On 08/24/2012 11:35 AM, Azelio Boriani wrote:
Magnus warned us about the use of very long time constant to act on
disciplining. In my opinion it is better to use quieter data coming from a
relatively fast sampling Kalman filter and correct as frequently as
possible then train the filter with long time constants.
It really, really depends. I usually warn about using _too_ long time
constants. As John has correctly pointed out, the PRS-10 has a good
crystal oscillator in it, allowing for longer time-constants to be used.
There are many benefits of using an SC-cut crystal oven in a rubidium.
I also agree with John about the hump, its bound to be there due to the
PLL action. There are two sources of humpiness at the cross-over. The
first is that at the cross-over you transition from the low-pass
filtered reference noise and the high-pass filtered oscillator and loop
noise. Since the noises is uncorrelated, their powers will add. The
cross-over filter does not suppress one noise before the other kicks in
to sufficient degree of suppressing the additive effect. Another aspect
is that the PLL Q-value creates a gain at the cross-over point, and
using too low Q values acts like an equalizer to bring noise up.
This is to be expected and comes out of standard control system math.
Cheers,
Magnus
_______________________________________________
time-nuts mailing list -- [email protected]
To unsubscribe, go to https://www.febo.com/cgi-bin/mailman/listinfo/time-nuts
and follow the instructions there.
