[time-nuts] OCXO retrace, how long is reasonable?

[Bob kb8tq]

Hi



> On Sep 4, 2020, at 3:37 PM, Matthias Welwarsky <time-nuts at welwarsky.de> wrote:
> 
> Dan,
> 
> I've attached the last 25000 seconds of data. The drift was about 15 DAC units 
> during this time. It's a 16 bit DAC, the pulling range is about 10Hz, so LSB 
> is worth 1.5e-4 Hz. 15 DAC units are 2.3e-3 Hz over 25000 seconds.

…. soap box mode on ….

When looking at this sort of data, by far the best thing to do is to convert it to relative
units like PPM, PPB or PPT. That makes the data independent of frequency. It allows
you to compare it to published data. Modern papers ( like since the 1950’s or so) pretty
much always present data in “parts per XX” sort of units. 

The other thing this allows you to do is to compare to measurements like ADEV
(which is also in these sort of units). 

Since there are a lot of people reading these posts, their “target frequency” may be
very different than yours. A frequency change of 0.0023 Hz on a 1 Hz output is 
quite a lot :) 

…. soap box mode off …. :)

> 
> Looking at your graph, it seems it took about  a week to
> express a slowdown in 
> drift, and maybe a month in total to sort of settle in? That's encouraging, 
> maybe I just need to give it a bit more time.
> 
> My temperature graph shows that there's some correlation, but not pronounced 
> enough to explain the behavior. The sensor is not thermally coupled to the 
> OCXO.

The data shown does not suggest a oven controller issue. 

One “gotcha” with DAC data: You have no way of knowing what’s drifting. It
could be:

1) The OCXO
2) The voltage reference to the DAC
3) The DAC it’s self 

Yes, there are some other less likely things, but those are enough to muddy 
the waters a bit.

Bob



> 
> BTW, the MDEV graph is mislabeled. It shows the DAC, not the TIC.
> 
> Regards,
> Matthias
> On Freitag, 4. September 2020 19:29:33 CEST Dan Kemppainen wrote:
>> Matthias,
>> 
>> How about a 'quick and dirty, hand wavey' graph?
>> 
>> This is a known 'kinda good' OCXO, that was off for about a year. The
>> DAC values are just that, DAC counts.  The black line should represent
>> about 8e-12 per day drift. (Sorry, I don't have better numbers than that
>> easily on hand). I think it's about 4uHz per count, or so.
>> 
>> Note 1: The first 13700 seconds of data were thrown away. That drift was
>> astronomic. (Basically vertical on this scale!)
>> 
>> Note 2: Also, note that this only one retrace, of one trimble OCXO.
>> 
>> Note 3: Notice the dates on the bottom of the graph. This is around 6
>> months of worth of drift. Small tics are 7 days apart, major tics 3 weeks...
>> 
>> Hopefully this helps!
>> 
>> Dan
>> 
>> On 9/4/2020 12:00 PM, time-nuts-request at lists.febo.com wrote:
>>> Message: 3
>>> Date: Fri, 04 Sep 2020 08:50:21 +0200
>>> From: Matthias Welwarsky<time-nuts at welwarsky.de>
>>> To:time-nuts at lists.febo.com
>>> Subject: [time-nuts] OCXO retrace, how long is reasonable?
>>> Message-ID:<5901813.EksvcnS80i at linux-5fgm.suse>
>>> Content-Type: text/plain; charset="us-ascii"
>>> 
>>> Hi,
>>> 
>>> I mentioned in a previous email that I seem to have picked a particularly
>>> bad (broken?) OCXO for GPSDO testing. It drifts at a very high rate, from
>>> the DAC graph I calculate it is around -4e-4 Hz/hour. It's been on for
>>> about 1 week now and the drift doesn't seem to come down, or maybe just
>>> marginally.
>>> 
>>> Of course, this being a surplus OCXO with unknown history, odd startup
>>> behavior is somewhat expected but so far I haven't seen anything this
>>> drastic.
>>> 
>>> Could this be OCXO retrace? How long can that phase last? I'm trying to
>>> decide if I should give it another week or just scrap it and move on.
>>> 
>>> Regards,
>>> Matthias
> 
> <graph.png>_______________________________________________
> time-nuts mailing list -- time-nuts at lists.febo.com
> To unsubscribe, go to http://lists.febo.com/mailman/listinfo/time-nuts_lists.febo.com
> and follow the instructions there.