'm not regulating
> the temperature yet, but it seems that should be possible) and during my
> absence the room temperature was also quite stable.  Accordingly the
> clock had only drifted a little bit less than 1ms (I recorded all PPS
> time stamps).  It probably would have drifted even less if there hadn't
> been a GPS burp shortly before the leap second and the loop hadn't
> stabilised to the correct offset again.

I've looked a that data in more detail today.  It looks like I was
within about ±0.2K of the zero-TC point.  There are no obvious signs of
any temperature related effects in the data from the residual
temperature swing of ±0.5K for the 47 hours the NTP loop was open.

Assuming an initial frequency offset where the loop opened regresses to
about 4ppb offset.  While that's a plausible number, it should produce a
linear time drift, and the actual drift is quite obviously quadratic.
Consequently, the residuals of the linearly corrected time stamps drift
to +200µs before going the other direction and heading to -200µs.

Plugging in drift from linear aging and no initial offset keeps the
residual changes within ±30µs.  The aging from that regression
calculates to about 7ppb/day.  The NTP frequency offset before and after
the loop was open changed by about 19ppb/47h, while looking at some
longer periods of undisturbed operation the XO drift has been between
3ppb/day and 10ppb/day.

Trying to model an initial frequency offset plus linear aging indicates
an initial offset around 1ppb (which is the last digit the loop stats
put out and so is within the rounding error), but doesn't change the
aging related parameters significantly nor the residuals in general.

So I guess the NTP loop actually was already nearly converged after the
GPS burp and before the arrival of the leap second.


Regards,
Achim.
-- 
+<[Q+ Matrix-12 WAVE#46+305 Neuron microQkb Andromeda XTk Blofeld]>+

Wavetables for the Waldorf Blofeld:
http://Synth.Stromeko.net/Downloads.html#BlofeldUserWavetables