FPGA Central

This is slightly off-topic:
I have mentioned before that I am in the process of designing (and
manufacturing in limited volume) a clock-generator box (1 Hz to 1.5
GHz in 1 Hz increments). We debated some jitter issues...
It's tough to get a very accurate reference xtal (2 ppm should be
possible, 20 ppm is easy)
I might include a simple calibration mechanism (any reasonable amount
of logic is free in the Virtex-5 FPGA), but from where do I get a very
accurate reference frequency input of any value ?

Decades ago, the major TV stations kept the color subcarrier very
accurate (much better than 0.01 ppm), but that service has ben
abandoned. WWV needs a short-wave receiver, or at least a good
selective amplifier.
I am asking this smart group for some creative ideas. GPS ?
I indicate the frequency on a 9-digit display, so it would be
desirable to be able to guarantee better than 1 ppm (after calibration
for at least a few days. Temperature drift is not a big issue in a lab
instrument...)
Peter Alfke

>Decades ago, the major TV stations kept the color subcarrier very
>accurate (much better than 0.01 ppm), but that service has ben
>abandoned...

Some radio carriers are maintained to very high accuracy for use as a
frequency reference. E.g. the BBC long wave transmission (previously
200kHz, now 198kHz) is controlled by a rubidium clock. The audio
modulation is AM but there is also a low-bandwidth phase modulation
used to broadcast data to electricity meters (for control of overnight
load-shedding). I don't know if that phase modulation would be a
problem for your application.

I'd guess that the signal is useable over all of western Europe. I
recall that when I stayed about 200km east of Moscow, I had forgotten
how to set my radio-controlled alarm clock to run only from the
internal crystal, but in the morning it had set itself from the 60kHz
transmitter in the UK.

Perhaps there is something similar in the USA.

http://en.wikipedia.org/wiki/Droitwi...itting_station
http://en.wikipedia.org/wiki/Radio_teleswitch

Mike

On Fri, 18 Jan 2008 14:49:46 -0800 (PST),
Peter Alfke <[email protected]> wrote:


>I might include a simple calibration mechanism (any reasonable amount
>of logic is free in the Virtex-5 FPGA), but from where do I get a very
>accurate reference frequency input of any value ?

I don't know what happens in the USA, but here in the UK we have
a timecode broadcast service (MSF) that provides a very stable
60kHz carrier, modulated with some timecode stuff that keeps
clocks up-to-date. Because the frequency is so low, coverage
is excellent and receivers very simple. If you're happy to
do the calibration over many hours, that might be all you need.
An A-D converter and synchronous demodulation sounds like an
afternoon's work for you guys - you could use it as a demo
of a high-resolution delta-sigma A/D... only a ferrite-rod
antenna needed. Add a temperature sensor near your oscillator
and then you could, over time, calibrate its temperature
coefficient, enabling you to keep it very accurate even if
you lose the time code signal.

I can easily imagine such a thing being subverted by nearby
CRT monitors or other sources of spurious-ness with components
close to 60kHz. I still carry the scars of attempts to build
readers for the TIRIS RF-ID tags, which worked at around 130kHz
and were pretty much inductively coupled. That frequency was
nicely at the second harmonic of the line scan rate of VGA
monitors, which made the whole thing fail horribly if it
was within a couple of yards of any cheap-and-nasty monitor.

Still, you could use synchronous demodulation to establish
a *very* narrow receiver bandwidth.
--
Jonathan Bromley, Consultant

DOULOS - Developing Design Know-how
VHDL * Verilog * SystemC * e * Perl * Tcl/Tk * Project Services

Doulos Ltd., 22 Market Place, Ringwood, BH24 1AW, UK
[email protected]
http://www.MYCOMPANY.com

The contents of this message may contain personal views which
are not the views of Doulos Ltd., unless specifically stated.

Jonathan Bromley <[email protected]> wrote:
> On Fri, 18 Jan 2008 14:49:46 -0800 (PST),
> Peter Alfke <[email protected]> wrote:


> >I might include a simple calibration mechanism (any reasonable amount
> >of logic is free in the Virtex-5 FPGA), but from where do I get a very
> >accurate reference frequency input of any value ?

