richard@karlquist.com said:

How many would you need?  Is 3 enough?

How well could you do with several low(er) cost oscillators relative to one
good but expensive one?  It might be an interesting experiment in a nutty
sort of way.

--
These are my opinions.  I hate spam.

On 4/11/2014 11:04 AM, Hal Murray wrote:

My guess would be 3 would be a minimum, so you
could have a majority vote.  Len Cutler's group
actually built an experimental ensemble of 9 or
10, but it didn't seem to come to fruition.
For this to make any sense, you would need to be
able to cherry pick 9 or 10 really good oscillators.
However, there was no way to get the production
line to sign on to this.

David Allan had
this interesting concept to the effect that if
you had a sufficient number of wristwatches
(maybe 1000) and you averaged them together
you could somehow get a quality clock, or at
least 31.6 times better.  Kind of like the
notion of 1000 monkeys with 1000 typewriters...

Rick

Interesting idea. It might be an interesting experiment to couple a large number of inexpensive xtals to see how it impacts effects such as sudden changes in a single xtal.

With sufficient monitoring of each one, you could even tune the coupling to amplify/attenuate the results of the 'good' and 'bad' ones over some interval.

Of course, what effect this has on things like phase noise, drift, and so on is a whole different matter.

Bob

Look at what NTP does to select "good" clocks when it has many to choose
from.  It does not simply average them.

It looks at the noise in each one and then sees which clocks have
overlapping error bars.  It assumes that all good clocks have the same time
within limits of their precision.  Then from the good clocks there is a
second level weeding out process then finally it does a weighted average of
the remainders where I think those with less jitter get more weight.

It would not be impossible to do this with 10MHz oscillators.  Certainly a
simple average is not a good idea as a broken unit can pull the entire
average way down.  I think you'd have to check reasonableness first and
eliminate outliers  I think today you might simply digitize the signals
and figure out which were best using software.

In short the output is "ensemble time" (not "average time") but there is a
careful selection of who is allowed to be  member of the ensemble.

I used a joke last week to explain to a class why we don't use averages,
with no other qualifications.  The joke is "Bill Gates walks into a bar....
What's the average net worth of everyone in the bar?  Maybe $250 million."

My point was that it is hard to describe a population that is not Gaussian
distributed.  "Stuck" and jumping crystals are not Gaussian.  You'd have to
detect the misbehaving devices.

David Allan had

--

Chris Albertson
Redondo Beach, California

On 11/04/14 21:38, Chris Albertson wrote:

NTP uses the ensamble clock style that Dave Allan developed for the NBS
AT time-scale and originally programmed on a PDP-7. Applying this type
of phase-comparison, estimate stability, weighing and updating ensamble
stability should indeed be possible to do. You need three or more
clocks, but one of these can be the GPS when you have it.

Jim Gray pointed out that it is important to watch your data. At
NBS/NIST they started to see some 1/f⁴ noise on one of their standards.
They could not figure it out. Turned out that the cleaning-lady was
pushing the standard over the floor once a week in order to clean under
it. This systematic "noise" where not in their standard model, but they
learned.

A frequency jump on the crystal oscillator in a control-loop will be
tracked in eventually, so it will look more like a phase-spike than a
frequency jump. Atomic clock FLLs will however not track in the full
phase difference, at least not guaranteed to do.

Indeed.

Cheers,
Magnus