[time-nuts] test mailing from a new member

Magnus Danielson magnus at rubidium.dyndns.org
Wed Jun 17 01:28:23 UTC 2009

David Bengtson skrev:
> On Tue, Jun 16, 2009 at 6:07 PM, Magnus
> Danielson<magnus at rubidium.dyndns.org> wrote:
>> Hal Murray skrev:
>>> david.bengtson at gmail.com said:
>>>>  I've recently gotten an opportunity to work on a GPSDO at work, and
>>>> so I now have a real need to pick up some more information on the
>>>> design tradeoffs and approaches in a GPSDO. I've plowed through a pile
>>>> of old emails, but it's difficult to get a good overview from email
>>>> threads.
>>>> I would appreciate links to discussions on the tradeoffs inherent in a
>>>> GPSDO. Sampling clock rate, filter bandwidths, jitter requirements
>>> Be sure you know about hanging bridges.  Start with the
>>>  "Timing for VLBI" slides at http://gpstime.com/
>> The key issue is, what type of specs is needed? What may it cost?
>> What kind of receiver may be considered? What's the intended application?
> Specs.. Hmmmm. Good question. The application is for a frequency
> reference, so frequency stability vs. PPS accuracy is important. Cost
> is perhaps less important than size. The OCXO under control is a
> small, surface mount part with +/- 0.5 PPM over temp, but flexibility
> to use higher accuracy parts would be good. As this is for an internal
> application, I'm not sure how much else I can disclose.

You should consider using a board having a 10 MHz input. Otherwise the 
normal route is to use the PPS pulse out of a board (adjusted by any 
form of offset value usually available in the bitstream) and make a TI 
compare to a divided down variant of the OCXO. The standard approach 
used by many is to use a PI-loop where the bandwidth and Q-value is set 
with thought. A more complex approach is to go for a Kalman filter, 
which has the benefit of being able to more dynamically adapt and tune 
it. A Kalman filter for a simple setup like this isn't particularly hard 
to design, even if the math may seem a bit complex at first.

Don't use to few bits in the DA.

There are a number of simple projects around. Check if their achieved 
level of performance is matching your needs. In that case you know more 
or less what you need to stretch for.

This is a topic which can go over the deep end if you let us... so 
asking for the required specs is really, what do you need? What can you 
settle with? It gives an indication of how deep you need to go.

One issue to keep in mind is what kind of hold-over performance do you 
need, if any. That is... how much may it deviate after how long a time 
without propper GPS signal.


More information about the time-nuts mailing list