> I don't know what happens in the USA, but here in the UK we have
> a timecode broadcast service (MSF) that provides a very stable
> 60kHz carrier, modulated with some timecode stuff that keeps
> clocks up-to-date.

In the US the same service comes from WWVB in colorado:

http://tf.nist.gov/stations/wwvb.htm

and clock devices that use it as a time reference are fairly common,
though these typically only preform a time sync once a day (typically
at night when propagation is better). In fact I'm wearing a Casio
watch that's both solar powered and gets the time nightly from WWVB
so I never have to think about it and I always have sub-second accurate
time on my wrist.

But I think GPS is generally the right solution for this problem these
days. There's a whole sub-genre of the GPS community (and devices from
manufacturers) that only care about when rather than where. Once you
get out of the consumer stuff, receivers that output a very accurate
1 pps (pulse per second) are common, and you can get any level of
exotic time keeping beyond this that you have the money to pay for :-)

G.

On Jan 18, 4:49 pm, Peter Alfke <[email protected]> wrote:
> This is slightly off-topic:
> I have mentioned before that I am in the process of designing (and
> manufacturing in limited volume) a clock-generator box (1 Hz to 1.5
> GHz in 1 Hz increments). We debated some jitter issues...
> It's tough to get a very accurate reference xtal (2 ppm should be
> possible, 20 ppm is easy)
> I might include a simple calibration mechanism (any reasonable amount
> of logic is free in the Virtex-5 FPGA), but from where do I get a very
> accurate reference frequency input of any value ?
>
> Decades ago, the major TV stations kept the color subcarrier very
> accurate (much better than 0.01 ppm), but that service has ben
> abandoned. WWV needs a short-wave receiver, or at least a good
> selective amplifier.
> I am asking this smart group for some creative ideas. GPS ?
> I indicate the frequency on a 9-digit display, so it would be
> desirable to be able to guarantee better than 1 ppm (after calibration
> for at least a few days. Temperature drift is not a big issue in a lab
> instrument...)
> Peter Alfke

You can get reference clocks that are locked to GPS. Here are a few
links:

http://ngn.symmetricom.com/products/...rces/index.asp
http://www.trimble.com/tmg_thunderbolt.shtml
http://cgi.ebay.com/Thunderbolt-GPS-...QQcmdZViewItem


If you want to Google for more, search on "primary reference clock" or
"telcom primary base reference".

It is pretty common to find them with a 10MHz reference clock output,
IRIG time code, and a one PPS signal. They discipline an internal
oscillator to the GPS clock, which is locked to an atomic clock. 10
MHz is also a common reference clock for frequency generators.



Regards,

John McCaskill
www.FasterTechnology.com

On 18 Jan., 23:49, Peter Alfke <[email protected]> wrote:
> This is slightly off-topic:
> I have mentioned before that I am in the process of designing (and
> manufacturing in limited volume) a clock-generator box (1 Hz to 1.5
> GHz in 1 Hz increments). We debated some jitter issues...
> It's tough to get a very accurate reference xtal (2 ppm should be
> possible, 20 ppm is easy)

Peter,

for our time to digital converters we use temperature compensated
oscillators (TCXOs) with 1 ppm.
We pay a premium because we have custom frequencies cut for us, but
for standard frequencies
in higher quantities the price should be below 10$.
These devices are available from many manucatures in 7mm x 5mm SMD.

I recently saw a new TCXO that only has 100ppb temperature drift.

Below that you need OCXOs. The packages of these are a bit larger,
they are more expensive
and draw quite a lot of current (500mW to 5W). But you can get them
down to 5ppb.

Another option is to lock the clock to GPS. There are boxed solutions
for that available.

