SAIDJACK@aol.com wrote:

Said

This was originally sent at 3am local time but doesn't seem to have made it.

The traditional method uses an AC transformer bridge with a platinum
resistance sensor and a stable low temperature coefficient reference
resistor.
However this is probably too bulky and expensive for your application.

As I understand it, you don't really need to measure the temperature but
some temperature dependent quantity that is a monotonic function of
temperature with a resolution and and stability equivalent to better
than 0.001K.

Measuring the ratio of the resistance of a platinum resistor with to a
stable reference resistor with a low temperature coefficient will
suffice as long as they both have the same temperature. A reference
resistor with a small non zero tempco will not affect the monotonicity
however the sensitivity will be affected slightly.

It is not even necessary to use a platinum resistance sensor, a copper
one will work just as well, and may even be convenient as a length of
enamelled copper wire can be would around an object (metal container?)
whose temperature one wishes to sense. Indeed the NBS (now NIST) used
copper resistance thermometers to sense the temperature of its 10pF
fused silica standard capacitors.

One could replace the variable variable ratio cascaded tapped
transformers used in the the AC bridge by a pair of multiplying DACs
with an AC reference, however the DACs would need to be 20 bit DACs that
are monotonic to better than 1 lsb. The DACs being adjusted to null the
output at the junction of the 2 resistors.

Another possibility is to use a 24 bit sigma delta ADC to measure the
ratio of the voltages across the RTD and the reference resistor. Off
course it would be prudent to measure this ratio also when the current
through the 2 resistors is reversed in direction so that residual
offsets due to thermoelectric and other causes cancel out. Since the
ratio of the voltages is being measured  the current flowing through the
2 resistors need only have good short term stability as must the ADC
gain. As long as it is monotonic the ADC need not have an integral
nonlinearity as small as 1 lsb. An LTC2412 or similar ADC should
suffice. The trick is to reverse the current flowing through the series
connected RTD and reference resistor without affecting its absolute
value. A current source feeding the pair of series connected resistors
via a suitably connected set of switches will suffice. Details will
follow later its 3am here.

An alternative technique is to use a quartz crystal cut with a high
tempco in an oscillator and just measure the oscillator frequency using
the OCXO as a timebase.
The HP2804 quartz crystal thermometer used to do this.

Bruce

Hi Brendan:

It's my take that there are two aspects of the Brooks (no relation)
design that need to be addressed for optimum operation:

1. the filter time constants of the stock design are not correct for a
   Rb oscillator, and that can be fixed by getting a custom PIC from Brooks.
2. the TIC although suitable for use with the older Motorola GPS
   receivers is not optimum for use with the newer M12+T receivers.  There
   is not a fix for this now, but maybe in 2007.

Keep in mind that this design does work and that the above items relate
to optimization not bug fixes.

Have Fun,

Brooke Clarke

w/Java http://www.PRC68.com
w/o Java http://www.pacificsites.com/~brooke/PRC68COM.shtml
http://www.precisionclock.com

Brendan Minish wrote:

On Tue, 2006-12-19 at 11:05 -0800, Brooke Clarke wrote:

I currently have the Brooks Shera design disciplining my LPRO-101
without any mods and using the ADC connected directly however the gain
is not high enough for the small EFC range of the LPRO.
It does work, it will keep it locked OK even with the longest
time-constant and the figures appear to make 'sense'. I guess that with
the excellent stability of the Rubidium Oscillator the lack of loop gain
may even be an advantage as it will reduce sensitivity to other sources
of short and medium term error

I certainly intend trying the modified PIC software to improve gain and
lengthen the filter time.

I would however be most interested in building and comparing with an
alternative discipline arrangement

Anyone care to comment on projects based around the SRS PRS10 module and
how it fares out in the scheme of things ?

