[time-nuts] Correcting measured PPS values with receiver sawtooth correction values

[Poul-Henning Kamp]

--------
In message <BN7PR07MB5297CAA7A73C8107C94B6D6FCE600 at BN7PR07MB5297.namprd07.prod.
outlook.com>, Mark Sims writes:

>For all the receivers that I have tested there is only one combination
>that jumps out as being the correct one.   However, for the new
>Ublox F9P, there are two combinations that produce virtually identical
>measurements and statistics (add current sawtooth or subtract
>previous sawtooth).

That is a pretty natural pattern, which has been evident to varying
degrees, from the very first Oncore-UT models.

>Any ideas on how to determine which combination is the "correct" one?

Absent input from the manufacturer, you will have to measure it. 

What you are looking at is off-by-one events caused by mispredictions
of the clock-rate which straddle the "use this or next flank" threshold
of the PPS signal generator.

If you temperature cycle the GPS, these events will happen more often.

After (mis-correction) these events show up as spikes of almost one
clock-period outside the normal band, at the "towers" of the "hanging
brides".

In other words:  Standard deviations and variances is not the right
tool, you are looking for outliers at a quite distinct distance
(+/-) from the mean.

-- 
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.