[time-nuts] Use ADEV or MDEV to set time interval of GPSDO?

[Erik Kaashoek]

For my home build GPSDO I'm searching for the best interval for updating 
the VCXO based on measuring the PPS over that same interval.
The GPSDO measures its internal PPS with 5ns accuracy using 10MHz from 
the internal (not so good) VCXO as reference resulting in a PPS accuracy 
of 5e-9 per second measured.
To understand the accuracy of the PPS, the drift of the VCXO and the 
performance versus interval I did the following measurements

  * PPS with U6200A/AR60A (dark blue line)
  * 10MHZ free running output of the VCXO in open loop with U6200A/AR60A
    (green line)
  * PPS as measured internally in the GPSDO (red line)
  * 10MHz GPSDO with 100s interval with U6200A/AR60A (pink line)
  * 10MHz GPSDO with 20s interval with U6200A/AR60A (light blue line)

In the document on his GPSDO, Lars states that for determining the best 
interval for a GPSDO one should use the MDEV plot instead of an ADEV.
So I made both (see attached)
On the ADEV plot the unregulated VCXO trace crosses the PPS traces at 
about 100s and when using either 100s or 20s as interval both traces 
converge with the downward trend of the PPS itself
On the MDEV plot the VCXO trace crosses the PPS trace at about 20s but 
both regulated VCXO's overshoot the PPS trace a lot.
The 20s interval seems to introduce incorrect adjustments as the blue 
line moves up around 20s and gets much higher than the 100s interval 
pink line. Also the frequency difference plot (FREQDEV.JPG )shows the 
20s interval trace being corrected incorrectly (too big jumps)
This all could be caused by the limited resolution (5ns) of the PPS 
measurement but I do not understand how this limited resolution should 
be reflected in shifting of the PPS line. Reduction in resolution seems 
to give a too optimistic measurement of the ADEV of the PPS where the 
MDEV seems to converge.

So my questions are:
Are my measurements correctly done?
Should I use the the ADEV plot to determine the optimal interval of a 
GPSDO?
Or is the bump at 20s due to the limited PPS resolution and should the 
MDEV be used with an adjusted PPS trace for the limited accuracy? If the 
latter, how to adjust?
Erik.
-------------- next part --------------
A non-text attachment was scrubbed...
Name: ADEV.JPG
Type: image/jpeg
Size: 135463 bytes
Desc: not available
URL: <http://febo.com/pipermail/time-nuts_lists.febo.com/attachments/20211208/c0666809/attachment.jpe>
-------------- next part --------------
A non-text attachment was scrubbed...
Name: MDEV.JPG
Type: image/jpeg
Size: 134903 bytes
Desc: not available
URL: <http://febo.com/pipermail/time-nuts_lists.febo.com/attachments/20211208/c0666809/attachment-0001.jpe>
-------------- next part --------------
A non-text attachment was scrubbed...
Name: FREQDEV.JPG
Type: image/jpeg
Size: 144566 bytes
Desc: not available
URL: <http://febo.com/pipermail/time-nuts_lists.febo.com/attachments/20211208/c0666809/attachment-0002.jpe>