73
Brendan EI6IZ

Hi Brendan:

It's interesting that the PRS10 can time stamp the 1 PPS input with a
resolution of 10 ns.  I wonder how they do that.
http://www.thinksrs.com/downloads/PDFs/Catalog/PRS10c.pdf

When I picked up my PRS10 at the factory during the tour I learned that
the 10 MHz oscillator in the PRS10 was an improved version of the SC10.
There's an Allan plot for the PRS10 at:
http://www.thinksrs.com/products/PRS10.htm <- overall PRS10 page
http://www.thinksrs.com/assets/instr/PRS10/PRS10diag2.htm  <- Just the plot

Have Fun,

Brooke Clarke

w/Java http://www.PRC68.com
w/o Java http://www.pacificsites.com/~brooke/PRC68COM.shtml
http://www.precisionclock.com

Brendan Minish wrote:

From: Brooke Clarke brooke@pacific.net
Subject: Re: [time-nuts] LPRO-101 with Brooks Shera's GPS locking circuit
Date: Tue, 19 Dec 2006 13:42:28 -0800
Message-ID: 45885CC4.30007@pacific.net

It actually says it does it with 1 ns resolution. A x100 analog interpolator
circuit for 1 ns to 100 ns (10 MHz) isn't rocket science. See the HP 5335A
operations and service manual. Getting 10 ns accuracy isn't supprising.

Do the PRS10 comes with the 5 dollar tour ticket or is it the other way around?
:D

The PRS10 seems like a good device.

Cheers,
Magnus

Magnus Danielson wrote:

The PRS10 is somewhat marginal for VLBI although it has been used for
short integration times to observe the L band hydroxyl line at around
1600 MHz.
Fortunately all stations but one used hydrogen maser clocks/frequency
standards. Some of the better crystal oscillators (Oscilloquartz 8607
BVA, FTS1200 etc) have better performance for integration times up to
100s or so. The PRS10 OCXO DAC has a fractional frequency resolution of
around 1E-12. This limits its performance for VLBI with long integration
times where it is necessary to have phase synchronisation between all
stations to within 10 degrees or so over the integration time.

Bruce

From: Magnus Danielson cfmd@bredband.net
Subject: Re: [time-nuts] LPRO-101 with Brooks Shera's GPS locking circuit
Date: Tue, 19 Dec 2006 23:00:55 +0100 (CET)
Message-ID: 20061219.230055.-706958800.cfmd@bredband.net

Actually, if you look in the PRS10 manual you find the actual answer. I wasn't
that far off just by my educated guess. They do a x2000 time expansion against
a 2,5 MHz (400 ns). You can read-out and write in the calibration value for
the slope. The actual resolution is 200 ps but they do not use up more than
1 ns in the "public" interface. It actually uses the CPU builtin counter, which
will do for the purpose. They could have spent a little more and got better
single shot resolution out of it, but I suspect they didn't see the need.

http://www.thinksrs.com/downloads/PDFs/Manuals/PRS10m.pdf

Cheers,
Magnus

From: Dr Bruce Griffiths bruce.griffiths@xtra.co.nz
Subject: Re: [time-nuts] LPRO-101 with Brooks Shera's GPS locking circuit
Date: Wed, 20 Dec 2006 11:24:42 +1300
Message-ID: 458866AA.3000209@xtra.co.nz

If you consider telecom and many instrument applications the PRS10 has
sufficient performance. I've actually sat at meetings and had the PRS10
behaviours as a reference frame.

VLBI has quite a different requirement on performance. Actually, I'd like to
hear what frequencies and performances you actually use and have.

Cheers,
Magnus

In message 20061219.235925.2053155698.cfmd@bredband.net, Magnus Danielson wri
tes:

They are limited by digital noise inside the box.

I used the "vernier trick" we talked about yesterday on the PPS
input of my first PRS10, trying to determine if the calibration was
still OK.

It was quite obvious that there were 10 MHz overtone noise affecting
the PPS timestamping.  There were both missing codes and jumps
corresponding to at least the 3rd and 5th overtones.

Interestingly, when you take an Oncore PPS and hook it up to the
PPS on a PRS10, and transport the negative sawtooth over the serial
port (early PRS10 firmwares cannot do this) you get incredibly good
performance, because the sawtooth "dithers" the systematics of the
PRS10 out of the picture.

If you feed the PRS10 with a 1PPS derived from a Cesium, you may
end up worse because the input PPS signal parks itself on one of
the unlucky points in the noise-spectrum.

