[600MRG] New GPSDO

[Dick Bingham]

Like many of us, I am messing-around with various ways to generate a stable
and accurate WSPR source for various bands.

The fellow whose comments I am adding to this note has been VERY busy and
is WSPRing successfully on 2-meters and higher. His signals on ANY band are
almost NIST-quality.

As a result, I have ordered several sets of Arduino UNO's, DDS modules,
display-modules (like Han's U3 display) and rotary encoders to try to
duplicate what the following guys are doing.

If I am successful, I will pass along my work to anyone interested . . .

73 Dick/w7wkr and wd2xsh/26
======================================
Dick -

Gardner is fine but before you jump into a bunch of servo system theory,
think about what you're doing.

Essentially the target is an LO with a very clean spectrum. That is, if you
look at it on a very good SA you'll see very low phase noise from extremely
low offsets out to many MHz. In the time domain, this corresponds to a lack
of unwanted/noise modulation from very long periods (days or weeks) to very
short ones (microseconds).

The problem is, you don't have a single oscillator with those attributes so
you have to "build" one using what you've got. What you have is the GPS
constellation as received by your local GPS receiver. Normally these have
simply outstanding very-long term accuracy. They may even have 0 error if
you consider this system (GPS time) to be a primary standard like NIST's
array of whatever-element-is-best-today clock's. BUT, as received, your
local GPS receiver which has only a 1 PPS output (pity it's so low, but
except for some of the latest Ublox chipsets which let you program the
output from sub-1 to 999 Hz, that's the 'standard') also has a bunch of
jitter on that time mark signal. So, it gives you a pulse that on average
over days is probably accurate to a part in 1e14 or something ridiculous
but which may have 10's to 100's of ns of jitter on any given pulse,
depending on the receiver, number of satellites being tracked, multi-path
etc etc.

So, what you do is you use that really good long term accurate/stable GPS
reference to discipline something with poorer long term but good mid-term
performance - say a 10 MHz OCXO. You count cycles gated by the 1 PPS signal
for some large number of seconds and then use the error, either frequency
or phase depending upon your system, to steer the long term frequency of
the OCXO to dead nuts on. This is what the HP 10 GPS 10 MHz reference,
Trimble Thunderbolt and similar GPSDOs do. You get a really good 10 MHz as
long as you don't expect to much accuracy sooner than a 50 ms or something.
The VE2ZAZ approach does this just fine. See the web page on the GPS10's I
made for more pointers. http://www.sonic.net/~n6gn/extpllpilot.html

That gets you part of the way there. The disciplined OCXO is good from long
term (close in) to mid-term (say 50 Hz offset) but isn't as good as some
other types of oscillators at shorter term (greater offsets). So, you do
the trick again. This time you use the good 10 MHz as a reference to
phaselock a better resonator, say a 100 MHz quartz oscillator, 600 MHz SAW
device or something like that. In the case of the U2/U3, you might get a
decent 125 MHz crystal (or 107+ that gives you .025 Hz exactly step-size,
per W6SFH) and build a clean, low noise VCXO. I have a schematic of a two
port one I did for some HP products that got used a bunch of places. Not
too many parts and no tuning. There are lots of possibilities.
Then you divide that VCXO by some smallish integer (division kills loop
gain and eventually kills you because of output phase noise of the divider)
and compare it to 10 MHz or some very small integer division of that. Say
you got down to 1 MHz reference for that PLL.

Now, the deal is, that with a PLL you have to keep the loop bandwidth an
order of magnitude or so below the reference. That's because if you don't,
the loop filter will not be able to keep the reference from modulating the
output and putting a sideband on the signal you're trying to generate. You
can use a notch filter or similar but even those have phase/delay down at
F/10 where you are trying to put the reference. That's why you don't go
much higher. Thus, there's a trade off between loop bandwidth, cleanliness
and reference frequency.

Put another way, this last loop is designed to have at least 50 Hz of
bandwidth since you want the characteristics of the GPSDO 10 MHz reference
that you made and are outstanding to be used to control the frequency and
spectrum of the final oscillator out to the offset where the final
oscillator is natively as good as the GPSDO. This is probably going to be
10-500 Hz. So, the final PLL has a reference that is at least 5 kHz
(probably more, and the 1 MHz is fine) and the resultant VCXO is as good as
the best parts of the constituent references: it's got the longterm of the
GPS constellation, the mid-term of the 10 MHz OCXO which is probably better
than the native 125 MHz VCXO, and the wider phase noise of the ~100 MHz
quartz (600 MHz SAW or whatever).

Having said all that, you can just go to the VE2ZAZ site, order a board and
discipline your HP 10 MHz OCXO for a shack standard. Then you make a second
lock loop with 50 Hz bandwidth to hold the 125 MHz in your U2/3 right on.
At that point, you can use the U2/3 in the normal manner or you can pick
off the 2nd image (125 MHz plus 19.. =144.489 etc) and the output signal
will be GPS referenced AND have good phase noise. You also have a really
good shack standard 10 MHz for any test equipment that takes it..

OK, after you look at my link above you can go read Gardner.

Glenn


On Sat, Apr 5, 2014 at 5:00 PM, Jim Miller <jim at jtmiller.com> wrote:

> I've been considering building my own GPSDO until this one was announced in
> the last few days. I'm going for this one and spending my time instead on a
> transverter for my K3.
>
> I'm ordering it Monday.
>
> http://www.jackson-labs.com/assets/uploads/main/LC-XO-Plus_PressRelease.pdf
>
> 73
>
> jim ab3cv (no financial interest...)
> _______________________________________________
> 600MRG mailing list
> 600MRG at w7ekb.com
> http://w7ekb.com/mailman/listinfo/600mrg_w7ekb.com
>