Delay through GPS antenna splitter/amplifier -- an answer, and a question
I had a chance recently to look at the performance of the two-port and
eight-port HP GPS antenna splitters on a super-duper network analyzer.
Screenshots of the results are at
http://www.febo.com/time-freq/pages/gps-splitter.
In short, the minimum delay (at the center of the passband) from antenna
port to output port is around 15 nanoseconds for the eight way unit, and
about 22 nanoseconds for the two way one. The delay seems consistent on
all the ports, with less than 1 nanosecond variation.
However, there is also a hump in the delay near the edges of the
passband, about 12 MHz above and below the center. The delay at the
edges increases by perhaps 5 nanoseconds, though depending on the port,
it's not always symmetrical.
So, an interesting question for any of you real GPS experts is what
effect a variation in group delay of the RF input has on the timing
solution? Is the true "length" of the amp/splitter some average of the
delay across the passband, or, given the spread spectrum nature of the
signal, does it not really matter? In fact, is the "length" of the
splitter even related to the measured group delay?
This also raises the issue that any GPS antenna that has RF filtering is
likely to have similar delays; I've never seen that sort of data published.
John
From my experience, your position and hence derived time is based on the
antenna centre. Cable, splitter, connector, and antenna filter delays all
need to be taken into account when looking at very accurate "nanosecond"
timing applications.
For most applications in the microsecond or tens of microsecond region it
isn't worth worrying about.
Rob K
----- Original Message -----
From: "John Ackermann N8UR" jra@febo.com
To: "Discussion of precise time and frequency measurement"
time-nuts@febo.com
Sent: Monday, March 12, 2007 3:03 PM
Subject: [time-nuts] Delay through GPS antenna splitter/amplifier -- an
answer, and a question
I had a chance recently to look at the performance of the two-port and
eight-port HP GPS antenna splitters on a super-duper network analyzer.
Screenshots of the results are at
http://www.febo.com/time-freq/pages/gps-splitter.
In short, the minimum delay (at the center of the passband) from antenna
port to output port is around 15 nanoseconds for the eight way unit, and
about 22 nanoseconds for the two way one. The delay seems consistent on
all the ports, with less than 1 nanosecond variation.
However, there is also a hump in the delay near the edges of the
passband, about 12 MHz above and below the center. The delay at the
edges increases by perhaps 5 nanoseconds, though depending on the port,
it's not always symmetrical.
So, an interesting question for any of you real GPS experts is what
effect a variation in group delay of the RF input has on the timing
solution? Is the true "length" of the amp/splitter some average of the
delay across the passband, or, given the spread spectrum nature of the
signal, does it not really matter? In fact, is the "length" of the
splitter even related to the measured group delay?
This also raises the issue that any GPS antenna that has RF filtering is
likely to have similar delays; I've never seen that sort of data
published.
John
time-nuts mailing list
time-nuts@febo.com
https://www.febo.com/cgi-bin/mailman/listinfo/time-nuts
In message: 001501c764bd$eb38e740$0202a8c0@quaddra
"Rob Kimberley" rk@timing-consultants.com writes:
: >From my experience, your position and hence derived time is based on the
: antenna centre. Cable, splitter, connector, and antenna filter delays all
: need to be taken into account when looking at very accurate "nanosecond"
: timing applications.
Getting cable delay wrong isn't the end of the world. However, rather
than seeing a few ns of variance in pps data, you'll see something
more in the tens of ns variance. The absolute offset isn't always a
good indication if you have the cable delay right. The variance in
your data tends to be a better indication, at least for the systems
that I've worked on.
Of course, if you have an "on-time" pps to test against, things are a
lot easier than if you just have a stable PPS to test against.
: For most applications in the microsecond or tens of microsecond region it
: isn't worth worrying about.
60m of cables + splitters, etc is only going to add 180-200ns to the
propigation time. That's well below 1us or 10us :-).
Warner
But we're time-nuts... we DO worry about those things. :-)
While we were at it with the network analyzer, we did FDR (frequency
domain reflectometry) to measure the cable delay to the antenna, and I
spent yesterday making up six matched cables to go from the splitter to
the receivers -- they all test within about 1 nanosecond of each other.