Noise like the oncore sawtooth isn't always a bad thing.

Dithering noise is a very important noise-cancellation technique,
in prof. software defined radios.

They add an analog version of a digitally generated PRNG signal to
the analog input signal, right before the A/D converter and then
subtract it right away again on the digital side in software.

That way any imperfections in the A/D converter gets spread out
over the frequency spectrum rather than appear as sharp spikes,
just like the GPS signal is spread out with a PRNG code.

But similar techniques are very useful in timing as well.

You can increase the resolution of time interval counter down to
the noise limit by adding a sinewave to your input signal on the
analog side remove it again from the counters result and average
long enough.

--
Poul-Henning Kamp      | UNIX since Zilog Zeus 3.20
phk@FreeBSD.ORG        | TCP/IP since RFC 956
FreeBSD committer      | BSD since 4.3-tahoe
Never attribute to malice what can adequately be explained by incompetence.

From: "Poul-Henning Kamp" phk@phk.freebsd.dk
Subject: Re: [time-nuts] LPRO-101 with Brooks Shera's GPS locking circuit
Date: Tue, 19 Dec 2006 23:17:03 +0000
Message-ID: 48601.1166570223@critter.freebsd.dk

Ofcourse they are. It is bleeding obvious. If they would have cared more about
it they should not have stuck the PPS as another pin in that connector for
starters. But then again, I think it is sufficient for what I perceive to be
their market and more power to them for it.

Interesting. Signal integrity is really crutial in acheiving the full potential
of such an interpolator.

A little noise or signal into such non-linearities do improve quantization
noise. Known fact and still valid.

I.e. don't forget to phasemodulate up some jitter on the PPS! :D

Indeed.

... and in tones of audio-gear etc. etc. etc. and your GPS receiver!!!

Dithering is a technique one should learn to handle for quantization noise and
similar small-scale non-linearities. What type of measure which is quantized
this way is really of a lesser important thing since the concept and related
math is more or less the same. It is just a method, which sometimes applies
itself without the explicit intent.

Actually, the thermal noise make a hell of a good quantization dithering
signal for the often 1.5 bit A/D. The integrate & dump is where there is bits
showing up.

Indeed.

Indeed.

Cheers,
Magnus

Hello folks,

i like to play the bad boy again: My claim is

a) that for most of us a GPSD Rb is of little to no use compared to a
good GPSD xtal oscillator

b) that it is a myth that you may use longer loop time constants with Rb
compared to a xtal oscillator

Let us first talk about b). It is necessary to have some basic knowledge
on temperature controllers in order to understand that. Unless the
temperature controller has an active cooling element (i.e. it may heat
AND cool) there is a very simple mechanical model for an temperature
controlled oven like being used in OCXOs:

Imagine a pot having a small hole through which a fluid can slowly pour
out of the pot. This hole represents the oven's insulation against the
surrounding world and the fluid pouring out represents the energy that
the oven looses to its surrounding due to the fact that the insulation
is not 100%.

Then there is a person with a second pot of fluid. This person can pour
out fluid from the second pot into the first pot. The second pot
represents the oven's heater by which thermal energy may be poured
inside the oven and the person is the temperature controller. The oven's
temperature is represented by the level of the fluid in the first pot.
The controller's task is to always pour just enough fluid from the
second pot into the first pot to keep the fluid level constant despite
the fluid lost through the small hole. One refinement of the model is
that we also consider that the amount of fluid pouring out of the hole
shall not only depend on the hole's size but also on the fluid level
itself inside the pot as well as the surrounding's temperature for which
there is no good counterpart in the model. While this mechanical model
is very easy it resembles everything very well what we need to
understand about oven controllers.

Now that we have this mechanical model, we can think about the
parameters that influence the model's behaviour. One parameter is the
size of the hole. The first idea that we might have is to make the hole
as small as possible = to make the insulation as good as possible. There
are lots of people who pack their HP10811 in big amounts of insulating
material in order to improve it. But is it an improvement? It is surely
an improvement in terms of energy because the fluid (=energy) pouring
out of the hole is lost and we constantly need to put an amount of fluid
(=energy) into the oven to keep the temperature constant.

