[time-nuts] Interesting paper: Don't GPSD' your Rb...

Magnus Danielson magnus at rubidium.dyndns.org
Mon May 7 16:52:14 UTC 2012


Azelio,

On 05/07/2012 09:56 AM, Azelio Boriani wrote:
> Magnus,
>
>> If you flip back and forth, then it makes sense because your phase
> deviations>will be less.
>
> Can you further explain this? Thanks.

Certainly!

Consider that you flip back and forth between two levels, let's just say 
50% "high" and 50% "low", the rate of flips considering the 
time-constant of the adjustment will matter, since if you let the time 
be so long that the alignment "ring out" you will get maxium frequency 
and phase deviation, where as if the rate of flipping is higher, then it 
drift just a little towards "high" when it gets a "low" and drift 
towards that... which gives you much smaller phase and frequency 
deviations... and you stay very close to the 50% level between "high" 
and "low". Essentially, the time-constant of the low-pass filtering will 
for a higher rate fairly well dampen the changes where as slow changes 
is in the pass-band with almost no low-pass filtering effects.

It is thus nothing stranger than a low-pass filter and a square-wave of 
different frequencies.

Also, if you update to slowly, you will build up larger errors before 
you have to steer back since you measured an error. Too low comparator 
rate in a PLL forces heavy filtering just to lower that comparator 
sub-frequency. Using high enough rate, the filtering process will be 
more continuous rather than very obviously sampled oriented.

I've been bitten by this before. I've seen what too low comparator 
frequency does to create instability and modulations.

Cheers,
Magnus



More information about the time-nuts mailing list