[time-nuts] PLL behavior

Jim Lux jimlux at earthlink.net
Tue Sep 18 23:35:58 UTC 2012


On 9/18/12 1:49 PM, Magnus Danielson wrote:
> On 09/18/2012 05:28 PM, Jim Lux wrote:
>> I'm looking for info on behavior of a PLL (with VCXO) when the reference
>> comes and goes periodically. When the reference is gone, the PLL will
>> "flywheel" according to whatever the loop filter does. (we can turn off
>> the "input" to the filter, so we're not trying to track noise)..
>>
>> What I'm particularly interested in is the behavior in the PLL when the
>> reference returns.
>>
>> The overall situation is where we are trying to make a frequency/phase
>> measurement over 10-100 seconds, where the reference has a 50% duty
>> cycle, and is on for a second, off for a second.
>>
>>
>> I can fairly simply model this, or just try it, but I'm looking for some
>> references to an analytical approach.
>
> The leakage of your filter will cause the frequency to have drifted a
> little during the off period, so one way of modelling it would be that
> you would treat it like a frequency step. However, if you think a little
> about it, the drift will most likely not be that great so you would only
> shifted a somewhat in phase, and what you get is a phase step response.
>
> It's really trivial to analyze and it has already been done to great
> extent.
>
> It helps if you realize that a dirac delta has the LaPlace form of I(s)
> = 1, and then that a phase step has the formula I(s) = /|phi / s and
> that a phase ramp/frequency step has the formula I(s) = /|omega / s^2.
> Applying these I(s) to you PLLs H(s) gives you the O(s) for your
> response to these stress-tests. Apply inverse LaPlace transform for
> impulse responces.
>

That is basically what I have now..   I guess the next question that 
leads to is "how big is the phase step", and that depends on what the 
oscillator did (in a statistical sense) during the flywheel time, which 
in turn, I should be able to figure out from the Allan Deviation data.

A lot of classical loop analyses (in terms of the statistics) makes the 
assumption that the phase detector response is linear (that is, that the 
error signal is linearly proportional to phase error), which is 
reasonable for small delta phase.   But in the phase step case, that 
might not be.

I suppose then, it's more like looking at the acquisition behavior analysis.



More information about the time-nuts mailing list