[time-nuts] DMTD Question1
Magnus Danielson
magnus at rubidium.dyndns.org
Wed Jan 27 01:33:08 UTC 2010
John Miles wrote:
>>> So why even talk about TC phase shifts at the ps / deg level? Who
>>> cares and who needs it?
>
> A picosecond is 1000 femtoseconds. When you're spending $100/ea. on parts
> rated for 60-70 fs jitter, a picosecond starts to look like a long time
> indeed.
>
>>> Does not sound like you are doing the Nuts any service to address
>>> things that are orders of Magnitude below usefulness.
>
> When designing something like this, it's useful to understand where, and
> why, your efforts fall short of the state of the art. That way you can be
> sure that the compromises you're making are the right ones. The idea is not
> to leave any obvious low-hanging fruit for optimization. You want to get
> the most out of the money you're spending, whether it's $200 or $20,000,
> right?
>
> I do think there's a lot of room between "state of the art" timing
> performance and conventional TIC-grade performance where corners can be cut
> and costs can be saved. It's helpful (and more interesting) if you can make
> those calls rationally, instead of just adopting circuits from papers and
> hoping for the best.
Let's recall that the type of measurement varies. It may be that focus
is on phase noise and short-tau instability. The concerns for long-tau
instability does not apply to the same degree. The time-span of the
measurement is an important aspect. Averaging techniques assumes that
propeties is relatively stable over time, such that short-time noise can
be filtered. Calibrations may go out of tune. Being able to crank out
numbers of reasnoble reliability calls for stability of the measurement
rig, beyond short-term performance. Understanding the hidden errors is
important. Drift and environmental dependencies is among the issues to
consider.
You can have a perfectly good schematic, but still not be able to get
the performance. Choice of components, mounting, etc. all add up.
I find that I personal preferences towards various methods at various
times, but I also find that I need to reevaluate things over and over as
I get more influences. None of them is the best. Some of them has
inheret drawbacks, so I need to learn how they work and what
difficulties has been addressed. How is each defect handled? What does
that handling imply?
It's non-trivial stuff, so there is only one mode to handle it... learn
more all the time.
Cheers,
Magnus
More information about the time-nuts
mailing list