Hi, Bob --

N-cornered hat measurement is a beta feature, not in the docs yet.  A single TimePod can generate the required three plots simultaneously, but you can use individual plots from TICs and frequency counters as long as you start with highly-reproducible measurements and are careful with your interpretation of the results.  If your setup returns inconsistent results from one run to the next, then any attempt at a separated-variance plot will be GIGO.

You can see some examples at http://www.ke5fx.com/gpscomp.htm .  Various finicky conditions need to be met before the 3-cornered hat display will become available; check out https://www.febo.com/pipermail/time-nuts/2015-September/094039.html for some tips.

-- john, KE5FX
Miles Design LLC

Hi John,
Thanks!  With lesser equipment, such as the 5370A, would longer runs be better?  I used a set of 1 hr runs and the result wasn't quite what I had expected.  However, it may be that I had mislabeled the files, and thus got the sources confused.  Of course, it may be that the ionospheric effect was grossly different between the three tests.  So, with a 5370, Source A would be the START input and Source B would be the STOP input, right?  For my testing, the sources are all 10MHz signals, and I'm driving the EXT input with 1PPS from a GPSDO.

Bob

  From: John Miles <john@miles.io>

To: 'Discussion of precise time and frequency measurement' time-nuts@febo.com
Sent: Thursday, April 13, 2017 1:07 AM
Subject: Re: [time-nuts] Three-cornered hat on timelab?

Hi, Bob --

N-cornered hat measurement is a beta feature, not in the docs yet.  A single TimePod can generate the required three plots simultaneously, but you can use individual plots from TICs and frequency counters as long as you start with highly-reproducible measurements and are careful with your interpretation of the results.  If your setup returns inconsistent results from one run to the next, then any attempt at a separated-variance plot will be GIGO.

You can see some examples at http://www.ke5fx.com/gpscomp.htm .  Various finicky conditions need to be met before the 3-cornered hat display will become available; check out https://www.febo.com/pipermail/time-nuts/2015-September/094039.html for some tips. 

-- john, KE5FX
Miles Design LLC

time-nuts mailing list -- time-nuts@febo.com
To unsubscribe, go to https://www.febo.com/cgi-bin/mailman/listinfo/time-nuts
and follow the instructions there.

Hi

There are a number of papers from the 70’s and 80’s digging into three corner hat data. The net result often
would turn out to be “less than zero” noise on one of the DUT’s. Since that’s physically impossible the technique
got a bit of “attention”.  The Cliff Notes version of the results is that simultaneous measurements were the key
to getting decent results. The closer to “same time” (as in microseconds or nanoseconds) the better. Even with very careful
data collection, odd things can still happen. Phase noise pops up at crazy low levels or ADEV goes to bizarre
numbers. In many ways a TimePod (or other ADC based setup) is ideal for getting the data synchronized. Running
all three devices on one is by far the best way I have seen to make the technique work. It still can have problems,
but less so that other ways of doing it.

Bob

Hi Bob,
OK, I give up.  Try as I might, I can't fudge things enough to make any sense.  I did run a set of 1 hour tests that seemed to confirm what I can infer from the phase plots in timelab, but anything longer than that and the curves either contradict the phase plots, or there are large gaps in at least one trace, or a trace is even missing entirely.  Oh well.  I seem to remember reading that the 3c-hat was only useful in comparing similar devices.  The KS just isn't close enough to what I'm trying to compare it to, I guess.

If you or anyone else has an ADEV plot of the KS against some local standard (for any length of time, any standard, even just a bare OCXO that is not a Trimble 34310-T) could you please share it with me?  I'm looking for relative peformance, not a definitive test.  Of course if you also have one of a 34310-T against the same standard, that would be great!

Bob 

  From: Bob kb8tq <kb8tq@n1k.org>

To: Bob Stewart bob@evoria.net; Discussion of precise time and frequency measurement time-nuts@febo.com
Cc: John Miles john@miles.io
Sent: Thursday, April 13, 2017 7:19 PM
Subject: Re: [time-nuts] Three-cornered hat on timelab?

Hi

There are a number of papers from the 70’s and 80’s digging into three corner hat data. The net result often
would turn out to be “less than zero” noise on one of the DUT’s. Since that’s physically impossible the technique
got a bit of “attention”.  The Cliff Notes version of the results is that simultaneous measurements were the key
to getting decent results. The closer to “same time” (as in microseconds or nanoseconds) the better. Even with very careful
data collection, odd things can still happen. Phase noise pops up at crazy low levels or ADEV goes to bizarre
numbers. In many ways a TimePod (or other ADC based setup) is ideal for getting the data synchronized. Running
all three devices on one is by far the best way I have seen to make the technique work. It still can have problems,
but less so that other ways of doing it.

Bob

Hi,

This problem have since been addressed by Francois Vernotte as I saw in
a presentation last year. He addressed the issue somewhat different than
the traditional calculations to achieve the same problem.
Turns out that correlation eats your noise if I recall things right.

Cheers,
Magnus

On 04/14/2017 02:19 AM, Bob kb8tq wrote:

Hi

The data I have on the KS boxes was posted to the list back when they were.a hot topic. I’m sure it is still in the archives.
I’m guessing it’s not quite what you are after.

The closer the devices are to each other the better the technique works. A simple way to look at it is as an attenuation. If
it knocks noise down 10:1, the worst unit should be no more than 10X noise than the best unit. How much things are knocked
down is a function of the length of the runs compared to the longest tau. For a 10:1 ratio of tau to run, attenuation of noise by 10:1
is very optimistic. You usually  need something beyond 100:1 to get that sort of performance. A lot depends on the noise involved.
Some types of behavior simply don’t work well with the technique.

