Hi Charles,
I agree with your statement regarding the determination of the optimum
time constant, but, as Bob Camp mentioned, temperature change has a
significant impact on setting the value. My 'lab' is a
non-airconditioned bedroom. My Tbolt doesn't have any active
temperature control. If I set the time constant to the point that Lady
Heather thinks is optimum, I see large swings in PPS offset when I open
the window and the temperature changes by a few degrees C. If I leave
the time constant at the default of 100 seconds, the swimgs are
drastically reduced. Active temperature control is on my 'round tuit' list.
I don't think you're correct about the Miller GPSDO containing a Shera
controller. The Miller design is famous for having no processor - just
simple analog hardware. It even says that on the page you referenced.
That's one reason why his settings aren't optimum. It's impractical to
get long time constants with the simple analog circuitry that he uses.
Despite that, his design still has a 'time nuts' level of performance.
Ed
On 8/22/2014 6:39 PM, Charles Steinmetz wrote:
Bob wrote:
The GPSTM is not as tweak friendly (no filter changes allowed) as
some of the other GPSDO's.
And that is a major problem. The correct filter settings for a Rb
local oscillator are very different from the settings for an OCXO,
which in turn are different from the correct settings for a TCXO.
As a general matter, almost all of the DIY GPSDO designs I have seen
use PLL loop filter settings that are not optimal. Many are not even
close (several orders of magnitude, or more, from optimal).
Generally speaking, the PLL loop filter cutoff should be set
approximately where the GPS xDEV curve intersects the local oscillator
xDEV curve. That puts the better device (GPS or local oscillator) "in
charge" of the composite xDEV at all tau -- the local oscillator at
short and medium tau, and the GPS at long tau. Optimal crossover tau
will generally be in the range of seconds for a TCXO, hundreds of
seconds for an OCXO, and hours to tens of hours for a Rb.
Sometimes, there are good reasons to depart from this general rule.
In particular, if a speedy recovery from holdover is required, then
one might choose a PLL filter cutoff tau that is lower than optimal.
The default crossover tau for the Trimble Thunderbolt is chosen quite
low, presumably for this reason. See, for example,
<http://www.leapsecond.com/pages/gpsdo/>, where the GPS contributes
significantly to the ADEV way down at tau = 1 second, where the local
oscillator is clearly much better than GPS and continues to be for
more than two decades. The Miller DIY GPSDO on that page is crossed
over about 3 decades lower than optimal. (The Miller GPSDO uses a
Shera DIY controller; I presume the Shera has the same crossover tau.).
Compare this to the HP z3801A and Jackson Labs Fury on the same page.
The HP crosses over about 2 decades higher than the Thunderbolt and
Miller GPSDOs, but that is still premature by about two decades given
the very high quality of the OCXO in that particular unit. The Fury
crossover is set well, but the overall ADEV is let down by the low
stability of the OCXO in that particular unit. (Note that the
crossover in commercially produced GPSDOs must accommodate the range
in production ADEVs of the local oscillators used, and are likely set
a bit lower than optimal for most of the actual OCXOs on this
account.) If the filter parameters are adjustable -- as they are in
the case of the Thunderbolt -- then a time nut can tune his or her
individual sample to get the best possible performance that particular
oscillator can deliver.
As I have mentioned before, rather than just setting the time constant
low to speed up holdover recovery, a better solution is to implement a
switchable PLL loop filter. A GPSDO designed this way uses a suitably
long time constant for normal locked operation to minimize xDEV at all
frequencies, and a faster time constant for turn-on warmup and
holdover recovery. It is rumored that the z3801 is designed this way.
Best regards,
Charles
_______________________________________________
time-nuts mailing list -- time-nuts@febo.com
To unsubscribe, go to https://www.febo.com/cgi-bin/mailman/listinfo/time-nuts
and follow the instructions there.