Now let us consider what the better insulation does for the control
loop: Unlike in other controller loops we cannot take fluid (=energy)
OUT of the oven. We can only put fluid (=energy) INTO the oven. Once the
controller has generated a overshot (a common effect in controller
loops) we cannot compensate for that overshot because the only way that
fluid (=energy) can leave the process is through the little hole.
Result: If a regulation overshot happens the time constant to get back
to the correct temperature depends on the size of the hole. The smaller
the hole the longer it will take to get back to the right temperature.
Because of that we need to make the loop time constant long enough to
hopefully avoid any overshots at all and near us the right temperature
very slowly from below. However, if there is now a sudden step in the
surrounding's temperature this will have change the amount of fluid
(=energy) leaving the hole per time unit and the long loop time constant
hinders the loop to react as fast as we would like. Most OCXOs are built
that way. Note that the hole (= the not 100% insulation) is the ONLY way
that the temperature inside the oven can be made smaller. People who
addionally insulate their OCXOs make the hole size smaller and that has
the effect that the loop time constant is now too fast for this oven.

In contrast to that one may have the idea to make the hole really big
compared to the situation above. Consider an oven that is not insulated
at all. Instead it has an heating element on one side and a heat sink on
the other side. Clearly, because the oven is thermically good coupled to
its surrounding it looses lots of energy to the surrounding (= big hole
size) so we need to put lots of energy (=fluid) into it permanently. Not
a good idea in terms of efficiency but note the effect on the
controller's loop time constant. Everything can be made fast compared
the small hole size. If there is overshot we can expect the heat sink to
remove the overshot quickly.

The concept of the big hole size is what we can find with temperature
controllers in Rb oscillators. Note: We are not talking about the
temperature of the xtal oscillator within the frequency contrtol loop.
Instead we talk about the temperature controller for the Rb lamp which
is made the 'big hole size way'. Ever had an FRK-L in your hand and
wondered about the heat sink on its back? Thats the big hole. Ever
wondered why an LPRO shall be mounted on a heat sink of sufficient size:
Thats the big hole.

Whats the whole story good for and what impact does it have on claim b)?
A thermically good insulated OCXO like an FTS1200 may show a 6 hr(!)
delay until its frequency reacts on a step change in surrounding
temperature. In contrast to that Rb reacts almost IMMEDIATLY to any
change in surrounding temperature. Big surprise!

Neverteless it would not be honest to state this without discussing HOW
MUCH the OCXO and the Rb reacts because their tempco is different. In my
flat i measured the following tempcos

HP10811: 1.4E-11/K @ 25 °C
FTS1200: 7.7E-12/K @ 25 °C
LPRO  :-6.9E-13/K @ 25 °C

With an LPRO and some daily degrees temperature change in your flat you
may expect frequency variations of some parts in 1E-12 which follow the
temperature immediatly, much more then the 1E-12 aging that we would
have expected from the Rb. Note that this is what i measured on a
relaive modern design like the LPRO. Older constructions like the FRK-L
may exhibit a bigger tempco. The FRK-L's specs suggest a typical tempco
of 5E-12/K giving raise to frequency changes of some parts in 1E-11
during a typical day.

As a result you may set the loop time constant

a) to a BIG value which will NOT compensate for the temperature driven
effects, then your standard's output frequency will be supperimposed by
the temperature driven effects

b) to a smaller value which will compensate the temperature driven
effects but will be no improvement in terms of loop time constant
against a xtal oscillator.

Of course the whole discussion does not apply if your flat is
temperature controlled to within a degree or better.

If a Rb cannot be used with a big time constant it is no improvement
against a xtal oscillator, the main difference being that it  needs more
energy.

Concerning a): Every Rb contains a xtal 'flywheel' oscillator. While
being a OCXO this oscillator is NOT of the same class and quality as a
OCXO as the FTS1200. This being due to the fact that it serves only as
the 'flywheel' for the atomic processes and the atomic processes
determine the long time behaviour of the complete Rb oscillator. That is
why a good OCXO may exhibit an ADEV being an order of magnitude smaller
that the output of a Rb at observation times tau=1-100s.