The KS box goes from “better than” to “worse than” and back to “better than” most atomic standards you would compare it to over
a range of tau from 0.1 S to 1,000,000 seconds. To get the 1,000,000 second data accurately, you would need a 100,000,000 second
run. The simple answer there is that nobody has that kind of time or that reliable a setup. Even the three month run to get good
100,000 second data is a challenge. None of that relates to three corner hat stuff, it’s just the confidence bars on ADEV. It gives
you another (say) 100:1 wait on top of the three corner stuff.

Now toss in the basics of GPS. Depending on the day, you will get <10 ns to  >100 ns swing over a  24 hour period. Today
may or may not be the same as tomorrow. That’s with a “perfect” L1 setup. The variation comes from the ionosphere and the fact that
the GPS data does not allow you to fully correct for it.  In addition, you will get some interesting bumps related to constellations and
your local antenna setup. Any GPSDO that is quartz based will happily follow the 24 hour swing in the GPS from the ionosphere. At
100,000 seconds, a 100 ns swing is 1x10^-12. That’s a lot of disruption. It most certainly is not the sort of thing that ADEV expects
to pop up in the middle of a run.

The simple answer to all this is “don’t go there”. Three corner hat is fine for short term stuff. It’s a mess for long term runs. Getting data
that is good enough for a long term ADEV run out of a three corner setup is a major struggle. The time for the correlation to knock down
the noise on top of the time to get good ADEV data gets you into impractically long runs.

Bob

Hi Bob,
Oh, I had completely forgotten about the many runs you gifted us with back then.  Fortunately, I kept all of them in my email archive.  I can't compare like for like, of course, but I think I can work up something that compares at the larger taus where the 5370 doesn't dominate.
I'm going to run another long term test of my GFS unit against my PRS-45A.  The problem, the issue that made me ask for data is that everything from phase plots to ADEV plots of my unit are just so much better than the KS.  In addition, I don't see the large ionospheric swings on my GFS unit that you and Bruce and others have spoken about.  This bothers me a lot.  Could it be my location here in Houston?  Could it be the Ublox LEA-6T compared to the much older Motorola in the KS?

Bob 

  From: Bob kb8tq <kb8tq@n1k.org>

To: Bob Stewart bob@evoria.net
Cc: Discussion of precise time and frequency measurement time-nuts@febo.com
Sent: Monday, April 17, 2017 7:55 AM
Subject: Re: [time-nuts] Three-cornered hat on timelab?

Hi
The data I have on the KS boxes was posted to the list back when they were.a hot topic. I’m sure it is still in the archives.I’m guessing it’s not quite what you are after. 

The closer the devices are to each other the better the technique works. A simple way to look at it is as an attenuation. If it knocks noise down 10:1, the worst unit should be no more than 10X noise than the best unit. How much things are knockeddown is a function of the length of the runs compared to the longest tau. For a 10:1 ratio of tau to run, attenuation of noise by 10:1 is very optimistic. You usually  need something beyond 100:1 to get that sort of performance. A lot depends on the noise involved. Some types of behavior simply don’t work well with the technique. 
The KS box goes from “better than” to “worse than” and back to “better than” most atomic standards you would compare it to overa range of tau from 0.1 S to 1,000,000 seconds. To get the 1,000,000 second data accurately, you would need a 100,000,000 secondrun. The simple answer there is that nobody has that kind of time or that reliable a setup. Even the three month run to get good100,000 second data is a challenge. None of that relates to three corner hat stuff, it’s just the confidence bars on ADEV. It givesyou another (say) 100:1 wait on top of the three corner stuff. 
Now toss in the basics of GPS. Depending on the day, you will get <10 ns to  >100 ns swing over a  24 hour period. Today may or may not be the same as tomorrow. That’s with a “perfect” L1 setup. The variation comes from the ionosphere and the fact thatthe GPS data does not allow you to fully correct for it.  In addition, you will get some interesting bumps related to constellations and your local antenna setup. Any GPSDO that is quartz based will happily follow the 24 hour swing in the GPS from the ionosphere. At 100,000 seconds, a 100 ns swing is 1x10^-12. That’s a lot of disruption. It most certainly is not the sort of thing that ADEV expects to pop up in the middle of a run. 
The simple answer to all this is “don’t go there”. Three corner hat is fine for short term stuff. It’s a mess for long term runs. Getting datathat is good enough for a long term ADEV run out of a three corner setup is a major struggle. The time for the correlation to knock downthe noise on top of the time to get good ADEV data gets you into impractically long runs. 
Bob

On Apr 16, 2017, at 10:16 PM, Bob Stewart bob@evoria.net wrote:
Hi Bob,
OK, I give up.  Try as I might, I can't fudge things enough to make any sense.  I did run a set of 1 hour tests that seemed to confirm what I can infer from the phase plots in timelab, but anything longer than that and the curves either contradict the phase plots, or there are large gaps in at least one trace, or a trace is even missing entirely.  Oh well.  I seem to remember reading that the 3c-hat was only useful in comparing similar devices.  The KS just isn't close enough to what I'm trying to compare it to, I guess.

If you or anyone else has an ADEV plot of the KS against some local standard (for any length of time, any standard, even just a bare OCXO that is not a Trimble 34310-T) could you please share it with me?  I'm looking for relative peformance, not a definitive test.  Of course if you also have one of a 34310-T against the same standard, that would be great!

Bob 

  From: Bob kb8tq <kb8tq@n1k.org>

To: Bob Stewart bob@evoria.net; Discussion of precise time and frequency measurement time-nuts@febo.com
Cc: John Miles john@miles.io
Sent: Thursday, April 13, 2017 7:19 PM
Subject: Re: [time-nuts] Three-cornered hat on timelab?

