[time-nuts] Best test setup?

[John Miles]

> I have severa crystal, rubidium, and GPSDO oscillators that I
> would like to
> characterize.  I have a 5370B and cesium that has 1, 5, and 10
> MHz output.
> The question is what is the best configuration of internal/external
> oscillator, start input, and stop input of the 5370B (and do I use an
> external 1pps trigger)?

I use my 5370 in +TI mode, with 1-pps at the START channel and 10 MHz at the
STOP channel.  This yields one reading per second between 0-100 ns.  (Or,
rather, about 6-106 ns on my particular counter, since the START pulse has
to come before the STOP pulse from the counter's point of view.  The
distinction isn't important.)

If you want to watch the oscillator during warmup, when it may be drifting
too rapidly for meaningful observation at 100ns/sec intervals, you can feed
1-pps inputs to both START and STOP.  This requires two 1-pps dividers.

If you have a HP 5061 cesium, be _very_ careful using its 1-pps output.  The
amplitude is something like 10V pk-pk and can easily fry a counter input. :(

> Then there is the issue of whether to use the cesium as the
> reference in all
> cases.  Seems the GPSDO would have better long term stability for
> very long
> tau.  Any insight appreciated.

It comes down to the question of what's worse, your cesium beam tube's SNR
or your GPS receiver's SNR?  You can think of your GPS receiver as a cesium
standard with a really awesome beam tube but a noisy radio link in the
control loop.

My 5061A/B is a bit more stable than my (modified/optimized) Thunderbolt
beyond a few hundred seconds but it's a close contest, and a GPS clock would
most likely be a better reference at timescales beyond a few hours.  OTOH if
you have a modern cesium standard it may well outperform the best local GPS
clocks at all timescales.

It seems that the best overall medium-term reference, on a cost/performance
basis, is a high-performance rubidium like the 5065A.

A telecom-grade LPRO-101 Rb module was tested in the attached plot against
both my 5061 and Thunderbolt, without any temperature stabilization.  The
tests ran concurrently using two different counters, one 5370B and one
homebrew.  The 5370's residual jitter (red) dominated the Cs-based
measurement (purple) below about 20 seconds.  There's good agreement beyond
that point, until the Thunderbolt's loop characteristic becomes worse than
either the Cs or Rb standard beyond 400 seconds (blue).  At 1000 seconds the
Rb becomes less stable than the Cs, and by 4000 seconds the Rb is unstable
enough, once again, to be characterized by both the TBolt and Cs references.

Moral of the story is that unless you have a really good reference at all
timescales of interest (which I don't), you need to be prepared to use
multiple references at different timescales.  Even so, you may not get all
the information you're after.  This graph looks nifty, but it conveys little
or no information about the LPRO-101's true performance between t=20s and
t=1000s!

-- john, KE5FX
-------------- next part --------------
A non-text attachment was scrubbed...
Name: cs_vs_tbolt.gif
Type: image/gif
Size: 26091 bytes
Desc: not available
URL: <http://febo.com/pipermail/time-nuts_lists.febo.com/attachments/20091229/a91c0c6f/attachment.gif>