Have fun,

Kolja
www.cronologic.de

On Jan 18, 2:49*pm, Peter Alfke <[email protected]> wrote:
> This is slightly off-topic:
> I have mentioned before that I am in the process of designing (and
> manufacturing in limited volume) a clock-generator box (1 Hz to 1.5
> GHz in 1 Hz increments). We debated some jitter issues...
> It's tough to get a very accurate reference xtal (2 ppm should be
> possible, 20 ppm is easy)
> I might include a simple calibration mechanism (any reasonable amount
> of logic is free in the Virtex-5 FPGA), but from where do I get a very
> accurate reference frequency input of any value ?
>
> Decades ago, the major TV stations kept the color subcarrier very
> accurate (much better than 0.01 ppm), but that service has ben
> abandoned. WWV needs a short-wave receiver, or at least a good
> selective amplifier.
> I am asking this smart group for some creative ideas. GPS ?
> I indicate the frequency on a 9-digit display, so it would be
> desirable to be able to guarantee better than 1 ppm (after calibration
> for at least a few days. Temperature drift is not a big issue in a lab
> instrument...)
> Peter Alfke

On Jan 18, 2:49*pm, Peter Alfke <[email protected]> wrote:
> This is slightly off-topic:
> I have mentioned before that I am in the process of designing (and
> manufacturing in limited volume) a clock-generator box (1 Hz to 1.5
> GHz in 1 Hz increments). We debated some jitter issues...
> It's tough to get a very accurate reference xtal (2 ppm should be
> possible, 20 ppm is easy)
> I might include a simple calibration mechanism (any reasonable amount
> of logic is free in the Virtex-5 FPGA), but from where do I get a very
> accurate reference frequency input of any value ?
>
> Decades ago, the major TV stations kept the color subcarrier very
> accurate (much better than 0.01 ppm), but that service has ben
> abandoned. WWV needs a short-wave receiver, or at least a good
> selective amplifier.
> I am asking this smart group for some creative ideas. GPS ?
> I indicate the frequency on a 9-digit display, so it would be
> desirable to be able to guarantee better than 1 ppm (after calibration
> for at least a few days. Temperature drift is not a big issue in a lab
> instrument...)
> Peter Alfke

(I wish I didn't have to use google groups to access from work...)

Peter,

Your own austin has quite a background in stable timing so he can
probably provide some good lunchtime conversation. The specs I've
seen on GPS timing references give 1 pulse per second outputs accurate
to within 100 ns. Perhaps this accuracy is better than those many
years ago when I was actively looking at the specs but it gives you an
idea of the accuracies you'd need to work with. GPS-trained frequency
sources use clean local oscillators to smooth out the uncertainty and
provide good accuracy under signal dropout conditions. For very high
accuracy phase stuff, the Allan variance can come into play (again,
seek guidance from austin).

It may be that for measurement accuracy, the low Allan variance isn't
a necessity; I don't have an appreciation for the scale of the
problem, only the problem itself. Heck - even rubidium oscillators
have close-in phase noise issues that are averaged out with external
help. Just this week I've been demodulating jitter and watching cheap
oscillators changing frequencies on a whim, changing from one
relatively stable value to the next.

The jitter generator I produced a decade ago went with a small OCXO
from www.mti-milliren.com that had nice close-in phase noise
characteristics within a reasonably small package but it was still a
$50 instrumentation solution.

Both a GPS reference and an OCXO will probably be a chunk of the
device cost and consume more power than you'd want for batteries.

A quick discussion with an apps engineer from someplace like MTI might
get you more precise information that from the lovely engineers that
frequent this group.

Let us know when we can order one of your devices!
- John_H

"Peter Alfke" <[email protected]> wrote in message
news:76bcd2dd-330e-47c8-9813-00fea9a4d6d1@v67g2000hse.googlegroups.com...
> Decades ago, the major TV stations kept the color subcarrier very
> accurate (much better than 0.01 ppm), but that service has ben
> abandoned. WWV needs a short-wave receiver, or at least a good
> selective amplifier.
> I am asking this smart group for some creative ideas. GPS ?
> I indicate the frequency on a 9-digit display, so it would be
> desirable to be able to guarantee better than 1 ppm (after calibration
> for at least a few days. Temperature drift is not a big issue in a lab
> instrument...)

