[time-nuts] Some results of PRS10 and Trimble Resolution
Tom Van Baak
tvb at leapsecond.com
Wed Jun 28 12:00:34 EDT 2006
> Hello Tom,
> We approached the sawtooth correction on the software
> side: we sample the 1PPS with 6.66ns resolution in our
> new Fury GPSDO and apply software adjusted sawtooth
> correction (post-capture), and this yields an easily visible,
> very significant reduction in the 1PPS capture noise with
> the Motorola M12+ receivers. You can really see the
> difference second-to-second when the correction is turned
> on/off. This yields an average sampling quantization noise
> of +-3.33ns.
Sounds good. Given that you still have many ns of
jitter in the 1PPS output you might want to consider
dividing your OCXO down to 1PPS and using that for
your master 1PPS user output; you can reduce the
output jitter 1000x that way. This is what HP did.
> The physical sawtooth on the 1PPS signal is actually
> helpfull in this setup as you mention since it dithers
> the LSB quantization noise, and thus actually improves
> the quantization resolution over time (with simple
> averaging low pass filtering of the captured data). This
> trick is called dithering in audio etc.
Loosely related to that is DAC dithering which
was mentioned here a while ago.
The one difference, though, is that the dithering
you get with a 1PPS output is not always random
enough to guarantee that you get a clean mean
after N seconds. Sometimes it works well, but
you can also get very unlucky (e.g., the hanging
bridge effect in that m12 web page). The longer
you average the less this is a problem; 10 or 30
minutes makes it a non-issue. But with much
shorter averaging times you can see how it might
> Using delay lines may be tricky and expensive, they
> usually are temperature sensitive, and only yield good
> results if the capture of the 1PPS is done in fast enough
> (<10ns capture resolution). But if your capture is fast
> enough anyways, there is no need to use a delay line
> since the correction can be done in software.
The 8-bit 1 ns resolution part from Dallas/Maxim looked
pretty cheap to me. $5, I think.
Yeah, you'd have to check the tempco but also weigh
it against the tempco of many other key components
of a GPSDO.
> Delay lines also have another disadvantage: the 1PPS
> correction from the GPS can be positive or negative in
> time, so in theory you would need a negative time delay
I wouldn't call it a disadvantage; you just advance
the GPS tick by the right amount to compensate.
The same way you compensate for antenna delay,
antenna cable delay, GPS engine delay, 1 PPS
output cable and 74AC buffer delays, etc.
> (Einstein would be happy :). So for the delay line to work
> correctly, you have to set the 0ns delay tap equivalent
> to the most negative pulse, all other pulses will incur more
> than 0ns delay. You are now effectively delaying the average
> 1PPS by 1/2 the spread of the pulses. Thus the output of
> the delay line is always late (by about 30ns for commercial
> receivers) on average for standard GPS receivers. This error
Right, actually any delay greater than half the sawtooth
jitter would work. Using half the range of an 8-bit delay
generator might be nice too; that allows you to give the
user a fine adjustment feature in the range +/- 128 ns
with 1 ns resolution.
> will have to be subtracted later. One trick when using Motorola
> GPS's is to set the cable delay to an additional 30ns to make
> the GPS receiver itself compensate for this delay by issuing
> the 1PPS output 30ns "too fast".
Yes, the antenna delay parameter is the way to do this.
> An interesting effect of sampling at +-3.33ns is that the
> GPS errors such as multipath, atmospheric, and GPS
> crystal temperature related issues become clearly visible
> with this kind of resolution...
Note the 58503 and Z3801A receivers (now ten years
old) have 100 ps resolution. But I'm guessing your
3.33 ns detector is cheaper than HP's!
More information about the time-nuts