[time-nuts] Temperature stability for Thunderbolt: results
Charles P. Steinmetz
charles_steinmetz at lavabit.com
Thu Jan 27 19:18:19 UTC 2011
>If all that's true, what you are watching on the EFC is the TBolt
>correct out the OCXO TC.
Agreed (well, superimposed upon its correction of all other drift mechanisms).
>By this (quite possibly imperfect) measure: When the EFC goes down
>3:1, you have improved the thermals by 3X. If you started at 1 C
>swings, the OCXO is now seeing 1/3 C swings.
Agreed (re: the EFC component that correlates to changes in Tbolt
>Ultimately what you would like is a setup that never needs to change
>the EFC (much), so you could run a nice long loop and still have the
>PPS stay lined up. That *should* give you a significant improvement
>in short term out of the TBolt.
Alternatively, if you lengthen the thermal time constant from ambient
to Tbolt so that it is substantially longer than the time constant of
the loop chosen for best drift performance, AND the Tbolt control
loop has sufficient gain, the Tbolt control loop will take care of
both. There is a limit to how long you want the Tbolt TC to be --
generally something in the neighborhood of 1000 seconds -50/+100% --
based on the tau at which the particular XO in your Tbolt transitions
to its random-walk behavior. Accordingly, you have a good idea what
sort of TC you need from ambient to Tbolt so that it is substantially longer.
I set mine up that way (as described in a previous post) before LH
had the temperature control feature, and it appears that the Tbolt
control loop has plenty of gain to suppress drift from the sort of
ambient temperature fluctuations one finds in a home shop environment
below the random fluctuations from the on-board XO. In this
connection, note that the oven will allow the crystal to vary in
temperature by a certain amount even if the ambient temperature
outside the oven is perfectly constant, and we have no control over
that loop. Also note that there will be temperature gradients within
the Tbolt between its internal temperature sensor and the outside of
the oven, and that these will vary over time. Thus, beyond a certain
point, holding the temperature at the Tbolt's internal temperature
sensor constant will not further improve the temperature regulation
of the crystal itself.
When this is done, the EFC will still show a component that tracks
the ambient temperature reported by the internal temperature sensor
-- but this will be an extremely slow variation, and indicates that
the control loop is successfully handling ambient temperature changes.
My point is not that one shouldn't try to hold the Tbolt ambient
temperature (as reported by its internal temperature sensor)
constant, just that it may not be a greater benefit in practice than
simply slowing the ambient-to-Tbolt time constant enough to let the
Tbolt's control loop handle variations in ambient temperature -- at
least for the modest swings likely to be encountered in a home shop
Ideally, one would like accurate temperature information from the
crystal itself, and would use external controls to hold the crystal
temperature constant directly. This could be tricky if the existing
oven loop were left intact, fully enclosed within our external loop
-- likely, one would need to redesign the entire loop structure.
>Certainly should add that without all the work done on Lady Heather,
>you wouldn't have a chance of doing any of this. Can't eyeball a
>plot you don't have...
Amen to that.
More information about the time-nuts