Other posters have mentioned the excellent GPS-based solutions available,
and/or WWV(B) options. Here are two others:
1) The TV carriers are still often dead-on in frequency, as they are now
locked to GPS.
2) The Loran-C system (100 kHz pulses) is being recapitalized with precise
(GPS-locked) references as a backup to GPS against potential jamming.

I've been very happy with my Symmetricom frequency references; for a
small/cheap solution, SigNav is now selling a GPS receiver based on the old
Motorola Oncore series (interface compatible) that provides a precise 10MHz
output. Trimble also sells something called the Micro-T with similar
capability. Both I believe are in the ~$200 range for an OEM, but prices
have been dropping. When locked to GPS, these jobbies can give you better
than 10 ppb accuracy.

Marty

On Jan 18, 8:02*pm, "Marty Ryba" <[email protected]>
wrote:
> "Peter Alfke" <[email protected]> wrote in message
>
> news:76bcd2dd-330e-47c8-9813-00fea9a4d6d1@v67g2000hse.googlegroups.com...
>
> > Decades ago, the major TV stations kept the color subcarrier very
> > accurate (much better than 0.01 ppm), but that service has ben
> > abandoned. WWV needs a short-wave receiver, or at least a good
> > selective amplifier.
> > I am asking this smart group for some creative ideas. GPS ?
> > I indicate the frequency on a 9-digit display, so it would be
> > desirable to be able to guarantee better than 1 ppm (after calibration
> > for at least a few days. Temperature drift is not a big issue in a lab
> > instrument...)
>
> Other posters have mentioned the excellent GPS-based solutions available,
> and/or WWV(B) options. Here are two others:
> 1) The TV carriers are still often dead-on in frequency, as they are now
> locked to GPS.
> 2) The Loran-C system (100 kHz pulses) is being recapitalized with precise
> (GPS-locked) references as a backup to GPS against potential jamming.
>
> I've been very happy with my Symmetricom frequency references; for a
> small/cheap solution, SigNav is now selling a GPS receiver based on the old
> Motorola Oncore series (interface compatible) that provides a precise 10MHz
> output. Trimble also sells something called the Micro-T with similar
> capability. Both I believe are in the ~$200 range for an OEM, but prices
> have been dropping. When locked to GPS, these jobbies can give you better
> than 10 ppb accuracy.
>
> Marty

Thank you, guys, for all the advice arriving in just a few hours.
I remember Droitwitch 200 kHz from my parents' (highly illegal)
listening to the BBC during WW2.
It covered most of Germany (!) very well, but is obviously silent here
in the US.
The 60 kHz timing transmissions in Europe and here from Fort Collins,
Colorado are rather slow, and burdened with their data modulation.
I like the GPS-based solutions from SigNav and Trimble best.
I have no intentions of burdening the box with a >>$10 precision
oscillator, I rather provide a separate calibration capability, for
manufacturing and for ultra-demanding users, where a few hundred
dollars do not matter.
The nicest solution would be over the internet, but I have not heard
about anything like that (yet).
I will keep you informed about our progress.
Peter Alfke

On Jan 18, 4:49 pm, Peter Alfke <[email protected]> wrote:
> This is slightly off-topic:
> I have mentioned before that I am in the process of designing (and
> manufacturing in limited volume) a clock-generator box (1 Hz to 1.5
> GHz in 1 Hz increments). We debated some jitter issues...
> It's tough to get a very accurate reference xtal (2 ppm should be
> possible, 20 ppm is easy)
> I might include a simple calibration mechanism (any reasonable amount
> of logic is free in the Virtex-5 FPGA), but from where do I get a very
> accurate reference frequency input of any value ?
>
> Decades ago, the major TV stations kept the color subcarrier very
> accurate (much better than 0.01 ppm), but that service has ben
> abandoned. WWV needs a short-wave receiver, or at least a good
> selective amplifier.
> I am asking this smart group for some creative ideas. GPS ?
> I indicate the frequency on a 9-digit display, so it would be
> desirable to be able to guarantee better than 1 ppm (after calibration
> for at least a few days. Temperature drift is not a big issue in a lab
> instrument...)
> Peter Alfke


Here is one more interesting link:

http://www.leapsecond.com/

Regards,

John McCaskill
www.FasterTechnology.com

>Here is one more interesting link:
>http://www.leapsecond.com/

Beware. That's much more than just an "interesting link".