Hi

There are a number of papers from the 70’s and 80’s digging into three corner hat data. The net result often
would turn out to be “less than zero” noise on one of the DUT’s. Since that’s physically impossible the technique
got a bit of “attention”.  The Cliff Notes version of the results is that simultaneous measurements were the key
to getting decent results. The closer to “same time” (as in microseconds or nanoseconds) the better. Even with very careful
data collection, odd things can still happen. Phase noise pops up at crazy low levels or ADEV goes to bizarre
numbers. In many ways a TimePod (or other ADC based setup) is ideal for getting the data synchronized. Running
all three devices on one is by far the best way I have seen to make the technique work. It still can have problems,
but less so that other ways of doing it.

Bob

Hi

The ionosphere is the culprit in terms of the daily swing. The swing is a function of the goodness of fit between the GPS broadcast data
and the ionosphere as it impacts the satellites you are using. There is no rime or reason to it beyond that. If you get “lucky” things don’t move
much. If you live in exciting times, things move quite a bit. Unless you go to something like an L1/L2 receiver, the GPS module you use
has little to do with it (unless it’s broke ….). Yes there are some fiddly little qualifiers relating to being at the north or south pole and GPS
coverage (along with space weather impacts). Very few of us do our runs at either location :) Just for reference, the area of concern also has
at least one day each year where the sun sets for < 1 hours.

Bob

Hi Bob,
OK, thanks.  I've kicked off a 7 day run of a GFS against the PRS-45A.  That should be long enough to separate out the GFS from the PRS' drift direction from the ionosphere.

Bob 

  From: Bob kb8tq <kb8tq@n1k.org>

To: Bob Stewart bob@evoria.net
Cc: Discussion of precise time and frequency measurement time-nuts@febo.com
Sent: Monday, April 17, 2017 11:28 AM
Subject: Re: [time-nuts] Three-cornered hat on timelab?

Hi
The ionosphere is the culprit in terms of the daily swing. The swing is a function of the goodness of fit between the GPS broadcast dataand the ionosphere as it impacts the satellites you are using. There is no rime or reason to it beyond that. If you get “lucky” things don’t move much. If you live in exciting times, things move quite a bit. Unless you go to something like an L1/L2 receiver, the GPS module you use has little to do with it (unless it’s broke ….). Yes there are some fiddly little qualifiers relating to being at the north or south pole and GPS coverage (along with space weather impacts). Very few of us do our runs at either location :) Just for reference, the area of concern also hasat least one day each year where the sun sets for < 1 hours. 
Bob

On Apr 17, 2017, at 11:33 AM, Bob Stewart bob@evoria.net wrote:
Hi Bob,
Oh, I had completely forgotten about the many runs you gifted us with back then.  Fortunately, I kept all of them in my email archive.  I can't compare like for like, of course, but I think I can work up something that compares at the larger taus where the 5370 doesn't dominate.
I'm going to run another long term test of my GFS unit against my PRS-45A.  The problem, the issue that made me ask for data is that everything from phase plots to ADEV plots of my unit are just so much better than the KS.  In addition, I don't see the large ionospheric swings on my GFS unit that you and Bruce and others have spoken about.  This bothers me a lot.  Could it be my location here in Houston?  Could it be the Ublox LEA-6T compared to the much older Motorola in the KS?

Bob 

  From: Bob kb8tq <kb8tq@n1k.org>

To: Bob Stewart bob@evoria.net
Cc: Discussion of precise time and frequency measurement time-nuts@febo.com
Sent: Monday, April 17, 2017 7:55 AM
Subject: Re: [time-nuts] Three-cornered hat on timelab?

Hi
The data I have on the KS boxes was posted to the list back when they were.a hot topic. I’m sure it is still in the archives.I’m guessing it’s not quite what you are after. 

The closer the devices are to each other the better the technique works. A simple way to look at it is as an attenuation. If it knocks noise down 10:1, the worst unit should be no more than 10X noise than the best unit. How much things are knockeddown is a function of the length of the runs compared to the longest tau. For a 10:1 ratio of tau to run, attenuation of noise by 10:1 is very optimistic. You usually  need something beyond 100:1 to get that sort of performance. A lot depends on the noise involved. Some types of behavior simply don’t work well with the technique. 
The KS box goes from “better than” to “worse than” and back to “better than” most atomic standards you would compare it to overa range of tau from 0.1 S to 1,000,000 seconds. To get the 1,000,000 second data accurately, you would need a 100,000,000 secondrun. The simple answer there is that nobody has that kind of time or that reliable a setup. Even the three month run to get good100,000 second data is a challenge. None of that relates to three corner hat stuff, it’s just the confidence bars on ADEV. It givesyou another (say) 100:1 wait on top of the three corner stuff. 
Now toss in the basics of GPS. Depending on the day, you will get <10 ns to  >100 ns swing over a  24 hour period. Today may or may not be the same as tomorrow. That’s with a “perfect” L1 setup. The variation comes from the ionosphere and the fact thatthe GPS data does not allow you to fully correct for it.  In addition, you will get some interesting bumps related to constellations and your local antenna setup. Any GPSDO that is quartz based will happily follow the 24 hour swing in the GPS from the ionosphere. At 100,000 seconds, a 100 ns swing is 1x10^-12. That’s a lot of disruption. It most certainly is not the sort of thing that ADEV expects to pop up in the middle of a run. 
The simple answer to all this is “don’t go there”. Three corner hat is fine for short term stuff. It’s a mess for long term runs. Getting datathat is good enough for a long term ADEV run out of a three corner setup is a major struggle. The time for the correlation to knock downthe noise on top of the time to get good ADEV data gets you into impractically long runs. 
Bob

On Apr 16, 2017, at 10:16 PM, Bob Stewart bob@evoria.net wrote:
Hi Bob,
OK, I give up.  Try as I might, I can't fudge things enough to make any sense.  I did run a set of 1 hour tests that seemed to confirm what I can infer from the phase plots in timelab, but anything longer than that and the curves either contradict the phase plots, or there are large gaps in at least one trace, or a trace is even missing entirely.  Oh well.  I seem to remember reading that the 3c-hat was only useful in comparing similar devices.  The KS just isn't close enough to what I'm trying to compare it to, I guess.

