[time-nuts] A Time-Nut's Worst Nightmare

[Magnus Danielson]

On 11/05/13 22:04, Poul-Henning Kamp wrote:
> In message<518E2C65.80204 at rubidium.dyndns.org>, Magnus Danielson writes:
>
>> The end result will be a clock which in long term is showing the right
>> time, but have short term variations. Since it is a lag scheme, it will
>> also on average be behind.
>
> Wouldn't it be easier to simply implement a random walk with a square-law
> sort of gravity anchored at the right time ?
>
> That way it would sometimes be ahead, sometimes behind, but keep
> correct time on average.
>
> And you could make the movement truly random and non-periodic.
>
> I'm sure that's how the watchmakers guild did it.
>

Having a leaky integrator, or in essence a low-pass filter with 
sufficiently low bandwidth would achieve that. Not to complex.

y = y + (x-y)*alpha

The original source isn't very complex anyway. There is a second 
fine-grained modulation as well, having a period of 15 steps every 
quarter of a second. That extends the complete loop to be 32*15 = 480 s 
or 8 min.

Never the less, I think it is an interesting exercise in modulation 
analysis in order to figure out the mechanism from the variations alone. 
Random or systematic is indeed interesting to figure out.

Oh, as to why doing it, well, it's a mockery of normal precise time, and 
fits the "why not?" purpose.

Cheers,
Magnus