It's a gigantic time sink.

--
These are my opinions, not necessarily my employer's. I hate spam.

In article <76bcd2dd-330e-47c8-9813-00fea9a4d6d1@v67g2000hse.googlegroups.com>,
Peter Alfke <[email protected]> writes:

>This is slightly off-topic:
>I have mentioned before that I am in the process of designing (and
>manufacturing in limited volume) a clock-generator box (1 Hz to 1.5
>GHz in 1 Hz increments). We debated some jitter issues...
>It's tough to get a very accurate reference xtal (2 ppm should be
>possible, 20 ppm is easy)

>I might include a simple calibration mechanism (any reasonable amount
>of logic is free in the Virtex-5 FPGA), but from where do I get a very
>accurate reference frequency input of any value ?

My two cents worth...

I assume your target market is pretty broad, some combination
of professionals/geeks with serious cash, hackers with time but no cach,
and professors/students and whatever in between.


A connector for an external clock is the obvious first step.
That lets you use whatever is available at the local environment.

Serious labs will have 10 MHz available from the wall next to the
110 V AC and ethernet.

Lots of lab equipment (for example counters) have a BNC jack
on the back for input of a reference clock andor output of
their internal oscillator. That sort of lab gear has good to
very good internal oscillators.

That's all pretty fancy/expensive. But it lets the geeks go
for the bleeding edge.


Most low cost digital crystal oscillator packages have 2
interesting error components. One is manufacturering offset.
The other is temperature. (Supply voltage is probably third,
but I'm not calibrated on that.)

The ballpark for tempertaure is 1 ppm per C. The ballpark for
manufacturing (at room temp) is 1/2 of the spec.

I'd expect you could get pretty close to a few ppm if you did
a calibration run at a known temperature on your final checkout
line and recorded the fudge factor.

You could do even better if you included a temperature sensor
near the local oscillator package and made a few test runs to get
a simple linear correction. (Maybe non-linear via table lookup
would be better. Mumble, TBD.)


I don't think you will get much from WWVB or WWV and friends
without a stable local clock to use as a refernce. They are
great for the long term, but you need a place to stand.


As others have suggested, ask Austin for high end ideas.

DDS type approaches are notorious for close-in spurs.

I think they are not interesting for longer term measurements.
(But lots of people are interestedin short term measurements.)



GPS is a good (low cost) straw man.

You can get GPS receivers with a PPS (pulse per second) output
for under $100. They are basically low cost commerical units
with an extra output pin.

You can get various boxes with a good crystal, a GPS input,
a pile of software, and ... They tend to be expensive
on the hobbyist scale.

An example:
http://jackson-labs.com/products_fury.html

Symmetricom (by way of Agilent) has Borged HP's old Cesium
clock business area and friends and lots of others.
http://www.symmetricom.com/Products
(I didn't run into their Flash crap until I got there.)


GPSDO is a good magic word: GPS Disciplined Oscillator.

GPS provides the long term stability. The local (good
crystal) provides the short and medium stability. Usually
there is a lot of software filtering involved.

HP pioneered that technology many years ago.

For more than you wanted to know:

http://www.hpl.hp.com/hpjournal/96dec/dec96a9.pdf
The Global Positioning System and HP SmartClock
December 1996 Hewlett-Packard Journal

http://www.realhamradio.com/GPS_Frequency_Standard.htm
(Z3801As were available cheap for a while as cell towers upgraded.)

HP app notes:
HP SmartClock Technology Application Note 1279
The Science of Timekeeping Application Note 1289
GPS and Precision Timing Applications Application Note 1272
(Ask me if google can't find them for you.)

