[time-nuts] Tuning a Trimble Thunderbolt
pete at heypete.com
Mon Apr 20 17:24:41 EDT 2015
On Mon, Apr 20, 2015 at 8:48 PM, Charles Steinmetz
<csteinmetz at yandex.com> wrote:
> Pete wrote:
>> when using the default
>> parameters it doesn't meet the holdover specifications of +/- 1 us
>> over 2 hours with a maximum of +/- 15C temperature change: it will
>> drift at least 20 us over 2 hours in holdover.
> Executive summary -- you are expecting way too much, way too soon.
Fair enough. It's new to me, so I wanted to tinker shortly after I got it.
>> 2. Is it typical for an oscillator in holdover to drift in a
>> non-linear way?
>> Can this non-linearity be corrected through the judicious choice of
>> tuning parameters or some other means?
> Probably not, but it should calm down significantly after the unit has been
> on continuously for several months.
Ok. I'll save the data from this measurement and see how things
compare in a few months.
>> That document says the units under test
>> were standard Thunderbolts (not Thunderbolt E's) and were on for three
>> days and had a "training period" of two hours prior to the test.
> I don't even know what "been on for three days and had a two hour training
> period" means. Tbolts are training whenever they are locked and operating
I parsed the statement to mean "the Thunderbolts had been powered on
for three days, so everything is warmed up and in thermal equilibrium,
but we did a factory reset to clear the training data two hours before
the test." I could be mistaken in my interpretation.
On a related note, is it possible to extract any data regarding the
training from the unit? It'd be interesting to see how the filter
coefficients change over time.
Also, are there any published details about how the training works?
The user guide is somewhat lacking when it comes to the nitty-gritty.
>> Thunderbolt has been on for a week and had been locked to the GPS
>> signal for at least 12 hours prior to the test.
> Quartz crystals are creatures of habit. They need to be on continuously for
> months before they settle down to their best behavior. And the Kalman
> filter needs to train long enough to see significant temperature changes and
> long periods of steady (that is, settled-down) drift. Thinking in terms of
> days and hours is orders of magnitude shorter than any realistic expectation
> of good behavior. If a Tbolt has been running locked and normally for
> months, and is turned off for a day and then turned back on, it may settle
> down in a day or three. But if you bought it surplus, it may have been off
> for years, and banged around considerably (or much worse) between the last
> use and you receiving it. In that case, you may well need to leave it alone
> for months before the crystal settles down and the Kalman filter is
Will do. Good advice.
Are the training parameters saved periodically to non-volatile memory,
or are they purely stored in RAM and so will be lost if powered down?
If the latter, does the RAM have any provisions for backup power (e.g.
a coin-cell battery or supercapacitor), even if it'd require some
soldering to wire in such a power source? Similarly, is there any
provision for applying power to the oscillator oven even if the rest
of the Thunderbolt is powered down? Considering that it's meant to
operate in telecom structures with reliable power and a fixed
location, I assume not.
>> 3. Is it normal for there to be "spikes" in the phase and frequency
>> error when the number of satellites being tracked changes?
> Yes. The better your survey is, the smaller they will be. But even with
> the best possible survey, you will see 10nS plus or minus when the
> constellation changes. And the poorer your antenna location is, the more
> often the constellation will change.
Good to know. Thanks for the details.
>> I observe
>> changes of ~100ns and 100-200ppt whenever there's a change in the
>> number of satellites. Can this be smoothed out?
> That's quite a lot. I suspect you do not have an accurate survey, or you
> have a marginal antenna location (poor carrier to noise ratio and/or
> multipath problems), or both (the latter pretty much guarantees the former).
> If you have multipath problems, a good choke-ring antenna can help
Alas, the location for the antenna is suboptimal: in the best location
available to me (an outdoor balcony) I have a clear view of the
southern sky from 150-300 degrees (az) and from horizon to zenith with
only a few low-elevation obstructions. However, this is only
accessible in warm months as it's in my living room and the antenna
cable needs to go out a window. My 10 month old daughter may soon make
the current location a bad idea as she gets more mobile and tries to
eat the Thunderbolt.
The only other option is a window on the north side of the building
with a much more limited view of the sky -- although that location
works reasonably well for my Garmin 18x receiver (which is much more
sensitive) the less-sensitive Thunderbolt would need to go into
holdover several times per day.
The surveyed position is within about 10 meters of the actual location
according to Google Maps and local building information.
Roof access is, unfortunately, not permitted by the apartment management.
> Find a good location for the antenna, find a good, out-of-the-way location
> for the Tbolt, and shield it from drafts (a simple cardboard box is a huge
> improvement). [Check the archives for other solutions. I have posted about
> my experiences with "cast aluminum box"es, "thermal mass," and "thermal
> capacitance," and many others have posted lots and lots of other ideas.]
> Then, let it run for 3 or 4 months without playing with it. Don't change
> any parameters, don't go into manual holdover, don't do ANYthing. THEN see
> how it works. Until then, you're just chasing your tail tracking the
> crystal as it settles down.
Interesting. I'll see if I can dig up a box. With all the baby stuff
around the house I'm sure to find something suitable.
The not playing with it for a few months will be the tough part. :)
Thanks again for the help.
> Best regards,
 Interestingly, the 18x rarely loses lock, even on satellites in
the southern sky. Since it is reporting a position a few tens of
meters away from its actual location (when used outside the apartment
it computes much more accurate locations), I assume it's because it's
tracking the signals reflected off of a conveniently located building
to the north that allows a partial, reflected view of the southern
sky. Less than ideal, yes, but it works well enough for NTP.
More information about the time-nuts