Keep in mind that the 7 min "warmup time" of the PRS-10 is jut the time for
the
unit to get the physics package close enough to final temperature for the
dithering
loop to lock to the atomic transition. But based on measurements of current
drawn
by the oven(s), it takes more like an hour for the ovens to settle down
very well.

My own experiments with a well-used "telecom mod" PRS-10 suggest that it
takes
my unit more like 3 or 4 days to really settle down to its background
frequency drift
rate.  This rate is specified as something like "< 5E-11 per month", which
is frankly
pretty awful.  Mine seems to be better than spec, more like 2E-11 per month
(frequency
increasing over time).  The point here is that there are apparently a
number of warm
up drift mechanisms operating, some of which take days to sensibly settle
down.

I've been wondering about the issue of whether it's best to operate a
PRS-10 only
at times of need, or continuously, with respect to life.  I posed the
question direct;y
to SRS, and they claim that it is better to leave it running all the time.
My own
"algorithm" has become "leave it on all the time" when I'm around and even
remotely
active with time-nuttery, but shut it down when I'm away on vacation.  This
is mostly
out of concern for something going wrong with the power supply and burning
the
house down.  Lightning storms are one such risk, for example.

Dana    K8YUM

On Fri, Mar 8, 2019 at 3:00 AM Forrest Christian (List Account) <
lists@packetflux.com> wrote:

On Fri, 8 Mar 2019 01:18:45 -0700
"Forrest Christian (List Account)" lists@packetflux.com wrote:

The warm-up time is from "I switched it on" to "there is a 10MHz signal
and it's 'stable'". I.e. at latest after 7 minutes you should be able
to use the ouptut of the PRS10. But keep in mind that the stable-after-warm-up
is not tha same as stable-for-a-time-nut. A Rb vapor cell standard has
a retrace (ie change of frequency after being switched on, for details see
John Vig's tutorial) of several weeks to a few months, depending on the
exact construction of the Rb cell, how long it has been in use and how
long it has been off.

Switching it on and off several times a month would be worse in
terms of longevity  than having it running continuously. The vapor
cell itself is usually run between 60°C and 100°C and the Rb lamp
usually between 120°C and 170°C. As you can imagine, power cycling
gives a huge thermal strain on all the components. The inner construction
of the PRS10 is not optimized for power cycling but thermal insulation
of the sensitive components (see e.g. http://time.kinali.ch/Rb/PRS10/ )

So I would rather feed the PPS to the PRS10 directly and let itself
do the work. This will also give you an optimized control loop for
the performance of the PRS10 without you having to go through the
process of desiging and optimizing the loop.

If you want to do just jitter measurement, then I wouldn't use a Rb standard
but a stable OCXO instead. At averaging times below 1-10s an OCXO will beat
any Rb standard, even one that has an OCXO like the PRS10 (admittedly, it's
not a high quality OCXO but it's one none-the-less). An OCXO has a shorter
warm-up time (usually 2-3 minutes until stable, usually less than a day
until time-nuts-stable, less than a month until nutty-time-nut-stable).

I personally, would use an OCXO as reference to the TICC (or TICCs?)
and use the PRS10 as an additional input to the TICC. This way you
can do a multi-way comparison between the GPSDOs and the Rb, removing
out the (in-)stability of the reference OCXO.

Beside those comments, you have not told us at what kind of performance
you are aiming at. There is a huge difference whether you need 1e-10 or
1e-13 and whether you need it at 1s, 1ks or even 100ks averaging times.

			Attila Kinali

--
It is upon moral qualities that a society is ultimately founded. All
the prosperity and technological sophistication in the world is of no
use without that foundation.
-- Miss Matheson, The Diamond Age, Neal Stephenson

Hi

Some of this gets into just what you are trying to do. Cycle to cycle jitter on the
PPS output of most GPS modules is in the “many ns” range. If sawtooth correction
is not being applied, a measurement that is good to an accuracy of 1 ns / 5 = 200 ps
would be plenty good enough. That is a 2x10^-10 (0.2 ppb) sort of accuracy.

If you are looking at longer term effects, you may need better accuracy / lower drift.
You can easily get past what a nice new 5071 can do heading down this road … For
now let’s assume the target above is correct.

The PRS-10 should be within 0.2 ppb within a few minutes after locking up. It should
hold this sort of accuracy “on the shelf / power off” pretty much forever and ever. In fact,
it’s a pretty good bet that it will hold 10X this accuracy for quite a long time. You probably
should warm it up for a day or so if you are after 0.02 ppb.