Obsessive compulsive? Me?
John
Rob Kimberley wrote:
From my experience, your position and hence derived time is based on the
antenna centre. Cable, splitter, connector, and antenna filter delays all
need to be taken into account when looking at very accurate "nanosecond"
timing applications.
For most applications in the microsecond or tens of microsecond region it
isn't worth worrying about.
Rob K
----- Original Message -----
From: "John Ackermann N8UR" jra@febo.com
To: "Discussion of precise time and frequency measurement"
time-nuts@febo.com
Sent: Monday, March 12, 2007 3:03 PM
Subject: [time-nuts] Delay through GPS antenna splitter/amplifier -- an
answer, and a question
I had a chance recently to look at the performance of the two-port and
eight-port HP GPS antenna splitters on a super-duper network analyzer.
Screenshots of the results are at
http://www.febo.com/time-freq/pages/gps-splitter.
In short, the minimum delay (at the center of the passband) from antenna
port to output port is around 15 nanoseconds for the eight way unit, and
about 22 nanoseconds for the two way one. The delay seems consistent on
all the ports, with less than 1 nanosecond variation.
However, there is also a hump in the delay near the edges of the
passband, about 12 MHz above and below the center. The delay at the
edges increases by perhaps 5 nanoseconds, though depending on the port,
it's not always symmetrical.
So, an interesting question for any of you real GPS experts is what
effect a variation in group delay of the RF input has on the timing
solution? Is the true "length" of the amp/splitter some average of the
delay across the passband, or, given the spread spectrum nature of the
signal, does it not really matter? In fact, is the "length" of the
splitter even related to the measured group delay?
This also raises the issue that any GPS antenna that has RF filtering is
likely to have similar delays; I've never seen that sort of data
published.
John
time-nuts mailing list
time-nuts@febo.com
https://www.febo.com/cgi-bin/mailman/listinfo/time-nuts
time-nuts mailing list
time-nuts@febo.com
https://www.febo.com/cgi-bin/mailman/listinfo/time-nuts
Since we are on the subject... I came across this page the other day and am
curious about what other's thoughts are about it.
http://tf.nist.gov/service/gpstrace.htm
If I understand correctly looking at the graphs, they are saying the time
from GPS is currently off about 24-30ns from the NIST frequency standard. So
does that mean that a person wanting every last ns should adjust their delay
according to the data on that page (and of course things like cable delay
and other circuitry)?
Jason
But we're time-nuts... we DO worry about those things. :-)
While we were at it with the network analyzer, we did FDR (frequency
domain reflectometry) to measure the cable delay to the antenna, and I
spent yesterday making up six matched cables to go from the splitter to
the receivers -- they all test within about 1 nanosecond of each other.
Obsessive compulsive? Me?
John
In short, yes. If you want true traceability to NIST, you need to take
into account UTC(GPS) versus UTC(NIST).
I don't think you would do this by adjusting the cable delay, because
that offset will change over time. Instead, you post-process by getting
the offset for the time during which you made the measurement, and
adding it to your solution.
John
Jason Rabel wrote:
Since we are on the subject... I came across this page the other day and am
curious about what other's thoughts are about it.
http://tf.nist.gov/service/gpstrace.htm
If I understand correctly looking at the graphs, they are saying the time
from GPS is currently off about 24-30ns from the NIST frequency standard. So
does that mean that a person wanting every last ns should adjust their delay
according to the data on that page (and of course things like cable delay
and other circuitry)?
Jason
But we're time-nuts... we DO worry about those things. :-)
While we were at it with the network analyzer, we did FDR (frequency
domain reflectometry) to measure the cable delay to the antenna, and I
spent yesterday making up six matched cables to go from the splitter to
the receivers -- they all test within about 1 nanosecond of each other.
Obsessive compulsive? Me?
John
time-nuts mailing list
time-nuts@febo.com
https://www.febo.com/cgi-bin/mailman/listinfo/time-nuts
If I understand correctly looking at the graphs, they are saying the time
from GPS is currently off about 24-30ns from the NIST frequency standard. So
does that mean that a person wanting every last ns should adjust their delay
according to the data on that page (and of course things like cable delay
and other circuitry)?
