[time-nuts] Testing frequency using NTP Bruce GPS ps
Mike Monett
XDE-L2G3 at myamail.com
Thu Oct 9 08:09:08 UTC 2008
"Tom Van Baak" <tvb at LeapSecond.com> wrote:
> Note another equivalent technique is to use two serial ports; one
> for the receiver (record sawtooth corrections) and one for a
> 53132A-style counter (record TI measurements) and then do the
> calculations in software. TAC32 does this. I think TBoltMon also
> allows it. Roughly, it's a trade-off in equipment and complexity.
> They give essentially the same performance result.
I have to learn more about how you do your measurements. A 53132A is
way out of my price range at the moment. But I do have a 53310A
which should give comparable results.
> There are still many other sources of noise, both short- and
> long-term in an OEM GPS receiver/antenna system. That's why even
> if there were zero quantization error, you would still see a
> couple of ns rms error in the 1PPS output.
I am interested in learning more about these errors. Is there a list
somewhere that describes the sources of error, and what can be done
to reduce it? I keep coming across hints of doing this in software,
but it would seem that has to be done in the signal processing part
of the receiver. But we don't have access to that. Is there anything
else we can do to help remove some of these errors?
>> Thanks. I had not found that page yet. Just judging by eyeball,
>> the pink and yellow traces don't seem to track very well. Any
>> reason?
> On that plot, the various runs weren't concurrent so you can't
> look for common-mode effects. Separate runs, showing typical
> levels of jitter, all plotted on the same x-y scale. You've got a
> good eye.
>> And what's the blue trace for? I can't seem to find the parent
>> page, so I don't know if you have already explained it.
> This will help a bit more:
>http://www.febo.com/pipermail/time-nuts/2005-April/018114.html
Thanks - I'll have to study that a bit more. Right now I can't make
heads or tails of it - maybe I'm too tired and need to get some
sleep. Sure - I just looked at the clock. It's 4am. That explains
it.
>> Overall, I was a little disappointed to find the sawtooth
>> correction only gives about a factor of 3 or so improvement.
> That's partly because the sawtooth of the M12+ is small to begin
> with, at least compared to the earlier VP. So there isn't much
> room for sawtooth correction to have a massive gain. Also the
> granularity of the quantization message from the receiver is 1 ns.
That is very interesting. What do you mean by granularity? I have a
file somewhere that talks about the algorithm to select the next
1PPS pulse. It added 500ps to the rounding, so the result would be
on 1 ns boundaries. Is that what you meant?
> And there's an internal delay of one or two seconds which has a
> slight effect on the quality of the correction.
That is getting very subtle. How much effect does it have, and where
on earth do you learn all this stuff?
> Another way to look at it is this. If a perfect M12+ had, say, 2.5
> ns of jitter, and you saw 10 ns rms without sawtooth correction
> and 3 ns rms with sawtooth correction you could say you achieved a
> 15x improvement! (10-2.5)/(3-2.5). Or if you didn't say it, the
> marketing department certainly would.
Don't you have to use the difference of squares in working with rms
values?
> The closer you get to a few nanoseconds of jitter anyway, the more
> all the other errors in a typical cheap OEM GPS receiver and
> antenna system come into play. So once you average beyond a couple
> of minutes the wander you see in the 1PPS has less and less to do
> with sawtooth and more to do with the sum of all the other subtle
> errors.
That's what I need to learn more about.
> Expensive geodetic or timing receivers use an assortment of
> techniques to reduce the effect of these remaining error sources,
> to the point where you'll hear of millimeters and picoseconds (and
> priced accordingly).
Where can I learn more about these techniques?
>> obtained with my method have given over two orders of magnitude
>> improvement in noise reduction. That's with no optimization.
> Averaging over multiple samples, of course, gives a reduction in
> noise. You can see this in the ADEV plots at the end of:
>http://www.leapsecond.com/pages/58503-cns2/
That is really outstanding work. You have a magic skill to take
complex data and present it in such a way to make it clear and
understandable.
But to be honest, all the noisy traces hurt my eyes. I am now used
to seeing signals with all the noise removed, so you can look at
more subtle things in the waveforms.
> On the other hand, if you found a way to get two orders of
> magnitude better performance out of an M12+ without the use of
> averaging then I'm all ears.
This is a technique I discovered last century. It is really very
simple, but some people may have a difficult time believing it can
be done. So rather than waste everyone's time trying to explain it,
a demonstration with a credible and unbiased person such as yourself
would probably work a lot better.
I have some stuff on order, and plan to make two simple systems
using an Oncore UT and a VCXO. It will take about two or three
months to get everything through Canadian customs and build the
prototype. It will be hand-wired, but I have developed a new method
of making very rugged low-noise prototypes suited for frequencies
well into the microwave region.
When it is working, I'll make a copy and send it to you. Then if you
would be kind enough to make your magical measurements, I would be
very pleased to see the results.
If it is what I think it will be, probably others would like to see
the results also.
If you are interested, I'll let you know my progress so we can
schedule a time that is convenient for you.
>/tvb
Best Regards,
Mike Monett
More information about the time-nuts
mailing list