For that reason the best choice in simplicity, ADEV, money and energy is
to combine a good GPS receiver with a good OCXO. If simplicity, money
and energy do not count an perhaps better design would use:

a) a Rb locked to GPS with a relative short loop time constant. That
would guarantee that the overall task of keeping the loop locked is
easier to fulfill with Rb less prone to environmental changes

and

b) a very good OCXO which is phase locked to the standard's output with
a loop time constant of say 50 s to improve the ADEV for small
observation times.

Regards
Ulrich Bangert, DF6JB

In message 000001c7244b$09275bf0$03b2fea9@athlon, "Ulrich Bangert" writes:

... if your hold-over requirement is trivial.

--
Poul-Henning Kamp      | UNIX since Zilog Zeus 3.20
phk@FreeBSD.ORG        | TCP/IP since RFC 956
FreeBSD committer      | BSD since 4.3-tahoe
Never attribute to malice what can adequately be explained by incompetence.

At 10:25 AM 12/20/2006, you wrote:

The model for the surounding temperature is a larger pot with a fluid level
which is lower than the level in the "oven" pot.  Flow rate is dependent on
the difference in the fluid levels, with the oven loosing less heat when
the 'outside' pot is at a higher level. (higher temperature)

Tom Buehl

Yes, agreed!

Ulrich,

That's a rather general statement, but I understand what
you mean. Consider, instead of a bold general assertion
which can be disproved with a single counter-example,
a simple list of advantages of quartz vs. rubidium. Then
let the user decide which is appropriate for their unique
application.

There's more to a GPSDO than its ADEV. For example,

Warm-up time --
Many Rb will lock in 5 minutes, typically. Some Qz
take much longer to get on-frequency from cold start.
This can simplify the initial loop locking algorithm.

Power consumption --
Probably Qz-based GPSDO have much lower power
consumption than Rb.

Physical size --
I would guess most Qz GPSDO are smaller than Rb
GPSDO. One could argue that the Casio GPS wrist
watches are a type of miniature GPSDO!

Hold-over performance --
For mid- to long-term, Rb is vastly superior to Qz;
most Rb have daily drift rates 100x better than Qz.

Stand-along performance --
Without GPS lock, a free-running Rb can be trusted
to be orders of magnitude more accurate than Qz.

Environmental --
Is it the case that Rb is less sensitive than Qz to
extreme environments?

Cost --
As a rule, Qz-based GPSDO are cheaper than Rb.

Phase noise --
I'd guess that Qz-based GPSDO could have better
short-term stability and phase noise than Rb.

Tuning range --
Could it be that Rb will run for many more years without
coarse adjust than Qz?

Lock indicator --
Most Rb give you a signal when frequency lock has
been achieved.

Mechanical sensitivity --
Rb GPSDO are much less sensitive to movement and
shock than Qz. If you don't believe me watch what
happens if you simply take your GPSDO and bump it,
or turn it over on your desk. This is especially important
during hold-over when no external correction is present.

Lifetime --
Is the MTBF of Qz much longer than Rb due to fewer
parts and simpler design?

Surplus --
There are so many cheap telecom Rb out there it is
near irresistible to make a GPSDO out of them.

You get the idea; feel free to add the ones I missed.

/tvb

In the matter of lifetime (outside of MTBF issues), is it correct that
Rb has a built-in life limiting mechanism (the lamp wears out), where Qz
does not? If so, Rb oscillators will eventually fail but one might hope
a Qz oscillator may not.

Didier

Tom Van Baak wrote:

In message 003501c72471$6fe791e0$3b8c8843@computer, "Tom Van Baak" writes:

Initial capture is best done with a looser timeconstant in any
circumstances.

Note that the integrator terms initial condition is undefined in a
PLL, this can be used to achieve lock without the initial overshot:

Clamp the integrator term to zero and let the proportional term
drag the offset into range (use a high rate for this, there is
no stability issues).

Once the second derivative of the offset approaches zero, unclamp
the integrator and switch to a normal but loose set of constants
for the PLL.

With properly chosen values, you can drag any frequency source
into submission of a PPS that way in a fraction of a minute.

