[time-nuts] Divide by five

[Chris Albertson]

On Sun, Nov 9, 2014 at 10:11 AM, Bob Camp <kb8tq at n1k.org> wrote:

>
> That may (or may not) give you the best ADEV on the output. My guess is
> that the filtering algorithm will need to be a bit more complex. NTP’s aim
> is mainly to throw out bad clocks and pick one as best. We would more
> likely want to combine the outputs and use all of the good clocks we have.
> The idea is to improve on the ADEV of the *best* source you have available.
>

No, not "pick one as best".  NTP does not pick the best clock.   NTP finds
the subset of clocks that track each other.  Clocks are allowed to enter
and leave this set.  NTP then computes a weighted average of the clocks in
the set.   Pretty much what you said you wanted.

These two algorithms are used.  Conceptually that are run one after the
other but actually they run together in a loop forever
1) clock selection <http://www.eecis.udel.edu/~mills/ntp/html/select.html>
2) clock cluster <http://www.eecis.udel.edu/~mills/ntp/html/cluster.html>

You would not want to use this exact algorithm but something like it.  I
think the basic idea is simple:  Reference clocks should tend to cluster
around correct behavior.  In other words our set of reference clocks is not
biased.  I don't think there is any way to detect bias unless you add more
reference clocks.

I would very much avoid an FPGA based CPU. You likely will need to add both
> flash and ram to the FPGA. Once you are done, you have a $30 gizmo that
> replaces a $1 chip. You also have a tool chain for your code that is far
> from “low barrier to entry”. It’s a great solution for something like video
> processing that needs TONS of bandwidth. We are very much on the other end
> of that stick.
>

I agree.  Unless the whole this is FPGA based and you have enough free
gates.  Otherwise it is not good to use something so complex to simulate a
$2 part.

>
> --

Chris Albertson
Redondo Beach, California