Just for fun, and this needs a serious time-sink warning,
here is my all time favorite paper about time:
http://www.physicstoday.org/vol-59/iss-3/p10.html
Daniel Kleppner
Physics Today, March 2006, page 10
[Don't say I didn't warn you.]

--
These are my opinions, not necessarily my employer's. I hate spam.

"Peter Alfke" <[email protected]> wrote in message
news:76bcd2dd-330e-47c8-9813-00fea9a4d6d1@v67g2000hse.googlegroups.com...
> I am asking this smart group for some creative ideas. GPS ?
> I indicate the frequency on a 9-digit display, so it would be
> desirable to be able to guarantee better than 1 ppm (after calibration
> for at least a few days. Temperature drift is not a big issue in a lab
> instrument...)
> Peter Alfke

Hi Peter,

I don't claim to be smart or creative, but some TCXOs are good for 1ppm and
not too expensive. If you can calibrate them after manufacture, which I
guess you can easily do with your small batch, and store the offset error in
your device, I bet you can easily keep a TCXO at the same frequency to a few
100ppb, if you don't change the temperature too much. I guess you could even
use a IC thermometer (Maxim make some cheap'n'cheerful ones, maybe with
nvrom to store offsets?) and adjust the cal accordingly.

My experience with TCXOs is just that; the initial accuracy is somewhere
within the spec., say +/-2ppm, but the frequency stays at that initial
frequency from then on (> a few days!). (A whack with a hammer can sometimes
'adjust' the initial frequency!)

Of course, OCXOs are good for a few ppb.
http://foxonline.com/pdfs/fts301ah.pdf

GPS is great, if you're outdoors. How many lab instruments meet this
requirement? ;-) Cabling an aerial can be a PITA.

HTH., Syms.

Actually, forget all that, use this:-
http://www.nist.gov/public_affairs/r.../miniclock.htm

comp.arch.fpga <[email protected]> wrote:
> On 18 Jan., 23:49, Peter Alfke <[email protected]> wrote:
> > This is slightly off-topic:
> > I have mentioned before that I am in the process of designing (and
> > manufacturing in limited volume) a clock-generator box (1 Hz to 1.5
> > GHz in 1 Hz increments). We debated some jitter issues...
> > It's tough to get a very accurate reference xtal (2 ppm should be
> > possible, 20 ppm is easy)

> Peter,

> for our time to digital converters we use temperature compensated
> oscillators (TCXOs) with 1 ppm.
> We pay a premium because we have custom frequencies cut for us, but
> for standard frequencies
> in higher quantities the price should be below 10$.
> These devices are available from many manucatures in 7mm x 5mm SMD.

> I recently saw a new TCXO that only has 100ppb temperature drift.

> Below that you need OCXOs. The packages of these are a bit larger,
> they are more expensive
> and draw quite a lot of current (500mW to 5W). But you can get them
> down to 5ppb.

While the OCXO have drifts down to 20 ppb, the default accurracy is only in
the ppm range...

--
Uwe Bonnes [email protected]

Institut fuer Kernphysik Schlossgartenstrasse 9 64289 Darmstadt
--------- Tel. 06151 162516 -------- Fax. 06151 164321 ----------

Peter Alfke wrote:

> I might include a simple calibration mechanism (any reasonable amount
> of logic is free in the Virtex-5 FPGA), but from where do I get a very
> accurate reference frequency input of any value ?

IMHO you should simply provide an external clock input
(perhaps of any configurable value, you have some spare
cells in your FPGA...) and let the user supply his own
reference clock. Which could be, for instance, a rubidium
atomic oscillator:

http://cgi.ebay.com/RUBIDIUM-SYMMETR...QQcmdZViewItem

for 260 USD (dollar is cheap now...). AFAIK there
are smaller models on the market.

Best regards
Piotr Wyderski

>While the OCXO have drifts down to 20 ppb, the default accurracy is only in
>the ppm range...

But many/most of them that are that stable have a voltage adjust input
pin. A good regulator and a pot lets you manually adjust it. A DAC
lets you servo it to something else, for example GPS. That's a good
first step down a very long steep slippery slope.