Jason
It means a person wanting every last ns should adjust their
expectations. There's no point in measuring your cable
delays to the foot when there may be tens of ns of delay in
the antenna, the survey, the cable, the splitter, the receiver,
etc. Not to mention multipath, antenna/cable tempco...
When you get to the sub-100 ns level UTC is not GPS. And
UTC(NIST) isn't UTC(USNO). Even the national labs don't
agree down to the last ns.
/tvb
From: John Ackermann N8UR jra@febo.com
Subject: [time-nuts] Delay through GPS antenna splitter/amplifier -- an answer, and a question
Date: Mon, 12 Mar 2007 11:03:12 -0400
Message-ID: 45F56BB0.90009@febo.com
John,
I had a chance recently to look at the performance of the two-port and
eight-port HP GPS antenna splitters on a super-duper network analyzer.
Screenshots of the results are at
http://www.febo.com/time-freq/pages/gps-splitter.
Hmm... I would have zoomed in closer to only have a +/- 20 MHz span or so.
I have an E5071A at work.
In short, the minimum delay (at the center of the passband) from antenna
port to output port is around 15 nanoseconds for the eight way unit, and
about 22 nanoseconds for the two way one. The delay seems consistent on
all the ports, with less than 1 nanosecond variation.
However, there is also a hump in the delay near the edges of the
passband, about 12 MHz above and below the center. The delay at the
edges increases by perhaps 5 nanoseconds, though depending on the port,
it's not always symmetrical.
As expected.
So, an interesting question for any of you real GPS experts is what
effect a variation in group delay of the RF input has on the timing
solution? Is the true "length" of the amp/splitter some average of the
delay across the passband, or, given the spread spectrum nature of the
signal, does it not really matter? In fact, is the "length" of the
splitter even related to the measured group delay?
The electrical length of the splitter needs to go into the cable delay
compensation.
The C/A signal has it's main peak at 1,57542 GHz (with doppler +/- 6 kHz) and
the first nulls is at +/- 1,023 MHz of that peak. The secondary peaks is at
+/- 2.046 MHz but much reduced in amplitude and additional peaks roll of fairly
quickly. There are some fancy plots showing this in the literature, I don't
have the numbers from the top of my head.
Most commercial receivers have a +/- 1,023 MHz window so the delay differances
outside of that is not really relevant. The flatness of their groupdelay within
that window is however an issue.
I'd expect the effect to be somewhat similar to multipath-effects.
This also raises the issue that any GPS antenna that has RF filtering is
likely to have similar delays; I've never seen that sort of data published.
Indeed. You should also ask yourself this question about the input filter and
MF1, MF2 and maybe even MF3 filters of the GPS receiver. What is the net
effect of these skews?
Cheers,
Magnus
I had a chance recently to look at the performance of the two-port and
eight-port HP GPS antenna splitters on a super-duper network analyzer.
Screenshots of the results are at
http://www.febo.com/time-freq/pages/gps-splitter.
Super nice plots, John.
Tom Clark, do you have comments?
John, next time you can borrow that instrument blow
some hot/cool air on the antenna and see what changes
and by how much.
I've heard that the older GPS antennas, the ones with zero
or less RF filtering, were much better for timing applications
but have never seen data to prove it.
For example, do you have one of the old Motorola hockey
puck ones you could test? If not, I can send you a bunch
of different GPS antenna to try.
/tvb
From: John Ackermann N8UR jra@febo.com
Subject: Re: [time-nuts] Delay through GPS antenna splitter/amplifier -- an answer, and a question
Date: Mon, 12 Mar 2007 12:04:06 -0400
Message-ID: 45F579F6.2070501@febo.com
But we're time-nuts... we DO worry about those things. :-)
Indeed!
Oh my goodness, the PPS is 10 ps late due to unmatched cable-lag! What should I
dooooo? (For the egg-clock in the kitchen)
While we were at it with the network analyzer, we did FDR (frequency
domain reflectometry) to measure the cable delay to the antenna, and I
spent yesterday making up six matched cables to go from the splitter to
the receivers -- they all test within about 1 nanosecond of each other.