If you or anyone else has an ADEV plot of the KS against some local standard (for any length of time, any standard, even just a bare OCXO that is not a Trimble 34310-T) could you please share it with me?  I'm looking for relative peformance, not a definitive test.  Of course if you also have one of a 34310-T against the same standard, that would be great!

Bob 

  From: Bob kb8tq <kb8tq@n1k.org>

To: Bob Stewart bob@evoria.net; Discussion of precise time and frequency measurement time-nuts@febo.com
Cc: John Miles john@miles.io
Sent: Thursday, April 13, 2017 7:19 PM
Subject: Re: [time-nuts] Three-cornered hat on timelab?

Hi

There are a number of papers from the 70’s and 80’s digging into three corner hat data. The net result often
would turn out to be “less than zero” noise on one of the DUT’s. Since that’s physically impossible the technique
got a bit of “attention”.  The Cliff Notes version of the results is that simultaneous measurements were the key
to getting decent results. The closer to “same time” (as in microseconds or nanoseconds) the better. Even with very careful
data collection, odd things can still happen. Phase noise pops up at crazy low levels or ADEV goes to bizarre
numbers. In many ways a TimePod (or other ADC based setup) is ideal for getting the data synchronized. Running
all three devices on one is by far the best way I have seen to make the technique work. It still can have problems,
but less so that other ways of doing it.

Bob

Bob S,

Here's an example of a one week GPSDO run:

http://leapsecond.com/pages/tbolt-8d/

IIRC, this was a default, untuned, self-surveyed TBolt. You can see some level of daily variations -- probably a mix of sky view, survey error, ionosphere, multi-path, sidereal effects [1], temperature (antenna, cable, GPSDO, reference), etc. It takes some time and equipment to sort out which is which, but even a simple test like this can give you an upper bound.

/tvb

[1] Fun GPS orbit stuff here:
http://leapsecond.com/pages/sidereal/index.htm
http://leapsecond.com/pages/sidereal/14years.htm
http://leapsecond.com/pages/sidereal/sv.htm
http://leapsecond.com/pages/gps-orbit/

----- Original Message -----
From: "Bob Stewart" bob@evoria.net
To: "Bob kb8tq" kb8tq@n1k.org; "Discussion of Precise Time and Frequency Measurement" time-nuts@febo.com
Sent: Monday, April 17, 2017 9:33 AM
Subject: Re: [time-nuts] Three-cornered hat on timelab?

Hi Bob,
OK, thanks. I've kicked off a 7 day run of a GFS against the PRS-45A. That should be long enough to separate out the GFS from the PRS' drift direction from the ionosphere.

Bob

  From: Bob kb8tq <kb8tq@n1k.org>

To: Bob Stewart bob@evoria.net
Cc: Discussion of precise time and frequency measurement time-nuts@febo.com
Sent: Monday, April 17, 2017 11:28 AM
Subject: Re: [time-nuts] Three-cornered hat on timelab?

Hi
The ionosphere is the culprit in terms of the daily swing. The swing is a function of the goodness of fit between the GPS broadcast dataand the ionosphere as it impacts the satellites you are using. There is no rime or reason to it beyond that. If you get “lucky” things don’t move much. If you live in exciting times, things move quite a bit. Unless you go to something like an L1/L2 receiver, the GPS module you use has little to do with it (unless it’s broke ….). Yes there are some fiddly little qualifiers relating to being at the north or south pole and GPS coverage (along with space weather impacts). Very few of us do our runs at either location :) Just for reference, the area of concern also hasat least one day each year where the sun sets for < 1 hours.
Bob

On Apr 17, 2017, at 11:33 AM, Bob Stewart bob@evoria.net wrote:
Hi Bob,
Oh, I had completely forgotten about the many runs you gifted us with back then. Fortunately, I kept all of them in my email archive. I can't compare like for like, of course, but I think I can work up something that compares at the larger taus where the 5370 doesn't dominate.
I'm going to run another long term test of my GFS unit against my PRS-45A. The problem, the issue that made me ask for data is that everything from phase plots to ADEV plots of my unit are just so much better than the KS. In addition, I don't see the large ionospheric swings on my GFS unit that you and Bruce and others have spoken about. This bothers me a lot. Could it be my location here in Houston? Could it be the Ublox LEA-6T compared to the much older Motorola in the KS?

Bob

  From: Bob kb8tq <kb8tq@n1k.org>

To: Bob Stewart bob@evoria.net
Cc: Discussion of precise time and frequency measurement time-nuts@febo.com
Sent: Monday, April 17, 2017 7:55 AM
Subject: Re: [time-nuts] Three-cornered hat on timelab?

Hi
The data I have on the KS boxes was posted to the list back when they were.a hot topic. I’m sure it is still in the archives.I’m guessing it’s not quite what you are after.