--
These are my opinions, not necessarily my employer's. I hate spam.

Peter Alfke wrote:
<snip>
> I have no intentions of burdening the box with a >>$10 precision
> oscillator, I rather provide a separate calibration capability, for
> manufacturing and for ultra-demanding users, where a few hundred
> dollars do not matter.
<snip>

How about a 2ppm accuracy reference for $2.55 as a first step? The
DS3231 is one of several 32.768kHz Maxim oscillator modules.

http://www.maxim-ic.com/quick_view2.cfm/qv_pk/4627

Peter Alfke wrote:
> This is slightly off-topic:
> I have mentioned before that I am in the process of designing (and
> manufacturing in limited volume) a clock-generator box (1 Hz to 1.5
> GHz in 1 Hz increments). We debated some jitter issues...
> It's tough to get a very accurate reference xtal (2 ppm should be
> possible, 20 ppm is easy)
> I might include a simple calibration mechanism (any reasonable amount
> of logic is free in the Virtex-5 FPGA), but from where do I get a very
> accurate reference frequency input of any value ?
>
> Decades ago, the major TV stations kept the color subcarrier very
> accurate (much better than 0.01 ppm), but that service has ben
> abandoned. WWV needs a short-wave receiver, or at least a good
> selective amplifier.
> I am asking this smart group for some creative ideas. GPS ?
> I indicate the frequency on a 9-digit display, so it would be
> desirable to be able to guarantee better than 1 ppm (after calibration
> for at least a few days. Temperature drift is not a big issue in a lab
> instrument...)

Hi Peter,
The Carrier precisions vary around the world, so I'd allow a flexible
calibration
scheme.

If you implement a reciprocal counter, then users can choose what they
like,
as reciprocal counters give appx constant precision-per-second

Most globaly common these days, is the 1pps on GPS, which gives appx
100ns,
or one part in 10^7 per pulse, so you'd need 1000 pulses to get 10^10

A tougher question is how does the user correct the result?. If you
use just a
vanilla crystal/module, they are not easy to trim. A VCXO would allow
a lock system
with a simple DAC.

Even nicer would be a dual-input system, (would need Dual reciprocal
counters, but
hey, this is a big FPGA! ) so a user could ratio-lock/check almost any
two
reference sources (as well as the local crystal).

9 Digits sounds ok for a per-second readout, but users might want to
run to 100s
or 1000s timebases, for highest precision, and 9 digits is light
then

Or, you could look for one of these!!

http://www.eurekalert.org/pub_releas...-oon090203.php
"The Ultra-miniature Rubidium (Rb) Atomic Clock, 40 cubic centimeters
in volume and using a minuscule one watt of power, doesn't weigh much
more than a matchbox either. And... it will lose only about one second
every 10,000 years."

