[time-nuts] A Time-Nut's Worst Nightmare
magnus at rubidium.dyndns.org
Sat May 11 07:48:22 EDT 2013
On 11/05/13 09:10, Lachlan Gunn wrote:
> From https://github.com/akafugu/vetinari_clock :
> The firmware has a 'random' sequence of pulses which over 32 seconds
> moves the second hand 32 times. This sequence is long enough to give
> the appearance of randomness without unnecessarily consuming
> processor cycles. I chose 32 seconds as it means the pattern is
> offset from the 60 second rotation of the minutes. The 'random'
> is 128 steps (and is therefore checked 4 times a second). This means
> clock moves between 0-4 times within each actual second. By moving
> more or less in each of the 32 seconds the clock looks random,
> the sequence always moves the clock exactly 32 seconds in the 128
Not quite what I guessed, but similar enough.
> So there is no long-term drift, it seems, beyond that of the crystal. I've
> pulled the pseudo-random tick sequence out of the source code and found the
> Allan variance, shown here:
Not completely a surprise for a phase-modulation, as phase-modulation
have that slope in the ADEV.
As a systematic modulation, MTIE analysis would be of more help, as you
have a looped sequence and analyze the pattern you would see how the
maximum time error will be a flat line beyond 32 s, making it easy to
believe that on long-term there is no increase in time.
You could also attempt the TDEV, which will just give a straight line
with wrinkles just as the ADEV.
More information about the time-nuts