Calibration wise, the “easy” way to do it depends a lot on what you have to do the measurement.
If you have a TICC (sounds like you do) - use it. You can also use a scope or a more
conventional counter.

If you are going after 0.02 ppb, you would want to set the unit to 0.002 ppb. That gives you
90% of the budget for things like temperature drift and aging. Indeed you might want to go
to a tighter number …. this is Time Nuts after all.

0.002 ppb is 2 ns over 1,000 seconds. It’s 200 ns over 100,000 seconds. (Amazing how that
works :) ). Is your TBolt PPS good to 2 ns? Probably not. Is it good to 200 ns? most certainly.
Since 100,000 seconds is just over a day, that makes for a fairly easy adjustment process.
Look at it once a day and tweak it. After a while it will not need tweaks any more.

The advantage of a “long term” adjustment like this is that it also takes in the temperature
swings in your lab over a one day period. At the 0.002 ppb level, they will be the dominant
part of what you see. Will you get to the 0.002 ppb level? It depends a lot on just how stable
your particular unit is and how drafty your lab is.

Running any sort of electronics gear 24 hours a day is a risk. It may wear out, it may cause
a fire, it may be fine … who knows. It also pulls power and you pay the electric company for that.
It’s generally cheaper / safer to turn stuff off if it is only going to be used rarely.

Lots of fun !!!

Bob

Hi

Putting some dimensions on this:

A “telecom” Rb should be in the 0.01 ppb at one second (ADEV) range. It should
improve by sqrt(tau) so at 100 seconds it will be at 0.001 ppb.  It should improve
some past that. Counting on 0.0001 ppb at 10,000 seconds …. not so much.

A cheap / small OCXO may be at the same 0.01 ppb at 1 second level. An OCXO
does not have a “predictable” improvement rate. You may have the same at 100
seconds as at 1 second.

A very good (but maybe cheap … who knows …) OCXO could get you to 0.001 ppb
at 1 second. You also could buy a few dozen to get one that actually does that. Testing
could be involved to sort them out.

Both the Rb and the OCXO have temperature stability specs on them. What those
numbers are may be unknown on a surplus item. The “best case” claimed temperature
stability on a telecom Rb is often better than the “best case” stability on your cheap
OCXO.

Indeed there are (rare) OCXO’s out there that get down to 0.0001 ppb  at 1 second
and hang in there out to 1,000 seconds. Finding one at all can be tough. Finding one
cheap … you would have to be very lucky. Working out that you do have one will involve
testing.

Bob

On Fri, 8 Mar 2019 06:31:48 -0600
Dana Whitlow k8yumdoober@gmail.com wrote:

Longer. I know of one measurement, where the Rb had a kind of stable
drift until it suddenly switched to another slope quite suddenly
(within a few days) about half a year after power up.

The aging mechanisms of Rb vapor cell standards are many and not
all of them are well understood, much less controlled. Compared
to that, an OCXO has "only" thermal stability of the oven, strain
relaxation of the holder/crystal and deposition/removal of contaminants
on the crystal surface. Ok, there are a couple more, but these three
are the main contributors for most OCXO out there. While for the
Rb vapor cell standard I could name you half a dozen just like that
and I am far from being an expert on these.

For those interested, John Vig wrote a couple of papers on the aging
of OCXO in the 80s and 90s. The topic of Rb vapor cell aging is a lot
more messy and I don't know whether there is any good paper that reviews
the main contributors.

		Attila Kinali

--
It is upon moral qualities that a society is ultimately founded. All
the prosperity and technological sophistication in the world is of no
use without that foundation.
-- Miss Matheson, The Diamond Age, Neal Stephenson

今日は

On Fri, 8 Mar 2019 14:09:20 +0000 (UTC)
Taka Kamiya via time-nuts time-nuts@lists.febo.com wrote:

There are techniques to cut the "first lock time" down quite a bit.
Most of them involve either switching of all the long time constant
parts of the control loop or switching between different configurations
(ie different multiplicative constants or even different topologies).

The working principle of all is, that once you are reasonably close
in frequency/phase, your long-time-constant loop will not have a
large difference to work with and its internal state converges quickly
to something close of what it should be.

One very simple approach (which not always works) is to decompose your
PI controller into a P and I part, and disable on startup the I part.
Wait until the P part has gotten the error to something stable (not
neccesarily small) then switch on the I term, with its internal state
initialized to zero. A more sophisticated variant of this, uses an I
part with a smaller time constant first to reduce the residual error,
then estimates from its internal state the state of the real I part and
switches over to that.

