Hi

There’s a seller on eBay ( besttestus ) who comes up with batches of 10811’s from who
knows where. Some are still attached to option boards for instruments. They just sold out
of the latest batch of about 200 or so. I’d guess they will get some back in eventually.
Asking price is typically right around $50 or so (with free shipping). They take offers so
you can indeed get them to sell you 10 at a whack for around $42 each.

Based on testing more than a few of them, there will be a couple that have significant
issues ( < 5%). There will be some that are in spec, but not by as much as when they
left the factory ( maybe another 5 %). The rest split into “pretty good” and “really good”.

The fact that the really good ones are still in the group suggests that these are indeed
NOS or something like that. A lot of batches you see have the best of the best pulled out
and you are just buying “the rest”. I’d guess that Fair is in the same mode, you get the
whole lot and can sort time as you wish.

Just how many show up in the sub 6x10^-13 ADEV range varies a bit by lot. To me,
anything that’s in the parts in 10^-13 counts as very good. How many would improve
if you burned them in for six months or so, who knows….

Bottom line: You probably do better buying closer to the source of the surplus in a
case like this.

Bob

Bob Camp via time-nuts time-nuts@lists.febo.com writes:

I went this route too. The one thing I'll say is be sure to test them
for at least basic functionality on arrival. besttestus says they test
them, but they don't. I received 1 of 3 with some part of the oscillator
circuit dead (no output). Of course, the seller is happy to reimburse or
send another in this case.

Matt

Bob, so the ADEV gets better than factory spec - better over time?  The
number stated below approaches NIST_F1 at 1s ~2.8E-13. (Not sure if its
correct - found on a slide presentation) and apparently exceeds most
commercial cesium clocks for time scales less than~10s..

This is all fascinating since I now have two functional (as far as I
know) opt 4 5061As (with, of course, 10811s inside). Plotting the spec
for the 10811 and the 5061 in time lab is interesting, as the ADEV for
the 10811A does indeed  exceed the 5061 for time scales less than 1s.

Maybe this is old news to all of you, but it's more than I knew yesterday!

Jim

On 2023-01-10 8:57 AM, Bob Camp via time-nuts wrote:

--

Jim Muehlberg
Senior Engineer

National Radio Astronomy Observatory
ngVLA Local Oscillator Lead

1180 Boxwood Estates Rd B-111
Charlottesville, VA 22903-4602
P 434.296.0270
C 434.422.2017

Hi

There always has been a pretty wide range of ADEV’s on the 10811’s. One of the fun things
they did at Santa Clara was to screen them for ADEV. Once they found enough “good ones”
for that month, they stopped screening. The ones that made the grade got this or that part number
sufix on them and headed into gear that needed them.

There are indeed OCXO’s that hit <= 1x10^-13 at 1 second. There aren’t a lot of them, but they do
exist. The ones I’ve seen take a bit of time to get to whatever level the best level is. I have not seen
any 10811’s that are past 3x10^-13 at 1 second. If you are going to go sort crazy, there are a lot of
“targets” on eBay from time to time. Some of them have significantly better yield than 10811’s.

The risk with any bulk purchase is ( as noted earlier ) that you are buying the tail end of a group that
already has been sorted. Working this out without buying the lot is pretty much impossible. It is very
much a risky business.

The whole “sort out 10811s” process is not in any way a new thing. Folks have been doing this for
decades. There are a lot of stories about this or that adventure looking at a population of a hundred
or more examples.

On the “normal”  10811, the package is not hermetic. It soaks up humidity ( and maybe other stuff )
as it sits in storage. Has this or that one been in storage for a year, a decade, or many decades? There’s
not a really easy way to tell. Burn in is going to dry out the boards and components. If you are interested
in ADEV out to 100 seconds ( or worse yet 1,000 seconds ) the units need to be pretty “dry” to do their
best.

Fun !!!

Bob

A couple of weeks is usually enough for an OCXO to reach its best stability at taus in the 100-1000 second range.  Sometimes you do encounter good crystals that are jumpy at first and need some extra time to settle down, and in those cases, the ones that show visible improvement in the first few days are worth waiting out.  But generally an OCXO that is not improving after a week is unlikely to straighten up and fly right in another month/year/decade.

Of course, if the oscillator is brand new or has been powered off for years, then it may take weeks or months to achieve its best performance at taus >> 1000 seconds due to rapid-aging effects.

Like Bob, I've tested quite a few 10811s and have seen a few reach the 3E-13 level at t=1s.  That's good enough to justify going through a lot of clunkers to find the gems.  Also as he points out, any 10811-xxxxx can be "one of the good ones."  The -60109 parts used in the atomic standards will be among the best but are not necessarily the best, so you've gotta test 'em all.

Regarding NIST-F1 and other exotic clocks, those are often designed to interrogate the underlying quantum process at discrete intervals and adjust their OCXO periodically, rather than maintaining a traditional closed loop.  So when you see low -13s or high -14s near t=1s, you are probably looking at the performance of an ultrastable OCXO rather than the clock transition itself.  Not sure if that's true for the fountain clocks in particular, but it's how some of the trapped-ion standards work.