The closer the devices are to each other the better the technique works. A simple way to look at it is as an attenuation. If it knocks noise down 10:1, the worst unit should be no more than 10X noise than the best unit. How much things are knockeddown is a function of the length of the runs compared to the longest tau. For a 10:1 ratio of tau to run, attenuation of noise by 10:1 is very optimistic. You usually need something beyond 100:1 to get that sort of performance. A lot depends on the noise involved. Some types of behavior simply don’t work well with the technique.
The KS box goes from “better than” to “worse than” and back to “better than” most atomic standards you would compare it to overa range of tau from 0.1 S to 1,000,000 seconds. To get the 1,000,000 second data accurately, you would need a 100,000,000 secondrun. The simple answer there is that nobody has that kind of time or that reliable a setup. Even the three month run to get good100,000 second data is a challenge. None of that relates to three corner hat stuff, it’s just the confidence bars on ADEV. It givesyou another (say) 100:1 wait on top of the three corner stuff.
Now toss in the basics of GPS. Depending on the day, you will get <10 ns to >100 ns swing over a 24 hour period. Today may or may not be the same as tomorrow. That’s with a “perfect” L1 setup. The variation comes from the ionosphere and the fact thatthe GPS data does not allow you to fully correct for it. In addition, you will get some interesting bumps related to constellations and your local antenna setup. Any GPSDO that is quartz based will happily follow the 24 hour swing in the GPS from the ionosphere. At 100,000 seconds, a 100 ns swing is 1x10^-12. That’s a lot of disruption. It most certainly is not the sort of thing that ADEV expects to pop up in the middle of a run.
The simple answer to all this is “don’t go there”. Three corner hat is fine for short term stuff. It’s a mess for long term runs. Getting datathat is good enough for a long term ADEV run out of a three corner setup is a major struggle. The time for the correlation to knock downthe noise on top of the time to get good ADEV data gets you into impractically long runs.
Bob

On Apr 16, 2017, at 10:16 PM, Bob Stewart bob@evoria.net wrote:
Hi Bob,
OK, I give up. Try as I might, I can't fudge things enough to make any sense. I did run a set of 1 hour tests that seemed to confirm what I can infer from the phase plots in timelab, but anything longer than that and the curves either contradict the phase plots, or there are large gaps in at least one trace, or a trace is even missing entirely. Oh well. I seem to remember reading that the 3c-hat was only useful in comparing similar devices. The KS just isn't close enough to what I'm trying to compare it to, I guess.

If you or anyone else has an ADEV plot of the KS against some local standard (for any length of time, any standard, even just a bare OCXO that is not a Trimble 34310-T) could you please share it with me? I'm looking for relative peformance, not a definitive test. Of course if you also have one of a 34310-T against the same standard, that would be great!

Bob

  From: Bob kb8tq <kb8tq@n1k.org>

To: Bob Stewart bob@evoria.net; Discussion of precise time and frequency measurement time-nuts@febo.com
Cc: John Miles john@miles.io
Sent: Thursday, April 13, 2017 7:19 PM
Subject: Re: [time-nuts] Three-cornered hat on timelab?

Hi

There are a number of papers from the 70’s and 80’s digging into three corner hat data. The net result often
would turn out to be “less than zero” noise on one of the DUT’s. Since that’s physically impossible the technique
got a bit of “attention”. The Cliff Notes version of the results is that simultaneous measurements were the key
to getting decent results. The closer to “same time” (as in microseconds or nanoseconds) the better. Even with very careful
data collection, odd things can still happen. Phase noise pops up at crazy low levels or ADEV goes to bizarre
numbers. In many ways a TimePod (or other ADC based setup) is ideal for getting the data synchronized. Running
all three devices on one is by far the best way I have seen to make the technique work. It still can have problems,
but less so that other ways of doing it.

Bob

time-nuts mailing list -- time-nuts@febo.com
To unsubscribe, go to https://www.febo.com/cgi-bin/mailman/listinfo/time-nuts
and follow the instructions there.

Hi Tom,
The reason I express so much confusion over this is because I don't see the wild phase excursions on my GFS units that people insist will happen due to ionospheric effects.  Is this because they are rare events, and I just haven't been saving data during a bad time?  I notice in your example page, you aren't seeing them, either during your 8+ day capture of the Tbolt.

Bob

  From: Tom Van Baak <tvb@LeapSecond.com>

To: Discussion of precise time and frequency measurement time-nuts@febo.com
Sent: Monday, April 17, 2017 3:05 PM
Subject: Re: [time-nuts] Three-cornered hat on timelab?

Bob S,

Here's an example of a one week GPSDO run:

http://leapsecond.com/pages/tbolt-8d/

IIRC, this was a default, untuned, self-surveyed TBolt. You can see some level of daily variations -- probably a mix of sky view, survey error, ionosphere, multi-path, sidereal effects [1], temperature (antenna, cable, GPSDO, reference), etc. It takes some time and equipment to sort out which is which, but even a simple test like this can give you an upper bound.

/tvb

[1] Fun GPS orbit stuff here:
http://leapsecond.com/pages/sidereal/index.htm
http://leapsecond.com/pages/sidereal/14years.htm
http://leapsecond.com/pages/sidereal/sv.htm
http://leapsecond.com/pages/gps-orbit/

----- Original Message -----
From: "Bob Stewart" bob@evoria.net
To: "Bob kb8tq" kb8tq@n1k.org; "Discussion of Precise Time and Frequency Measurement" time-nuts@febo.com
Sent: Monday, April 17, 2017 9:33 AM
Subject: Re: [time-nuts] Three-cornered hat on timelab?

Hi Bob,
OK, thanks. I've kicked off a 7 day run of a GFS against the PRS-45A. That should be long enough to separate out the GFS from the PRS' drift direction from the ionosphere.

Bob

      From: Bob kb8tq kb8tq@n1k.org
To: Bob Stewart bob@evoria.net
Cc: Discussion of precise time and frequency measurement time-nuts@febo.com
Sent: Monday, April 17, 2017 11:28 AM
Subject: Re: [time-nuts] Three-cornered hat on timelab?
 
