Mark wrote:

It appears that most do but some don't.  Between the results I have
seen posted on the list (Lady Heather screen shots) and my own data,
they seem to fall into two groups.  I have two with excellent
oscillator tempcos, but a third unit I have is about 100x worse and
the sign of its tempco is reversed compared to the first two.  That
one appears to consume approximately the same oven power as the other
two, and heats the housing approximately equally, so the oven does
not appear to have a gross failure.  All three have Trimble 37265 OCXOs.

I initially thought the third unit's oven controller was broken (low
gain).  Then I noticed that the great majority of posted Lady Heather
plots appear to be from units similar to that one, with the much
higher tempco and reversed tempco sign compared to my two low-tempco
units.  But I have seen a few other plots that appear to be from
units similar to my first two.  The two with the low tempcos do not
appear to be inferior to other Tbolts with respect to stability, PN, or aging.

I'm inclined to think that all 37265 OCXOs are supposed to work like
my first two, and that the ones with large tempcos are the result of
a supply or manufacturing error (most likely, a mismatch between the
oven set point and the crystal).  But who knows?  There do seem to be
many more of the ones with large tempcos around.  It would be
interesting to take a few 37265s apart to see if there are any
obvious differences between the high- and low-tempco units, and if
tweaking the oven set points would reduce the tempcos of the high-tempco units.

Best regards,

Charles

Mark, Charles,

This is interesting and I'd like to pursue it. I still have a pile of original TAPR-Thunderbolts here so without too much effort I can gather tempco statistics for you. I tested the TAPR lots for phase noise and ADEV and DS1620 granularity, but did not test free-run tempco of the OCXO.

<rant> Please don't use the word "TERRIBLE". It's relative to XO, TCXO, OCXO, DOCXO, etc. This is time-nuts, not some feel-good audiophile list. We are allowed to make objective experiments and use scientific notation. The units for tempco involve Hz and C, or dF/F per K, etc. Given that, do either of you have actual tempco numbers? </rant>

Bob, do you have experience or an explanation for the bi-model tempco grouping that they have observed?

Thanks,
/tvb

tvb wrote:

I checked my notes and found that I did not record any free-running
tempco values.  My observations were based on the scale factors I had
to use to get the temperature and DAC graphs in Lady Heather to
overlay each other.  I initially noticed it because there was a very
pronounced tracking of the two graphs for one Tbolt and for the other
two there was not (the temperature-compensating component of the DAC
voltage is mostly lost in the noise).  I had checked the actual EFC
sensitivity of each oscillator in the vicinity of the operating
point, so all relevant variables were more or less controlled.

My impression is that the better ones are comparable to a single-oven
10811, maybe even a bit better.  LH typically reports tempcos of
1e-12/C to 1e-11/C.  My worse unit (and, from what I can infer from
LH plots posted to the list and on-line, it appears many others as
well) typically reports a tempco of 1e-10/C to 1e-9/C.  Of course,
the LH numbers are all to be taken with some caution since LH does
not have any a priori means to separate tempco and drift.

Best regards,

Charles

Hi

LH can get a bit confused about OCXO tempo. It's not really the software's fault, as you point out - the data just isn't there.

Bob

On Dec 14, 2012, at 5:36 AM, Charles P. Steinmetz charles_steinmetz@lavabit.com wrote:

I think what I'll do is collect some data on a batch of TBolts that I have soaking here. It seems to me there's enough information that, over time, the tempco can be accurately determined. I mean, when you see LH plots with glaring diurnal patterns in both temp and DAC it's easy to roughly calculate the correlation by eye.

Alternately, when the TBolt is in disciplining-disabled mode, the tempco can be inferred from temp and quadratic PPS offset residuals (EFC gain is not a factor in this case). Or for greater precision, a simple match of variations in ambient temp and externally measured frequency would do the job.

I use different software to talk with my TBolts anyway, logging all communication in its native TSIP binary. I'll modify or write an automated tool to take the guesswork out of it.

/tvb

----- Original Message -----
From: "Charles P. Steinmetz" charles_steinmetz@lavabit.com
To: "Discussion of precise time and frequency measurement" time-nuts@febo.com
Sent: Friday, December 14, 2012 2:36 AM
Subject: Re: [time-nuts] Thunderbolt oven / non-stable operating temperature

On 12/11/2012 10:33 PM, Charles P. Steinmetz wrote:

(( sorry to single out that one line ))

Just a curiosity. Is there any way to check that via software? Did you
just physically look under the cover, or how did you figure out which
type of oscillator your thunderbolt has?

Thanks,
Sarah

Hi

The OCXO is a "dumb" version. It does not talk to the TBolt. There's no way to check it in software. There are a few examples out there that have "late model" stickers on the outside and earlier parts on the inside. There's pretty much no way to know what you have without opening up the box.

When you open the box, be careful not to loose the hardware that goes on the F connector….

Bob

On Dec 14, 2012, at 1:42 PM, Sarah White kuzetsa@gmail.com wrote:

Sarah wrote:

You need to open it up.  There is a sticker on the OXCO can:

Emacs!

Best regards,

Charles

tvb wrote:

It would be interesting (to me, at least) to know the spread of EFC
gains from a reasonable population of Tbolts.

Best regards,

Charles

Hi

The real answer to that is going to be a "that depends" kind of thing. The population of units in the basement are all within 20% of each other as measured by LH's auto tune process.

Bob

On Dec 14, 2012, at 2:51 PM, Charles P. Steinmetz charles_steinmetz@lavabit.com wrote:

On 12/11/2012 10:33 PM, Charles P. Steinmetz wrote:
((...snip...))

((...snip...))

I'm pretty sure LH estimates the temperature coefficient in a rather
inelegant way:

Seems based on the graph / display window, and affected by filtering
(RMS is affected by display filtering too) I've been doing measurements
this past week, and posted the first set of results here:

http://spaceclock.tumblr.com/post/38439854001/

As you can see, with minimal, or filtering completely disabled, LH has a
significant jitter problem in the tempco, and sometimes even shows a
reversed sign.

In my testing, a high or low time constant seemingly has no effect on LH
ability or inability to mangle its calculation of the temperature
coefficient...

The time constant just happens to be 4096 in this case because I'm doing
long-term measurements on the effects of using a value other than 100,
and ultimately employing a binary search algorithm to determine the most
optimal time constant for the crystal in this particular thunderbolt.

--Sarah