-- john

On 1/11/23 8:55 AM, Bob Camp via time-nuts wrote:

Cherry picking oscillators is a "standard thing" (although one that you
hope you'd not have to do).

My project is in the middle of looking at oscillators in an X-band
transmitter for phase noise/short term ADEV.

It turns out that DSN only has narrow symbol loop bandwidths (up to 25
Hz) and these transmitters (2 Mbps) weren't designed with that in mind,
so they'll lose symbol lock occasionally. In a 400 Hz loop bandwidth
there's no problem.   But DSN was designed and built for 10 bits/second,
insanely tight loop bandwidth to get the noise down.

And, of course, it's just on the ragged edge.  Some work, some don't. Of
course, the crystal mfr (I have no idea who it is, that's 3 subcontract
tiers down) probably doesn't specify anything - maybe they have an
overall jitter spec with some wide bandwidth.  And all the usual things

• how do you measure it without opening up the transmitter? How do you
  set up a test jig? What is the "real requirement"?

Had we realized this earlier, it's conceivable we could have flowed a
requirement down to the radio mfr, who would then flow that to the board
designer, who would flow that to the crystal company.  And turn $20
crystals into $2000 crystals, because the mfr would probably have to do
the cherry picking, and the yield would be terrible. (Of course, they
could sell the other ones per usual, but when you're ordering 10-20
crystals, and you have to screen 1000 to get the good 10, it might not
be worth even saving the ones that don't make the cut. They're now out
of the usual flow.

Another strategy would be for the radio mfr to make lots of boards, with
crystals as they come, and sort those.  That might actually be cheaper -
I'll bet the "as mfr" cost of the boards is <$150, which is a lot less
than $2000. and the "reuse" is easier.

This whole oscillator business is MUCH more complicated than a lot of
people think.

Hi Jim --

Cesium standards aren't necessarily all that good at short tau as
virtually all use the Cesium transition to steer an OXCO in a phase
locked loop.  The short term stability can be only as good as the OCXO,
which dominates the stability within the loop bandwidth.  So it's at
longer tau, outside the loop bandwidth, that Cesium stands out; the ADEV
improves decade-for-decade as the measurement interval increases out to
a final noise floor.  And of course Cs has the advantages of negligible
drift, good retrace, and of course that it defines the second (for now).

The HP5061A had a switch between (IIRC) 1 and 60 second loop time
constants.  You could use the long mode to get better short term
stability, if you were in a stable environment, had a good Cs tube and
OCXO, and the stars were aligned correctly.  If not, the loop would go
out of lock at some indeterminate time and the alarm light would go on.
The 1 second time constant was much more reliable, but noisier at short tau.

John

On 1/11/23 10:40, Jim Muehlberg via time-nuts wrote:

Hi

…. it’s worse …. a lot worse ….

If the parts are destined for space, each and every step along the way, they get a pile of paperwork.
There are lots of steps and thus lots of paper. It starts back at blank cutting ( if not at quartz growing)
and moves forward from there. Each one is part of a group and the group gets paperwork as well.
Pretty quickly, you have a $2 blank and $20 in paperwork for that specific blank.  That paper just keeps
piling up as it turns into a finished crystal.

Decide to screen a hundred crystals to find the best 1%? Unless the rest are going into space as well,
all that paperwork heads to the scrap pile. Even if they are headed to space, that’s a whole lot of
crystals. You better have a high volume customer to push them off on. ( good luck with that ….).

Since a crystal is pretty unique, both the custom guy and the high volume guy would need to be buying
a very similar end product. Obviously, a frequency change is a killer. A temperature range change
may well be. Shock / vibration / warmup / pull range ….. any one of those might kill the swap.

If you want to keep accounting happy, both the screen outs and the rest need to head out the door
pretty close to each other. The days of "put it in inventory and hope” are long gone. ( and for good
reasons ….. tons of those inventory crystal eventually went to the dumpster).

Bottom line is that if you screen 100 to get 1, in this case, you pay for the 100 …..

Bob

On 1/11/23 12:11 PM, Bob Camp wrote:

Yes in a "full Class A Space" build with traceability to sand - good
example is the Ultra Stable Oscillators from APL. Start 1000 blanks to
get a dozen good oscillator physics packages to get 4 finished USOs.
Blanks are cheap, finished oscillators are $1.5M each.

These days, there's a lot of "qualify as a box" with limited Class S/QML
parts.  We'll keep paper to know what parts were installed, so if
someone else has a problem then we can see if we have the same parts.
(bad luck if it's "generic 2N222" or "100 ohm resistor")

You pick parts that have industrial or automotive temperature ranges,
seem to have the right performance, buy a few batches and test them,
then build your boxes and  hope for the best (particularly if you're
depending on a "not on data sheet" number).  And that's what we're
facing. (I would imagine that if the mfr had realized that this was a
problem, they might have ordered something or screening or whatever. 
But yes, buy 100 oscillators to get 1 good one.