Hi
The ionosphere is the culprit in terms of the daily swing. The swing is a function of the goodness of fit between the GPS broadcast dataand the ionosphere as it impacts the satellites you are using. There is no rime or reason to it beyond that. If you get “lucky” things don’t move much. If you live in exciting times, things move quite a bit. Unless you go to something like an L1/L2 receiver, the GPS module you use has little to do with it (unless it’s broke ….). Yes there are some fiddly little qualifiers relating to being at the north or south pole and GPS coverage (along with space weather impacts). Very few of us do our runs at either location :) Just for reference, the area of concern also hasat least one day each year where the sun sets for < 1 hours.
Bob

On Apr 17, 2017, at 11:33 AM, Bob Stewart bob@evoria.net wrote:
Hi Bob,
Oh, I had completely forgotten about the many runs you gifted us with back then. Fortunately, I kept all of them in my email archive. I can't compare like for like, of course, but I think I can work up something that compares at the larger taus where the 5370 doesn't dominate.
I'm going to run another long term test of my GFS unit against my PRS-45A. The problem, the issue that made me ask for data is that everything from phase plots to ADEV plots of my unit are just so much better than the KS. In addition, I don't see the large ionospheric swings on my GFS unit that you and Bruce and others have spoken about. This bothers me a lot. Could it be my location here in Houston? Could it be the Ublox LEA-6T compared to the much older Motorola in the KS?

Bob

      From: Bob kb8tq kb8tq@n1k.org
To: Bob Stewart bob@evoria.net
Cc: Discussion of precise time and frequency measurement time-nuts@febo.com
Sent: Monday, April 17, 2017 7:55 AM
Subject: Re: [time-nuts] Three-cornered hat on timelab?
 
Hi
The data I have on the KS boxes was posted to the list back when they were.a hot topic. I’m sure it is still in the archives.I’m guessing it’s not quite what you are after.

The closer the devices are to each other the better the technique works. A simple way to look at it is as an attenuation. If it knocks noise down 10:1, the worst unit should be no more than 10X noise than the best unit. How much things are knockeddown is a function of the length of the runs compared to the longest tau. For a 10:1 ratio of tau to run, attenuation of noise by 10:1 is very optimistic. You usually need something beyond 100:1 to get that sort of performance. A lot depends on the noise involved. Some types of behavior simply don’t work well with the technique.
The KS box goes from “better than” to “worse than” and back to “better than” most atomic standards you would compare it to overa range of tau from 0.1 S to 1,000,000 seconds. To get the 1,000,000 second data accurately, you would need a 100,000,000 secondrun. The simple answer there is that nobody has that kind of time or that reliable a setup. Even the three month run to get good100,000 second data is a challenge. None of that relates to three corner hat stuff, it’s just the confidence bars on ADEV. It givesyou another (say) 100:1 wait on top of the three corner stuff.
Now toss in the basics of GPS. Depending on the day, you will get <10 ns to >100 ns swing over a 24 hour period. Today may or may not be the same as tomorrow. That’s with a “perfect” L1 setup. The variation comes from the ionosphere and the fact thatthe GPS data does not allow you to fully correct for it. In addition, you will get some interesting bumps related to constellations and your local antenna setup. Any GPSDO that is quartz based will happily follow the 24 hour swing in the GPS from the ionosphere. At 100,000 seconds, a 100 ns swing is 1x10^-12. That’s a lot of disruption. It most certainly is not the sort of thing that ADEV expects to pop up in the middle of a run.
The simple answer to all this is “don’t go there”. Three corner hat is fine for short term stuff. It’s a mess for long term runs. Getting datathat is good enough for a long term ADEV run out of a three corner setup is a major struggle. The time for the correlation to knock downthe noise on top of the time to get good ADEV data gets you into impractically long runs.
Bob

On Apr 16, 2017, at 10:16 PM, Bob Stewart bob@evoria.net wrote:
Hi Bob,
OK, I give up. Try as I might, I can't fudge things enough to make any sense. I did run a set of 1 hour tests that seemed to confirm what I can infer from the phase plots in timelab, but anything longer than that and the curves either contradict the phase plots, or there are large gaps in at least one trace, or a trace is even missing entirely. Oh well. I seem to remember reading that the 3c-hat was only useful in comparing similar devices. The KS just isn't close enough to what I'm trying to compare it to, I guess.

If you or anyone else has an ADEV plot of the KS against some local standard (for any length of time, any standard, even just a bare OCXO that is not a Trimble 34310-T) could you please share it with me? I'm looking for relative peformance, not a definitive test. Of course if you also have one of a 34310-T against the same standard, that would be great!

Bob

      From: Bob kb8tq kb8tq@n1k.org
To: Bob Stewart bob@evoria.net; Discussion of precise time and frequency measurement time-nuts@febo.com
Cc: John Miles john@miles.io
Sent: Thursday, April 13, 2017 7:19 PM
Subject: Re: [time-nuts] Three-cornered hat on timelab?
 
Hi

There are a number of papers from the 70’s and 80’s digging into three corner hat data. The net result often
would turn out to be “less than zero” noise on one of the DUT’s. Since that’s physically impossible the technique
got a bit of “attention”. The Cliff Notes version of the results is that simultaneous measurements were the key
to getting decent results. The closer to “same time” (as in microseconds or nanoseconds) the better. Even with very careful
data collection, odd things can still happen. Phase noise pops up at crazy low levels or ADEV goes to bizarre
numbers. In many ways a TimePod (or other ADC based setup) is ideal for getting the data synchronized. Running
all three devices on one is by far the best way I have seen to make the technique work. It still can have problems,
but less so that other ways of doing it.

Bob

time-nuts mailing list -- time-nuts@febo.com
To unsubscribe, go to https://www.febo.com/cgi-bin/mailman/listinfo/time-nuts
and follow the instructions there.

time-nuts mailing list -- time-nuts@febo.com
To unsubscribe, go to https://www.febo.com/cgi-bin/mailman/listinfo/time-nuts
and follow the instructions there.

Hi

If you look at Tom’s data, he very definitely has a peak to peak in the 5 to 7 ns range over a 24 hour period. He also gets the expected auto
correlation spikes.

Bob

Bob,

First of all, there is a first degree of compensation from the GPS
transmitted Klobuchar ionspheric model. There is a limit to how well
those would match the actual values at the time and behavior for your
spot on the globe. The GPS models this to a fair fit for the globe.
Use of WAAS/EGNOS or even DGPS would allow for a better correction.