I would use TDR/TDT for that. :)
Obsessive compulsive? Me?
Naaa....
Cheers,
Magnus
Tom Van Baak wrote:
John, next time you can borrow that instrument blow
some hot/cool air on the antenna and see what changes
and by how much.
I've heard that the older GPS antennas, the ones with zero
or less RF filtering, were much better for timing applications
but have never seen data to prove it.
That would make sense. The HP splitter has a single filter of some sort
(probably SAW) on each output port. It is probably very similar to the
filters used in current antennas.
Just a swag, it's probably reasonable to assume that each stage of
filtering adds about 10ns of delay.
The tempco of the antenna amplifier/filter is certainly another factor
to consider. If I can get hold of the analyzer again, we can do a hair
dryer test on the splitter to see how it's affected. My mantra has
become "every resonator is a thermometer".
For example, do you have one of the old Motorola hockey
puck ones you could test? If not, I can send you a bunch
of different GPS antenna to try.
My hockey puck died and has been replaced by a Timing 2000 antenna. I
do have a couple of others sitting on the shelf that I could test.
But, I'm not sure how to non-destructively test an antenna, as the
measurements we did require two-port analysis. I'm not sure if we could
radiate enough signal into the antenna to get a usable output.
By the way -- for anyone who's considering similar tests of amplified
GPS antennas or splitters, note that common network analyzer output
levels will drive the amplifier into compression and cause messed up
readings. You need to shove as small a signal as possible into the
device; 0dBm will definitely hose the readings (though it probably won't
damage anything). I think we found that -30dBm was about the maximum we
could use without seeing gain compression.
John
Magnus Danielson wrote:
I would use TDR/TDT for that. :)
My buddy the expert tells me that using frequency sweep rather than
pulse techniques is preferred these days. With multi-GHz sweeps, you
can get better resolution, and you also minimize the problems that
strong external signals can cause when you're measuring a feedline with
an antenna at the end.
But that's just what I'm told...
John
From: John Ackermann N8UR jra@febo.com
Subject: Re: [time-nuts] Delay through GPS antenna splitter/amplifier -- an answer, and a question
Date: Mon, 12 Mar 2007 13:17:50 -0400
Message-ID: 45F58B3E.5000103@febo.com
Magnus Danielson wrote:
I would use TDR/TDT for that. :)
My buddy the expert tells me that using frequency sweep rather than
pulse techniques is preferred these days. With multi-GHz sweeps, you
can get better resolution, and you also minimize the problems that
strong external signals can cause when you're measuring a feedline with
an antenna at the end.
The TDR I have access to is far supperiour to the FDR in the E5071A. The main
reason is that the bandwidth of the E5071A is limited (8.5 GHz).
Network analysers have a far supperiour dynamics, but timing precission miss
out a little. Measuring the group delay would however be a much better solution
if possible (i.e. access to both sides and with suitable connectors).
Cheers,
Magnus
I used to be bothered when I was working on this stuff full time, especially
when we were testing equipment in labs for customers on a "mission". Now, as
long as all my clocks are on the same second.....
:-)
Funny story however.. Was at NPL on Thursday for our bi-annual Time &
Frequency Club meeting. There was a small exhibition running and one of the
companies - Time & Frequency Solutions had a natty new T&F analyser - nice
box - touch screen etc. The sales guy wanted to demo it and show me how it
checked the delay on a sample piece of cable (with BNCs and a BNC T on the
end). He asked me to guess the delay. I looked at it, did a mental
calculation based on the assumed length and the connectors and told him 6nS.
He put it on the machine, and guess what....6nS. He said I'd been watching
and knew the result! Had to explain that I'd been around the T&F "block" a
few times and that sometimes experience (and an inspired guess!) count more
than hardware!!