Analysis of these dynamic loops can be a bit cumbersome and I recomend
to read a good textbook on this topic before attempting to do something
like this, as it's very easy to shot your foot.

			Attila Kinali

--
It is upon moral qualities that a society is ultimately founded. All
the prosperity and technological sophistication in the world is of no
use without that foundation.
-- Miss Matheson, The Diamond Age, Neal Stephenson

We have operated  about 40 PRS10s over the past 20 years or so. These are
all run continuously in benign environments and monitored via GPS
time-transfer . Some have died after just over a year; others have operated
for more than 12 years. A new one generally takes a few months to burn in,
before its frequency drift stabilises to something like the specifications.
They do not behave very predictably, showing sudden changes in frequency
and so on. One exhibited frequent steps for a few years and then the
problem went away. Trying to predict what they would do at the level of
better than a few parts in 10^11 seems difficult.

Cheers
Michael

On Sat, 9 Mar 2019 at 3:03 am, Attila Kinali attila@kinali.ch wrote:

This seems like as good of a message to reply to as any.  Thanks to
everyone who has replied so far -  lots of good information, which is
right in the category of what I was looking for.

Let me clarify what I'm hoping to measure.  I apologize if some of
the terminology below is wrong, I'm still learning correct use of some
of the terms - I've tried to explain in a way where it is obvious what
I'm talking about even if the terms are slightly wrong.

The GPS receivers I'm hoping to measure are used in an application
where the 1PPS is used to synchronize radio transmissions.  Generally
one wants the pulse to be within a couple uS or so of the top of the
second.    Older radios cared less, newer ones want it to be more
precise.  Sorry about the soft spec, it unfortunately is what it is.
One note is that there is some expectation that there will be be
some fixed offset from the UTC second due to cable delays and the
like, so I'm less worried about a reasonably low fixed offset  and
more about the variable (jitter) offset.    I.E. 1uS +-100nS is
perfectly fine, 1uS +-1uS isn't so hot.

In evaluating a new set of GPS receivers for this application, I'm
hoping to gather Pulse-to-Pulse delay figures for every pulse the GPS
receiver outputs over a week or so.  I might end up repeating it in
various forms, including moving the GPS receiver into an
"environmental chamber" (aka modified chest freezer) to verify that
they still behave in low or high temperature - or if I can figure out
a reliable way to do so, to simulate various signal impairments such
as partial sky view.

My intent is to then process the raw data to get a feeling for the
statistical nature of the 1PPS out of that particular GPS receiver.
For instance, how many pulses were outside the +-ns spec for the
receiver.  What did the histogram of the in-spec pulses look like?
And so on. For some of the processing it's pretty obvious that I need
a clock source for the TICC which is going to be stable over the
entire period or else I'll end up measuring errors in the clock source
instead of the GPS receiver.    It seems to me that if I'm looking for
a histogram supposedly confined to +-100nS, I'm probably going to want
to be able to measure around +-1nS with confidence.  Much more than
that (0.1nS) seems a bit excessive, less than that (10nS) is too
coarse.

Intuition and past experience implies that if I want to measure +-1nS
with confidence, I really need a slightly more accurate clock source
than the measurement itself.  0.1ppb over the measurement period seems
reasonable, 0.2ppb is probably good enough.  I don't think I need
0.02ppb.

I replied in particular to Bob's message below since it seems to
confirm what I was thinking, even without me being verbose enough
about my application.  It also sounds like powering this on for a day
or so before I start (or getting started and just throwing the first
day's data away) might be reasonable to do and I don't need to let
this thing run for months ahead of time to get to the accuracy I need.
Of course, I'll do a bit of analysis on the data to try to verify
that the unit isn't drifting.

Out of curiosity, I'm also going to do some measurements on the couple
of standalone OCXO's in my collection to see if either would be
accurate enough for this particular measurement.  The best one in my
(admittedly limited) collection seems to be spec'd worse than I'd
need, although I'm not sure how much aging it's done.  None of them
have seen any meaningful poweron time just because I've been too busy
to play with them.  I guess that at some point I should also up my
game as far as good quality OCXO's since I do see how the right
well-aged low-drift OCXO would probably be fine for this measurement.

After reading everyone's response I'm still considering what to do as
far as long-term runtime on the PRS10.  I do know that I'm going to
need this off and on for the next few months so once I get it in a
permanent enclosure and get it powered up it will probably stay on
continuously at least until I am done with this round of experiments,
at which point it might get turned off with the hopes of extending
it's service life.

On Fri, Mar 8, 2019 at 8:06 AM Bob kb8tq kb8tq@n1k.org wrote:

--

• Forrest