Second, these changes is slow, so you better measure them compared to a
cesium or maybe rubidium rather than the GPS itself or another GPS. A
GPS tracks in these deviations, so it will only be visible when compared
to an independent source. The frequency error and drift of the reference
clock can be compensated, but the remainder will dominantly be remaining
delay variations.

You can fair better if you have a double-frequency GPS setup, as it can
first-degree measure and compensate the ionospheric shifts, which allows
for a benefit over L1 CA only receiver.

Cheers,
Magnus

On 04/17/2017 10:38 PM, Bob Stewart wrote:

Hi Magnus,
Today I started a long run against my PRS-45A.  Maybe this time I won't have a power outage.  I'll see what it tells me in a few days.
Bob 

  From: Magnus Danielson <magnus@rubidium.dyndns.org>

To: Bob Stewart bob@evoria.net; Discussion of precise time and frequency measurement time-nuts@febo.com
Cc: magnus@rubidium.se
Sent: Monday, April 17, 2017 5:28 PM
Subject: Re: [time-nuts] Three-cornered hat on timelab?

Bob,

First of all, there is a first degree of compensation from the GPS
transmitted Klobuchar ionspheric model. There is a limit to how well
those would match the actual values at the time and behavior for your
spot on the globe. The GPS models this to a fair fit for the globe.
Use of WAAS/EGNOS or even DGPS would allow for a better correction.

Second, these changes is slow, so you better measure them compared to a
cesium or maybe rubidium rather than the GPS itself or another GPS. A
GPS tracks in these deviations, so it will only be visible when compared
to an independent source. The frequency error and drift of the reference
clock can be compensated, but the remainder will dominantly be remaining
delay variations.

You can fair better if you have a double-frequency GPS setup, as it can
first-degree measure and compensate the ionospheric shifts, which allows
for a benefit over L1 CA only receiver.

Cheers,
Magnus

On 04/17/2017 10:38 PM, Bob Stewart wrote:

Hi Bob,

That is a good solution indeed. Good luck with that measurement run!

One of the fun stuff with Timelab is that you can walk by and check the
developments. I've found that very useful for long measurements (as in
hours and days).

I prepared a cesium for one vendor, and initially they did not care so
much, but then they saw more deviations between the receivers, so they
wanted to sort it out, but discovered that they could not cancel out the
common mode of GPS signals (and its shifts), so then firing up that
cesium was the right thing. I remember writing support emails while
waiting for the airplane in Madrid airport, happy that they was doing a
first run for the right measurement reason. :)

Cheers,
Magnus

On 04/18/2017 04:25 AM, Bob Stewart wrote:

Hi Magnus,
Try as I might, the weather and the local power company had other ideas about my long term capture.  I'm running everything but the 5370 from a UPS.  I guess I'm going to have to get batteries for my other UPS and run the 5370 from that.  A one second power loss was all it took to stop the test.

Anyway, I did manage to get 376,238 points of data.  The data is captured on a 5370A.  The external clock input and the STOP channel are fed by the 10MHz from my PRS-45A.  The START channel is fed by the 10MHz from one of my GPSDOs.  The EXT channel is fed by the 1PPS from another of my GPSDO units.  "EXT ARM" is enabled.  So, essentially, at every 1PPS pulse, the phase difference between the two 10MHz feeds is captured.

I've attached a screenshot of the phase plot which can also be found here:http://evoria.net/AE6RV/Timelab/Screenshot.png
I've also made the timelab file (compressed by 7z) available here:
http://evoria.net/AE6RV/Timelab/GFSvsCS.4.22.17.7z

So, back to my question:  Where are the large ionospheric phase moves?  This question has been causing me doubt since I started on this project.  Or don't I still have enough data collected for this to happen?

Bob

AE6RV.com

GFS GPSDO list:
groups.yahoo.com/neo/groups/GFS-GPSDOs/info

  From: Magnus Danielson <magnus@rubidium.dyndns.org>

To: Bob Stewart bob@evoria.net; Discussion of precise time and frequency measurement time-nuts@febo.com
Cc: magnus@rubidium.se
Sent: Tuesday, April 18, 2017 1:09 AM
Subject: Re: [time-nuts] Three-cornered hat on timelab?

Hi Bob,

That is a good solution indeed. Good luck with that measurement run!

One of the fun stuff with Timelab is that you can walk by and check the
developments. I've found that very useful for long measurements (as in
hours and days).

I prepared a cesium for one vendor, and initially they did not care so
much, but then they saw more deviations between the receivers, so they
wanted to sort it out, but discovered that they could not cancel out the
common mode of GPS signals (and its shifts), so then firing up that
cesium was the right thing. I remember writing support emails while
waiting for the airplane in Madrid airport, happy that they was doing a
first run for the right measurement reason. :)

Cheers,
Magnus

On 04/18/2017 04:25 AM, Bob Stewart wrote:

Hi

You have roughly 25 ns p-p in the data you show. There are a number of 10 ns “cycles” in the data.
Any of this may be due to ionosphere. They also could be due to other issues.  With ~4.4 days of noisy
data, it may be tough to spot a trend. Since the ionosphere is a bit random, there is no guarantee that
you will always see a pretty sinusoidal trend line through the data. It’s a good bet that things quiet down
around midnight. There is no guarantee that they always go nuts (or go nuts to the same degree) around noon.

Bob

Hi Bob,

On 04/27/2017 06:48 PM, Bob Stewart wrote:

Annoying, but you got some good values never the less.

OK, this seems like a good setup.

Thank you for providing the data, I downloaded it so I can play around
with it, which I naturally did. :)

Your data seems to be more affected by constellation shifts, as the
period of about 43080 s seems to be a period of the constellation.
You either have averaged out to a somewhat incorrect position of your
antenna or you have sub-optimal position of your antenna.