Cheers
Rob
-----Original Message-----
From: time-nuts-bounces@febo.com [mailto:time-nuts-bounces@febo.com] On
Behalf Of John Ackermann N8UR
Sent: 12 March 2007 16:04
To: Discussion of precise time and frequency measurement
Subject: Re: [time-nuts] Delay through GPS antenna splitter/amplifier -- an
answer, and a question
But we're time-nuts... we DO worry about those things. :-)
While we were at it with the network analyzer, we did FDR (frequency domain
reflectometry) to measure the cable delay to the antenna, and I spent
yesterday making up six matched cables to go from the splitter to the
receivers -- they all test within about 1 nanosecond of each other.
Obsessive compulsive? Me?
John
Rob Kimberley wrote:
From my experience, your position and hence derived time is based on
the antenna centre. Cable, splitter, connector, and antenna filter
delays all need to be taken into account when looking at very accurate
"nanosecond"
timing applications.
For most applications in the microsecond or tens of microsecond region
it isn't worth worrying about.
Rob K
----- Original Message -----
From: "John Ackermann N8UR" jra@febo.com
To: "Discussion of precise time and frequency measurement"
time-nuts@febo.com
Sent: Monday, March 12, 2007 3:03 PM
Subject: [time-nuts] Delay through GPS antenna splitter/amplifier --
an answer, and a question
I had a chance recently to look at the performance of the two-port
and eight-port HP GPS antenna splitters on a super-duper network
analyzer.
Screenshots of the results are at
http://www.febo.com/time-freq/pages/gps-splitter.
In short, the minimum delay (at the center of the passband) from
antenna port to output port is around 15 nanoseconds for the eight
way unit, and about 22 nanoseconds for the two way one. The delay
seems consistent on all the ports, with less than 1 nanosecond variation.
However, there is also a hump in the delay near the edges of the
passband, about 12 MHz above and below the center. The delay at the
edges increases by perhaps 5 nanoseconds, though depending on the
port, it's not always symmetrical.
So, an interesting question for any of you real GPS experts is what
effect a variation in group delay of the RF input has on the timing
solution? Is the true "length" of the amp/splitter some average of
the delay across the passband, or, given the spread spectrum nature
of the signal, does it not really matter? In fact, is the "length"
of the splitter even related to the measured group delay?
This also raises the issue that any GPS antenna that has RF filtering
is likely to have similar delays; I've never seen that sort of data
published.
John
time-nuts mailing list
time-nuts@febo.com
https://www.febo.com/cgi-bin/mailman/listinfo/time-nuts
time-nuts mailing list
time-nuts@febo.com
https://www.febo.com/cgi-bin/mailman/listinfo/time-nuts
time-nuts mailing list
time-nuts@febo.com
https://www.febo.com/cgi-bin/mailman/listinfo/time-nuts
In short, yes. If you want true traceability to NIST, you need to take
into account UTC(GPS) versus UTC(NIST).
And it gets uglier yet. If you want UTC you have to take
into account the UTC - UTC(NIST) delta, which was about
16 ns in January. See the full 2006 record:
http://tf.nist.gov/timefreq/pubs/bulletin/nistutc2006.htm
The point is GPS, NIST, USNO often vary up to tens of
ns over the span of days and months. It highlights the
problem of knowing what the "true time" is.
Also note that, strictly speaking, there is no such thing
as "UTC(GPS)" and BIPM frowns on the phrase UTC(k)
where k isn't one of the national time labs (e.g., NIST,
USNO, NPL, PTB).
What we call "GPS time" is in fact a good approximation
to UTC(USNO) but since the official UTC(USNO) clock
goes through so many levels of distribution, Kalman filters,
uplinks, downlinks, 24+ individual SV clocks, and PLL'd
disciplined oscillators before it reaches your 1 PPS BNC
output, it is no longer "UTC(USNO)".
For ns-level work the key is post-processing, where many
of the systematic and variable offsets can be estimated
days or months later and mathematically backed out from
your archived physical time interval measurements.
/tvb
UTC-GPS is based on UTC-USNO, and therefore there will be a small offset
between other establishments like NIST, NPL who contribute to the UTC.
As you say, the offset will vary with time, so post-processing is the answer
in critical applications (which is the way UTC is determined anyway).
