[time-nuts] oscillators

Magnus Danielson magnus at rubidium.dyndns.org
Sat Sep 1 21:01:21 UTC 2012


On 09/01/2012 05:53 PM, Jim Lux wrote:
> On 9/1/12 8:32 AM, Bob Camp wrote:
>> Hi
>>
>> Observing a curve and being able to compensate it are often two
>> different things. Hysteresis is one very obvious example. Another is
>> simple sensor lag. A some what less obvious one is that the
>> temperature performance is also influenced by the rate of change in
>> temperature.
>>
>> Here's another thing to consider:
>>
>> If your crystal is running 3 ppm / C, and you are after 3.0 x 10^-11
>> stability at one second - You will need to either have a rate of
>> change at ~ 1x10^-5 C/sec (0.6 mC / min) or you will need to
>> compensate for some pretty small changes. That of course makes a bunch
>> of assumptions ….
>>
>>
>
> In this application, the requirement for frequency accuracy has to do
> with initial acquisition.. that is, you want the signal (or receiver
> tuning) to be within some few hundred Hz of where it's expected to be
> (because the receiver is narrow band).
>
>
> The ground station typically has a Doppler predict based on orbit
> knowledge, that predict has some uncertainty. Added to the radio
> frequency uncertainty. (SNUG - Space Network User Guide, has more info)
>
> Once you've acquired, the receiver and ground station will track (i.e.
> the ground station puts in the estimated Doppler, so all you're really
> tracking is the variation in the local oscillator). (for a LEO satellite
> at 2.3 GHz, the 7km/s orbital velocity already puts tens of kHz
> variation on it)
>
> (and this completely neglects that a modern radio could use something
> like an FFT for acquisition)
>
> Temperature changes are pretty slow.. I'm seeing 5-10 degree cyclical
> variation over 90-100 minutes. Actually, the bigger change is during the
> warm up transient, going from off and cold to on and warm over 10
> minutes or so.
>
> In other applications, where you're not going in and out of the sun
> every revolution (i.e. deep space, rather than LEO) and you were
> interested in Allan deviation type measurements for gravity science
> (where we're looking for 1E-13 over 100 sec sort of performance), what
> we'd probably do is warm up early.. Turn it on, compensate based on the
> measured temperature, and then hold the compensation fixed during the
> measurement, letting the ground worry about the apparent frequency
> change due to Doppler. We'd have a high quality narrow band signal, just
> at an unknown (but reasonably stable) frequency. What the science team
> is usually interested in is small relative changes in phase &
> amplitude(occultations) or in small changes in frequency (Doppler, for
> gravity science).
>
> (we regularly measure velocity to cm/sec precision for outer planet
> orbiters like Cassini, Juno, etc.)

If you can make reasonable predictions of the heating and cooling 
profile, you could use that and hopefully gain a decade or so, and any 
slew in detectors and would mostly affect the remaining error.

You could also make use of uplink carrier and GPS to improve the model 
state of your crystal. That way the predicted and feed forward values 
could be kept fairly well adapted. Naturally, you would want a fall-back 
scenario to back out for wider offset search if you failed to maintain 
the model. You don't want to loose your bird due to temporary system 
failure. Another alternative would be to allow for the downlink 
frequency to be a hint for the uplink where the frequency is such that 
you can "pull in" the bird if need to.

Cheers,
Magnus



More information about the time-nuts mailing list