It gives you a peak-to-peak amplitude of about 10 ns or so.

The ionospheric errors has a period of 86400s, so to get a clear
separation of these would take more data. However, playing around with
the data in TimeLab allowed me to filter out some of the other systematics.

The day-to-day variations is noticeable. I wonder how much of that is
thermal though. The building variations was filtered out in the process.

One has to identify a number of these potential disturbances, estimate
their size in order to more clearly see other things. TimeLab has a
notch filter to notch out a particular frequency. It would be nice if an
alternative approach would be to give the notch a period.

One has to recall that even and odd harmonics to a disturbance frequency
can be there, as it is not always a pure sine disturbance.

Cheers,
Magnus

On Thu, Apr 27, 2017 at 12:48 PM, Bob Stewart bob@evoria.net wrote:

Bob, my test setup is a good deal simpler than yours, but attached is a
plot that I think shows the variations you are looking for quite clearly.
This is data from my homebrew GPSDO, which uses an Adafruit non-timing GPS
module and a run-of-the-mill surplus OCXO. The plot records the phase
comparator output over a period of about 1 week. The time constant of the
PLL is 1024 seconds and it is plotting the 5-minute average TIC values.

The full horizontal scale is 24 hours.

The vertical scale shows the data from several days with the traces for
successive days offset upwards by the equivalent of 40 nsec.

As you can see there is pretty good correlation of the phase error from day
to day and the wiggles migrate to the left a little, corresponding to the
23:56:04 siderial repeat time of the GPS constellation.This is with a
pretty good antenna location, under a shingle roof in the attic. I
calculate the day-to-day correlation at about 0.8.

Making the time constant larger increases the variations somewhat, because
the loop does not adjust as much, and they definitely get worse if I use a
less optimal antenna location.

--

--Jim Harman

Hi Magnus,
OK, a couple of things about my location.  I'm in West Houston, and it's not summer yet, so there's a lot of variation in temperature from day to day.  Some nights it's in the 40sF and some nights it's in the high 70s or low 80sF.  Lots of variation in the days, as well.  My antenna is not optimal, at all.  The best I could do was to remove the dish from an unused DishTV antenna and install my GPS antenna on top of the little mast they use.  It's about the best I can do.  In fact, it's better than I expected.
The receiver is a LEA-6T that was put through a 24 hour survey and the position was saved in flash memory.  However, there have been lots of power cycles since that survey.  Whether or not that affects the result, I don't know.

Still, the point of the test was to understand why I'm not getting these large phase swings.  And I think Bob Camp's explanation was good.  Maybe in another 5 years the sunspots will be back up and I can see the comparison to now. Bob

  From: Magnus Danielson <magnus@rubidium.se>

To: Bob Stewart bob@evoria.net; Discussion of Precise Time and Frequency Measurement time-nuts@febo.com
Cc: magnus@rubidium.se
Sent: Saturday, April 29, 2017 6:45 AM
Subject: Re: [time-nuts] Three-cornered hat on timelab?

Hi Bob,

On 04/27/2017 06:48 PM, Bob Stewart wrote:

Annoying, but you got some good values never the less.

OK, this seems like a good setup.

Thank you for providing the data, I downloaded it so I can play around
with it, which I naturally did. :)

Your data seems to be more affected by constellation shifts, as the
period of about 43080 s seems to be a period of the constellation.
You either have averaged out to a somewhat incorrect position of your
antenna or you have sub-optimal position of your antenna.

It gives you a peak-to-peak amplitude of about 10 ns or so.

The ionospheric errors has a period of 86400s, so to get a clear
separation of these would take more data. However, playing around with
the data in TimeLab allowed me to filter out some of the other systematics.

The day-to-day variations is noticeable. I wonder how much of that is
thermal though. The building variations was filtered out in the process.

One has to identify a number of these potential disturbances, estimate
their size in order to more clearly see other things. TimeLab has a
notch filter to notch out a particular frequency. It would be nice if an
alternative approach would be to give the notch a period.

One has to recall that even and odd harmonics to a disturbance frequency
can be there, as it is not always a pure sine disturbance.

Cheers,
Magnus

Hi Bob,

This is a phase plot of a rubidium to an M12 from a test that I did
back in 2008. The offset and ageing have been removed, but there is
still a bit of wander.
As with your plot, constellation related issues appear the most
obvious. Peak to peak is fairly similar too.

Angus.

On Thu, 27 Apr 2017 16:48:37 +0000 (UTC), you wrote:

Jim,

Errors aligning up due to sidreal time is most likely due to the
multipath errors and repeated upload shifts.

Solar shifts would align up on 24 h basis.

The filtering of the time-constant is expected.

Cheers,
Magnus

On 04/29/2017 04:14 PM, Jim Harman wrote:

What other patterns, if any, are uncovered if one removes a smoothed sidereal variation?

— Eric

Hi,

Upload rate could be one. There is some systematics due to jumps from
the old to new estimation. I think I recall a 2 h upload rate, but no
guarantee for that number.

Cheers,
Magnus

On 04/30/2017 06:32 PM, Eric Scace wrote:

Hi

A lot of the upload and estimate processes have changed over the years. What was true
at "maybe every 2 hours" four years ago might be wrong today. Oddly enough it could be
wrong in either direction.  What they can’t change is the coarse resolution of the estimate.
That’s embedded in the rather limited bit stream they have coming down from the sat’s.

Bob

Hi,

Upload rates have been important for GALILEO to distinguish itself from GPS.

As GPS has been looking into the future, they have asked about the
bitstream rates and asked increased speed vs. improved redundancy. I
think I recall going towards the later.

Cheers,
Magnus

On 04/30/2017 10:23 PM, Bob kb8tq wrote: