[time-nuts] Achievable temperature stability for Thunderbolt environment?

Charles P. Steinmetz charles_steinmetz at lavabit.com
Sat Jan 15 19:59:00 UTC 2011

Achim wrote:

>I am just wondering, if I should rather worry more about high 
>temperature gradients rather than excursions from a mean value, as 
>slow variation can be compensated by the control loop while for 
>quick changes, the loop is just too slow :)

This has been discussed quite a bit on the list, so you will find 
information in the archives.  The current version of Lady Heather 
facilitates adding overall temperature regulation, if you so 
desire.  Many of us have found that excellent performance can be 
achieved simply by ensuring that changes to the Tbolt's enclosure 
teperature are filtered by a longish time constant.

First, note that the reported temperature is measured at the 
periphery of the Tbolt pc card, not inside the crystal oven.  Second, 
the GPS disciplines the oscillator in a high-gain loop through a time 
constant and filter Q set in software (which can be determined by the 
user), typically from 200-2000 seconds.  For best performance, this 
time constant will be set so that it allows the GPS to control the 
oscillator at longer averaging periods (tau), where the oscillator's 
drift reaches its "random walk" phase, but leaves the crystal 
substantially to its own devices at shorter tau.

Assuming sufficiently high gain in the disciplining loop, one only 
needs to make the thermal time constant between the Tbolt pc card and 
ambient temperature significantly longer than the control loop for 
the control loop to handle temperature changes as well as random 
oscillator drift.  This can be accomplished by housing the Tbolt in a 
somewhat larger enclosure.  Note that insulating it too well from 
ambient temperature (resistive loss) is not desirable because it 
allows too much of a temperature rise and thereby deprives the oven 
control loop of the temperature sinking it needs to operate properly, 
as well as subjecting the components outside the oven to 
unnecessarily high temperatures that could tend to create reliability 
problems.  In my tests, reported temperatures up to 45 degrees C work 
well.  What you are after is thermal reactance to slow down the 
change experienced by the Tbolt, not so much thermal resistance.

I settled on a cast aluminum project box that provides about 1" (25.4 
mm) of clearance around the Tbolt and has 1" rubber feet to minimize 
conductive heat transfer to the environment.  I mount the Tbolt on 
standoffs (I used nylon standoffs to minimize thermal conduction, but 
I suspect metal standoffs would be fine).  One can then ask whether 
the convection currents within this outer enclosure work against 
you.  I have tried (i) adding a small fan to circulate the air within 
the enclosure, and (ii) adding foam with large open cells to disrupt 
the convection.  I could detect no difference other than another 0.75 
degree C rise with the foam, so I returned to a simple air space.

In my case, the outer enclosure is mounted inside a rack-mount 
enclosure that also contains buffers, dividers, and signal 
conditioners to provide the various reference signals I need.  There 
is a minimal temperature rise inside this box -- it is cooled with a 
very slow fan.  Overall, I get about a 1 degree C swing over 24 hours 
as reported by the Tbolt temperature sensor, superimposed on a 2 or 3 
degree C semi-annual swing.  There is very little change in the 
temperature data at time scales below several hours.

Best regards,


More information about the time-nuts mailing list