[ntp:hackers] A stop-gap authenticated time service
magnus at rubidium.dyndns.org
Sun Nov 15 10:57:19 UTC 2015
On 11/15/2015 10:35 AM, Poul-Henning Kamp wrote:
> In message <56484F6A.1000807 at rubidium.dyndns.org>, Magnus Danielson writes:
>>> The "Allan intercept" is a rule of thumb it has no formal mathematical basis.
>> For NTP it relates back to an article by Judah Levine on the ACTS
>> system, which got lifted into the NTP world without validating the noise
> If I recall correctly, that article didn't leave any space for
> the steering-mechnisms noise and time delay.
No, it didn't.
> The PI steering used in typical PLLs introduce a very non-random
> noise due to the (exponential-ish) tapering off of the P term
> corrections and adds a significant time-delay, both of which must
> be compensated for relative to the Allan Intercept.
> I've found nobody who did the math to get from Allan Intercept to
> optimal PI parameters.
The noise-shaping of a PLL into Allan deviation is ehm, nowhere to be
found. I've looked. It is however fair to assume that high Q will give a
ringing tone, thus giving the ringing response in the ADEV as for a sine
modulation. Whatever you do, you want a low Q / high damping factor for
best phase-noise or ADEV performance. The Allan intercept is really the
cross-over point that just as in phase-noise matching of slopes achieves
the lowest noise as you high-pass filter one noise type (of your locked
oscillator) and lowpass filter the other (your reference oscillator +
transfer noise) with your PLL. See for instance Ulrich Rohde. Doing this
in phase noise domain or Allan deviation domain is about the same thing.
You will get a little bias because the noise forms does not have the
same conversion factor as you go from phase noise to Allan deviation,
but in the big picture, it's about the same method.
If I get the time, I might derive the various ADEVs for noise through
PLL. Might be a useful article.
More information about the hackers