[time-nuts] Practical considerations making a lab standard with an LTE lite
SAIDJACK at aol.com
SAIDJACK at aol.com
Sun Nov 23 15:46:32 EST 2014
this is the kind of lively discussion I was hoping for! I enjoyed this.
Some comments (these are my opinions only):
* Thanks much for Tom publishing the plots, and spending a lot(!) of time
evaluating and helping improve the units significantly. Tom's unit was a
pre-production unit. We added RTV (some units black, some a combo) to the TCXO
production units based on his suggestion of the successful TP
modification. The RTV will help keep airflow away, but additional shielding will help
* There is a point where thermally stabilizing the unit does not help
anymore. I suspect that point is reached shortly before burying the unit 50
meters underground :) At the point of diminishing marginal returns the GPS and
loop noise will be larger than the thermally induced phase offsets. Also
local heating from the GPS receiver (which is not constant) will swamp
external thermal effects at some point. For us in our lab, the point of
diminishing returns is reached when we simply slide the unit into its ESD packaging,
then put some pink ESD padding on top of it. With that simple shielding
we can get ADEV at 5x to 8x its rated 1ppb performance out of most units.
* Temperature changes are typically not the problem with TCXOs, simple
airflow and convection turbulence is what causes most of the phase drift
problems. As shown by Tom simply putting a layer of TP on top of the unit made a
huge difference in stability by keeping convective flow away from the TCXO,
while it probably did nothing for temperature insulation. These convective
flows are very fast and high-frequency so inside the GPS loop time
constant, whereas temp changes are usually easy to low-pass dampen to the point
that the GPS loop will hide them.
* Actively heating the units' enclosure to some stable temperature is
counter-productive in my opinion for two reasons: first higher temperatures
cause convective airflow inside the enclosure. We want as little convective
flow as possible. Second CMOS slows down at higher temperatures, and noise
levels go up with temperature. As mentioned before temperature changes (other
than instant changes such as when the sun almost sudden hits the enclosure)
usually are easily low-pass filtered to be slower than the GPS loop time
constant which is below a couple 100 seconds, so keeping the enclosure at
some high temperature is probably going to make things worse. There are other
items to consider such as the AT-cut TCXO crystal probably has its most
stable operating point at around 25C, and the lifetime MTBF of electronics
typically gets cut in half with every 10C Degree increase in temperature.
* The 10MHz units have a different RF output than the 20MHz units. The
20MHz units have a 50 Ohms series-terminated and buffered RF output, while the
10MHz units have the TCXO output drive the MMCX connector directly without
series impedance matching. Both drive the line with 3.0V CMOS levels. This
means the cable on the 10MHz unit should be kept as short as possible, and
that impedance matching for maximum power-transfer is not required nor
desired. The suggestion that Charles made for checking the impedance by
progressively loading the output more and more is valid for Sine Wave outputs, but
not for CMOS outputs as implemented on the LTE Lite. One issue is that the
TCXO is driving a 1.8V CMOS input through a capacitive voltage divider,
and if you load the TCXO so much that its output voltage goes to 1/2 the
no-load voltage then the input of the processor will likely not get enough
voltage range to operate properly.
I mentioned 1M Ohms input impedance simply for convenience as it is a
standard input impedance as Charles mentions. You can significantly reduce that
impedance since the 10MHz TCXO can drive a handful of mA no problem, and
the 20MHz buffered output can drive 20mA or more. This means a 1K Ohms load
is also no issue as it would load the output only with 3mA, however(!) the
more you load the CMOS output the more heating will happen in the 3.0V
linear regulator close to the TCXO and inside the 10MHz DIP-14 TCXO. This will
cause load-induced instability.
The best input for the LTE-Lite output is simply a 3.3V or 5V powered CMOS
gate. No input termination resistance required. Cable lengths should be
kept short (less than a foot) to prevent ringing and loading the TCXO output
for more than a couple of nanoseconds as the edges traverse into the coax.
I like to put a weak pull-down of 470K to 1M on those CMOS gate inputs so
the input does not float when its not connected to anything. There is
absolutely no need to load down the output with 100 Ohms, 1K, or even 10K. For
CMOS inputs, the only thing that makes a difference in phase noise seems to be
the rise/fall time and voltage swing. The faster swing and higher voltage
the better. Loading down the output will reduce this voltage swing and
slow down the internal CMOS switching process, and thus increase phase noise.
Simply use a 74AC04 buffer as your 10/20MHz receiver/buffer, and keep the
coax feed to less than 1 foot and you should see the performance we had
measured in our plots. No need to over-design this. Using a couple of the
74AC04 gates in parallel with each going through its own series resistor to
increase the drive power then combining the outputs of several of the 74AC04
gates, then adding a series capacitor to get rid of the DC offset will allow
you to drive 50 Ohms terminated systems properly. If you use four gates in
parallel then use 180 Ohms or 200 Ohms series resistors on each gate and
that will combine to close to a perfect 50 Ohms output impedance with very
high current drive capability. Any noise on the power supply for that IC will
end up as AM noise on the output signal, so an ultra low noise LDO is
suggested (that's NOT a 7805 regulator!).
* On the MMCX cables, we were informed that the factory sent a number of
incorrect straight versus right-angle cables by accident. I sincerely
apologize for that. They will work just as well electrically, but may not
mechanically fit into the Hammond enclosure we mentioned. If you like please let us
know and we can replace your straight cables with RA cables for your.
In a message dated 11/23/2014 11:26:37 Pacific Standard Time,
dave.martindale at gmail.com writes:
Did you use one-ply, two-ply, or three-ply TP?
More seriously, your LTE-Lite differs in a couple of respects from the
batch of "production" ones, or at least my example. Your TCXO seems to be
in a metal package (shiny gold colour) and open to the air, if I'm
interpreting the photo on your LTE-Lite page correctly (and also the photo
that Said posted in his divide-by-two document). The production units have
the TCXO in a solid black package, probably black epoxy, with a blob of RTV
rubber on top. So the "production" units are probably already somewhat
better shielded against drafts.
(Thanks for doing the tests, particularly for those of us who can't do
these tests ourselves. I can only watch the 1 PPS of the LTE-Lite wander
with respect to the 1 PPS from my old Thunderbolt (Piezo oscillator), and
look at the worst-case variation, but I have no way of knowing how much of
the drift is due to each GPSDO).
On Sun, Nov 23, 2014 at 11:24 AM, Tom Van Baak <tvb at leapsecond.com> wrote:
> The short-term performance is 10x worse if you don't shield the TCXO from
> air, even if the ambient air is "still". I suggested Said sell the
> with some sort of engineered shield in place. Instead each of us will
> the problem in our own way; which is ok for a dev kit.
> For plots and photos showing performance with, and without, and with
> insulation see:
> The difference is dramatic, especially if you are used to working with
> OCXO where this sort of effect does not occur.
> The insulation may be found in convenient rolls at many local stores. I
> used TP, which for this application is an acronym for Thermal Paper.
> time-nuts mailing list -- time-nuts at febo.com
> To unsubscribe, go to
> and follow the instructions there.
time-nuts mailing list -- time-nuts at febo.com
To unsubscribe, go to
and follow the instructions there.
More information about the time-nuts