-jg

On Jan 19, 2:25*pm, -jg <[email protected]> wrote:
> Peter Alfke wrote:
> > This is slightly off-topic:
> > I have mentioned before that I am in the process of designing (and
> > manufacturing in limited volume) a clock-generator box (1 Hz to 1.5
> > GHz in 1 Hz increments). We debated some jitter issues...
> > It's tough to get a very accurate reference xtal (2 ppm should be
> > possible, 20 ppm is easy)
> > I might include a simple calibration mechanism (any reasonable amount
> > of logic is free in the Virtex-5 FPGA), but from where do I get a very
> > accurate reference frequency input of any value ?
>
> > Decades ago, the major TV stations kept the color subcarrier very
> > accurate (much better than 0.01 ppm), but that service has ben
> > abandoned. WWV needs a short-wave receiver, or at least a good
> > selective amplifier.
> > I am asking this smart group for some creative ideas. GPS ?
> > I indicate the frequency on a 9-digit display, so it would be
> > desirable to be able to guarantee better than 1 ppm (after calibration
> > for at least a few days. Temperature drift is not a big issue in a lab
> > instrument...)
>
> Hi Peter,
> The Carrier precisions vary around the world, so I'd allow a flexible
> calibration
> scheme.
>
> If you implement a reciprocal counter, then users can choose what they
> like,
> as reciprocal counters give appx constant precision-per-second
>
> Most globaly common these days, is the 1pps on GPS, which gives appx
> 100ns,
> or one part in 10^7 per pulse, so you'd need *1000 pulses to get 10^10
>
> A tougher question is how does the user correct the result?. If you
> use just a
> vanilla crystal/module, they are not easy to trim. A VCXO would allow
> a lock system
> with a simple DAC.
>
> Even nicer would be a dual-input system, (would need Dual reciprocal
> counters, but
> hey, this is a big FPGA! ) so a user could ratio-lock/check almost any
> two
> reference sources (as well as the local crystal).
>
> 9 Digits sounds ok for a per-second readout, but users might want to
> run to 100s
> or 1000s timebases, for highest precision, and 9 digits is light
> then
>
> Or, you could look for one of these!!
>
> http://www.eurekalert.org/pub_releas...-oon090203.php
> "The Ultra-miniature Rubidium (Rb) Atomic Clock, 40 cubic centimeters
> in volume and using a minuscule one watt of power, doesn't weigh much
> more than a matchbox either. And... it will lose only about one second
> every 10,000 years."
>
> -jg

I really appreciate all the inputs.
Here is where I stand today.

We have not yet finished the prototype (got delayed by OLED interface
issues, and also explored more creative ways to reduce jitter and
wander), but that's now behind us.
We generate 2 output channels (since it is almost for free) and have a
dual-16 character (really: 128 x 64 pixel) display. We will use a
312.5 MHz xtal oscillator that promises single-digit ppm accuracy.
For the first dozen prototypes I will ignore the accuracy question,
since the average user seems to be content with 20 ppm. But we do not
want to remain "average"...
We will provide for a calibration input, and also for the non-volatile
storage of calibration factors (We already store hundreds of
frequencies for recall, again because it is almost for free).
Calibration will be by arithmetic number crunching, nothing analog
here. We don't change the internal clock frequency, we just change the
DDS/PLL interpretation.
For the later calibration strategy, we use the fact that we can adjust
our signal output to any frequency with 1 Hz granularity. So we can
match any comparison frequency "anywhere in the world".

We need a Virtex-5 LXT50 chip, due to the way we (ab)use the GTPs, but
that gives us an enormous amount of spare logic, RAMs and multipliers.
There is also a chip thermometer with digital read-out on every
Virtex-5 die. It really is an embarrassment of riches. And we can fine-
tune the design as often as we have time and patience.
Aren't FPGAs fun !
Thanks again. I have a few things to read and study.
Peter Alfke

>The nicest solution would be over the internet, but I have not heard
>about anything like that (yet).

If you have a stable local clock, you can calibrate it
over the net. Use NTP - Network Time Protocol.

If you have a T1/T3 link, the bit rate on that is usually
very accurate.

--
These are my opinions, not necessarily my employer's. I hate spam.

"Hal Murray" <[email protected]> wrote in message
news:[email protected]...
>
>
> If you have a T1/T3 link, the bit rate on that is usually
> very accurate.
>
I wouldn't rely on that. T1 is only required to be Stratum 4 compliant,
which is 32 ppm accuracy.

Peter Alfke wrote:
> We will use a
> 312.5 MHz xtal oscillator that promises single-digit ppm accuracy.
> For the first dozen prototypes I will ignore the accuracy question,
> since the average user seems to be content with 20 ppm. But we do not
> want to remain "average"...

You could maybe allow for something simple like maxim's DS32KHz which
comes straight out of the box at 2ppm over 0-40'C, or the DS3231
which
has more trim options, but may over-complicate the details.
Maxim also have a DS76KHz, but that's not so widely stocked.

-jg

>Maxim also have a DS76KHz, but that's not so widely stocked.

What's interesting about 76.8 KHz? Who uses that, or some
(sub)multiple of it?

--
These are my opinions, not necessarily my employer's. I hate spam.