Rob K
-----Original Message-----
From: time-nuts-bounces@febo.com [mailto:time-nuts-bounces@febo.com] On
Behalf Of John Ackermann N8UR
Sent: 12 March 2007 16:51
To: Discussion of precise time and frequency measurement
Subject: Re: [time-nuts] Delay through GPS antenna splitter/amplifier -- an
answer, and a question
In short, yes. If you want true traceability to NIST, you need to take into
account UTC(GPS) versus UTC(NIST).
I don't think you would do this by adjusting the cable delay, because that
offset will change over time. Instead, you post-process by getting the
offset for the time during which you made the measurement, and adding it to
your solution.
John
Jason Rabel wrote:
Since we are on the subject... I came across this page the other day
and am curious about what other's thoughts are about it.
http://tf.nist.gov/service/gpstrace.htm
If I understand correctly looking at the graphs, they are saying the
time from GPS is currently off about 24-30ns from the NIST frequency
standard. So does that mean that a person wanting every last ns should
adjust their delay according to the data on that page (and of course
things like cable delay and other circuitry)?
Jason
But we're time-nuts... we DO worry about those things. :-)
While we were at it with the network analyzer, we did FDR (frequency
domain reflectometry) to measure the cable delay to the antenna, and
I spent yesterday making up six matched cables to go from the
splitter to the receivers -- they all test within about 1 nanosecond of
each other.
Obsessive compulsive? Me?
John
time-nuts mailing list
time-nuts@febo.com
https://www.febo.com/cgi-bin/mailman/listinfo/time-nuts
time-nuts mailing list
time-nuts@febo.com
https://www.febo.com/cgi-bin/mailman/listinfo/time-nuts
Funny story however.. Was at NPL on Thursday for our bi-annual Time &
Frequency Club meeting. There was a small exhibition running and one of the
...
Rob,
That's a great story. And then this is how I was humbled...
I've had a few people over to my time lab and most are really
impressed. But I was showing an expert mechanical engineer
friend that one could actually calculate the length of a cable
using nano- and picosecond time interval measurements
(drum roll...). So it came to 10 feet and he simply said, holding
the cable up in the air, "well that seems about right", but was
not the least bit impressed that I took two minutes and used
thousands of dollars of weird blinking timing gear to measure
something his experienced eye saw immediately or that a $1
measuring tape could have confirmed...
From: "Tom Van Baak" tvb@LeapSecond.com
Subject: Re: [time-nuts] Delay through GPS antenna splitter/amplifier -- ananswer, and a question
Date: Mon, 12 Mar 2007 10:46:40 -0700
Message-ID: 002b01c764ce$64606cf0$0300a8c0@pc52
Tom and Rob,
I often estimate time/distance relationships using simple rule-of-thumb
calculations, and I often find that it is more than sufficient from getting the
basic hang of things.
But both your stories indicate that you should never select a problem to demo
which has an obvious, simple and low-tech solution to it when you want to demo
your high-tech toys^H^Hols. It needs to be utterly important, hard to measure
except with your toys^H^Hols. :)
Cheers,
Magnus
In message: 20070312.180426.-1621061084.cfmd@bredband.net
Magnus Danielson cfmd@bredband.net writes:
: Oh my goodness, the PPS is 10 ps late due to unmatched cable-lag! What should I
: dooooo? (For the egg-clock in the kitchen)
Put a tiny kink in the cable. That should be good for an offset of
that magnitude.
:-)
Warner
Magnus Danielson wrote:
From: "Tom Van Baak" tvb@LeapSecond.com
Subject: Re: [time-nuts] Delay through GPS antenna splitter/amplifier -- ananswer, and a question
Date: Mon, 12 Mar 2007 10:46:40 -0700
Message-ID: 002b01c764ce$64606cf0$0300a8c0@pc52
Tom and Rob,
I often estimate time/distance relationships using simple rule-of-thumb
calculations, and I often find that it is more than sufficient from getting the
basic hang of things.
But both your stories indicate that you should never select a problem to demo
which has an obvious, simple and low-tech solution to it when you want to demo
your high-tech toys^H^Hols. It needs to be utterly important, hard to measure
except with your toys^H^Hols. :)
This is the basic problem with our hobby... I proudly tell people that I
can measure time to trillionths of a second, but am hard pressed for a
good answer when they ask "why do you need to?"