[time-nuts] PCB layout question for GPSDO

[Matthias Welwarsky]

On Samstag, 29. Februar 2020 16:48:30 CET John Ackermann N8UR wrote:
> On 2/29/20 7:40 AM, Matthias Welwarsk wrote:
> > For example, there's the TDC7200 itself. When I tested my design first,
> > the
> > measurements had a distinct relation to temperature, in the order of maybe
> > 300ps/°C, but there's other stuff as well, for example, the 74ALVC74 in
> > yours, the whole analog output section etc.
> 
> The TDC7200 should compensate for temperature changes, though in
> practice I'm not sure how quickly.
> 
> Because the ring oscillator in the chip that provides the
> picosecond-level clock is free running and violently temperature
> sensitive, it needs to be calibrated and the TDC7200 does that
> automatically after each measurement.
> 
> After the STOP pulse is received, the chip measures the time interval of
> several cycles of the external 10 MHz clock, and works that result
> backward to determine the actual frequency of the ring oscillator.  The
> TI calculation uses that measured frequency as an input and thus should
> take into account any tempco.
> 
> Now, I've tried sticking my finger on the chip and that causes a wild
> tempco for at least some seconds, so there is some inertia in the
> correction, but the calibration process should take care of more typical
> slowly changing temperatures.  Isolating the chip from moving air should
> be helpful.

There's two interesting implications here, one general, one in the context of 
a GPSDO:
- If the temperature gradient is sufficiently steep, the calibration will be 
outdated when the next measurement starts.
- Mine as well as Tobias GPSDO design use the disciplined LO as the reference 
clock for the TDC7200. The control loop perturbs the LO as a consequence of 
the measurement. If the response is very fast, this perturbation might fall 
into the calibration window. This window is 40 cycles of the reference clock 
at max. So, a total of 4000ns worst case. For my own software I know that the 
response will not happen faster than 1ms, because everything is driven from a 
mainloop with 1ms ticker granularity so it will not be a problem.

BR,
Matthias

> 
> John
> 
> 
> _______________________________________________
> time-nuts mailing list -- time-nuts at lists.febo.com
> To unsubscribe, go to
> http://lists.febo.com/mailman/listinfo/time-nuts_lists.febo.com and follow
> the instructions there.