After this the PLL can adapt its constants based on the statistics
(remember what I said earlier about looking at the ADEV shape).

Single oven: Maybe, double oven: certainly.

I is likely to be the difference between replacing the GPS antenna
now or after the snowstorm is over.  Given the price difference,
this may be a no-brainer advantage to the Rb.

Also, if the qz in the rb jumps, the Rb is very likely
to tell you it lost lock.  A Qz unit will jump and you
will not know it, unless the resulting phase jitter
kills your microprocessor or similar.

No significant difference with proper design.  The Rb's
cooling requirements are tricker to design for than an
Qz units "bolt down and forget".

That's actually not a given.  For a decent Qz performance
new price approaches $1k and a PRS10 is only $1.5k.

Depends on your PLL more than anything else.

Yes, no chemical stress and with proper drive levels,
no mechanical stress either.

--
Poul-Henning Kamp      | UNIX since Zilog Zeus 3.20
phk@FreeBSD.ORG        | TCP/IP since RFC 956
FreeBSD committer      | BSD since 4.3-tahoe
Never attribute to malice what can adequately be explained by incompetence.

Let me add a note.  The Shera controller also has a hold mode.  This
mode will freeze the DAC output at its current voltage.  So when its
engaged with a rubidium oscillator,  your rubidium becomes a normal
unit, except it has been "super tuned" to UTC !  Short term specs return
to normal, etc.

Somebody asked a question about the GPS sawtooth bridge and the
controller.  If the controller detects a problem during the measurement
period ( and remember it looks at the last value, the present value and
the future value), it will throw out that solution, as it is not valid.
It holds the last used value in the DAC.

There was this post a while back for a RFG:

http://www.febo.com/pipermail/time-nuts/2005-October/019674.html

But I could of sworn I found a more complete pinout elsewhere. I'm about to
search through my browser history to see if I can find the other URL(s) with
info.

Jason

On Wed, Dec 20, 2006 at 10:58:02AM -0600, Jason Rabel wrote:

I think you just steered me down a blind alley...

I grabbed the last two of these at the newer higher price of

$185 BIN but more careful searching shows photos of a box without any
evident GPS antenna input.

I think the part with the GPS is the RFTG-M-XO which contains a

disciplined OXCO with optional rubidium input from the RFTG-M-RB box.

I suppose there must be a EFC voltage input or, much more

useful, a 1 PPS input that syncs the Rb.  But they appear to be a pair.

I suppose that isn't too bad a price for a Rb oscillator if they

have reasonably low hours and are the long lived telco kind...

And a Rb oscillator with a PLL that expects a 1 PPS might be quite

useful at that price.

--
Dave Emery N1PRE,  die@dieconsulting.com  DIE Consulting, Weston, Mass 02493
"An empty zombie mind with a forlorn barely readable weatherbeaten
'For Rent' sign still vainly flapping outside on the weed encrusted pole - in
celebration of what could have been, but wasn't and is not to be now either."

Oh fudgknockers.... I could have sworn I saw RB models with the GPS antenna
inputs, but looking at the few pictures I can find, I think you are right.
Please don't burn me at the stake, at least not until after Christmas.

I sincerely apologize, I honestly thought they had the antenna input, but I
guess it was the RFTG XO that has it, and the RB ties into it. For the $130
BIN (before the raised the price), it seemed like a great deal even just for
the Rb unit.

It has a 1PPS label on the front, I do not know if it is an input or output.
I'm going to cross my fingers and hope it is an input.

The RB unit itself is a "special" FE-5680A and the life is 20 years. (That
is what the FEI guy said.) I know these Lucent modules were made for mobile
/ temporary / emergency cell sites, so they probably have not had much use.

http://www.freqelec.com/pdf/12pager.pdf

http://www.gillam-fei.be/products/pdf/others/rubidium/FEI%20-%205680A.pdf

I did also see this on eBay a while back, but I think the DDS is a special
option. Still good info to have on hand just incase:

eBay Item number: 160048938286 - $960
eBay Item number: 160062108623 - $515

Jason (Don't Kill Me) Rabel