[hpsdr] Some GPS Timing Receiver comments

[Graham / KE9H]

Repost of Tom Clark's GPS Timing Receiver email, as pure text.
There is so  much material, it deserves to be reposted.

Thanks, Tom...

--- Graham

============================================

 On 4/6/2009 5:27 PM, Graham / KE9H, in the thread entitled
[hpsdr] GPS 1PPS board, wrote:
 > The Motorola UT+ (8 satellite tracking) was replaced by the
 > M12+ (12 satellite plus improved jitter) and is now replaced
 > by the iLotus M12M.(3 volt power, and more improved jitter.)
 > The Motorola design was sold to iLotus in Singapore.  So new
 > 12 channel GPS timing receivers with high accuracy timing
 > outputs are available, quoting 10 ns (1 sigma) jitter without
 > granularity (sawtooth) correction and 2 ns. jitter with the
correction.
 >
 > http://www.ilotus.com.sg
 >
 > sold in US by
 >
 > http://www.synergy-gps.com/
 >
 > Trimble also makes a line of timing receivers called Trimble-
Resolution
 > that have been out there for 5 years or so.
 >
 > http://www.leapsecond.com/pages/res-t/
 >
 > http://www.trimble.com/timing/resolution-t.aspx?dtID=overview
 >
 > Development kits for the receivers in current production are
available.


Just to add some factoids to this discussion, I'd suggest that
you look at all the material that Rick Hambly (W2GPS) and I
have posted on our http://gpstime.com website. In particular I
draw your attention to

    * My "Low-cost, High Accuracy GPS Timing" talk from the ION
2000 conference. This pretty well documents the pre- and post-
S/A performance of the 6 & 8 channel Motorola Oncore receivers.
    * Rick's "Critical Evaluation of the Motorola M12+ GPS
Timing Receiver" ... paper presented at the 2002 PTTI meeting
which discusses the performance and accuracy (as opposed to the
precision) of the Motorola M12+ timing receiver. Motorola
dropped out of the timing GPS receiver market and sold the
design to iLotus who now manufacture it in Singapore, sold in
the US by Art Sepin @ Synergy (URLs are all in Graham's message
copied above).
    * The take a look at my "Timing for VLBI" series. I recommend
the 2007 version (there should be a 2009 update in a month or
so). In the 2007 paper, starting on Slide #20 and going thru
#24, I describe the reason that the old Oncores have ~100 nsec
peak-to-peak sawtooth "dither" while the newer M12+ shows about
¼ of the dither. In #23, you see a "hanging bridge" when the
receiver's internal xtal clock passes thru a zero-beat with the
GPS "true" clock.
    * In the M12+ (and in the earlier Oncores), the serial
binary data tells you the "sawtooth" error that will be present
on the next 1PPS pulse with 1 nsec resolution. If this
correction is applied to an M12+ receiver, the sawtooth
disappears and the resultant 1PPS signal shows only ~1½ nsec
jitter -- i.e. the 1pps pulse-to-pulse (corrected for dither)
is a  frequency standard at a level ~1x10e-9 (i.e. 1 Hz @ 1
GHz). In my paper, on slide  #4, I introduce the Allan
Deviation as a way to graphically view the relative performance
of several high-performance oscillators over time scales
ranging from 100 msec to a couple of weeks.
    * On slides #26 & 27, I describe the hardware circuit that
Rick has built into his CNS Clock 2 to provide a de-dithered
1pps signal on a connector. It works. Slides #30-38 may also be
of interest. Slides from 39 onwards are quite specific to VLBI
and should be ignored.
    * The "Improving the Performance of Low Cost GPS" ... paper
is pretty much the same as the 2007 VLBI paper, with one
exception; on slides #16-18 you will see us pondering why the
M12+'s message telling of the error on the next pulse, when fed
into the programmable delay line, results in only a ~90%
correction to the M12+'s sawtooth. To make a long story short,
Rick replaced the Maxim delay line with a similar part with 4x
the resolution and changed the algorithm that generated the
programmed correction. We found that ALL Maxim part had the
same scale error. In essence, Maxim has a definition of 1 nsec
about 10% different from the rest of the world ;-) .

Concerning the Collins/Rockwell timing receivers with 10 kHz
outputs:

    * Yes, you can use the 10 kHz signal to program a GPSDO and
it works fairly. One design I did used a ~100 second time
constant to lock some pretty good xtals. I had to take special
care in the 10 MHz => 10 kHz divider chain. I found that a long
ripple counter with cascaded 74HC390 dual 10x dividers (like
several of the designs cited here) exhibited large temperature
driven variations in the phase of the 10 kHz  signal -- upwards
of 200 nsec on the 1PPS counter stage when used in a typical
lab environment!
    * The fix for the temperature-driven drift  (which also
took up a lot less board space) was to use Tom VanBaak's
deterministic PIC-based "software" design. With one cheap PIC,
he provides all the decade steps from 1 MHz to 10 seconds with
very little temperature problems. For details, the design,
including firmware code, is available at
http://www.leapsecond.com/pages/ppsdiv/.
    * The sawtooth dither on the 10 kHz signal appears like an
interpolated version of the dither on the Motorola. It's
magnitude between adjacent 10 kHz pulses is 1/10,000 of what it
would be at 1PPS. The 1PPS and 10 kHz signals are synchronous..
    * So the logical question to ask is "can we de-dither the
signal just like the Motorola?"; the answer is NO. After
Collins/Rockwell (CR) dropped their timing receiver was picked
up by Connexant. I met with some of the Connexant engineers and
was able to garner some idea of the internal architecture.
Where the Motorola design (see slide #20 in the VLBI Timing
paper) uses a long counter with the 1PPS output based on the
integrated count, the CR (and its derivatives) uses a
programmable NCO. When the CPU in the receiver determines the
epoch of the next 1PPS tick and it sets the clock rate to steer
the clock to match GPS. In short, the Motorola et al design
matches the integrated time, the Collins/Rockwell et al scheme
changes the rate to minimize the epoch error. It's rather like
the difference between a full A/D converter (Moto) and a
Sigma/Delta A/D converter (the CR). The CR loses track of the
integrated frequency. The CR design has no integrator to inform
the user of the predicted error :'( .

I have never studied the innards of the Trimble Resolution-T so
I can't offer any comment on its performance. I did get one of
the Trimble GPSDOs via TAPR, and it's performance looks pretty
good.

Hope that provided some useful GPS Timing factoids.

73 de Tom, K3IO (ex W3IWI) -- original developer of the TAC
("Totally Accurate Clock") which yielded timing at levels
better than 30 nsec even when the DoD still enforced Selective
Availability. But that's another story!
Hope that helps.


==

 1239321736.0