[time-nuts] The difficulty of low noise measurements

[Magnus Danielson]

Fellow time-nuts,

As I now have a hydrogen maser sitting here, it triggers me to measure
things. Essentially I try to measure the stability of the maser, and
well, that will be very hard since one need very quiet reference sources
to do that, but that then triggers the question of how quiet are my
sources anyway. So, I decided to start a measuring campaign to figure
this out and to measure both ADEV stability and phase-noise stability,
and with that build reference for the future of what tools to use for a
particular measurement. It also has the side-effect that it forces me to
get more of my lab in order.

Anyway, so as I measure, I find that I have problem getting good ADEV
measure. I operate my TimePod such that I can do the three-corner hat. I
see some disturbance in the ADEV with a re-occurrence of about 0.5 s and
that part of the ADEV never converges to something useful even when
doing a 48 h test. At the upper end, the BVAs I use for reference is
clearly drift limiting, so I have to restort to Hadamard to see somewhat
beyond the drift limit, but that is not helpful. So, I decide to setup
phase-noise measurments and do that while the BVAs settles, as drift
does not obstruct the phase-noise as much as ADEV.

As I measure phase-noise of my 5065, OSA8600, OSA8601 and EFOS10 it
becomes very clear that the EFOS10 does not have even the same shape of
the phase noise as the other three, and being far worse than the
reference BVAs despite the output oscillator is a OSA8600 and for higher
frequencies this should completely dominate the phase-noise behaviors.
It comes to the point where I decide I need to break a measurement and
investigate this deeper.

So, a tell-take of the phase-noise plot is that it has some strange
spikes at about 0.1 Hz and 0.3 Hz. It then has ripples that eventually
goes smooth as the filter-bandwidth of the phase-noise measurement
increase, so it just becomes an almost smooth noise energy which slopes
down. To put this in context, it is about 30 dB higher than expected
when it is as worst. Surely, this is something other than the intended
noise of the maser.

It now becomes time to shift view again, remember we started with
long-term stability in ADEV and HDEV and then shifted to phase-noise.
So, I shift over to view the phase data, and in particular the residue
phase as the linear phase-slope has been removed (press 'r' in the phase
view of TimeLab). I then see regular phase-spikes at 10 s appart. I
measure these to be in the range of 10 ps high.

For a while I have my suspicions falsly on the PFD detector in the PLL,
as I have a compassionate dislike/hate for them, but Bob and John talked
me out of doing that hack right now. Better first collect more evidence.

Then it hits me. Could it be... what happens in 10 s interval here, the
only process going on with that type of rate is the logging of data from
the maser. So, I pull the cable and start new measurements... and that
is it. I now have ADEV operating smoothly in the region where it was
clearly fighting, and I have a nice smooth 1/tau slope down into the
abyss from the maser, clipping the 1E-13 line at tau=3 s at which time
it is so quiet that as it progresses into the E-14 range of measures it
becomes hard to measure using BVAs. Phase-noise measurement is a
spitting image of the BVA I have, and those matches up with TvBs
measurements of OSA8600.

Now, I have already took the precaution to sprinkle the cables with
common mode suppression chokes. I have not used any fancy
power-supplies, but just a standard HP/Agilent bench supply for the
BVAs, and they are being fed from the same supply even.

The EFOS-B masers was shipped with optical isolators for RS232, and this
may help for common mode and "ground loop" but not for the transitions
themselves. I will have to investigate this further to figure out how
the main action that makes the bursts of data on the serial port to
appear in the measured 5 MHz, and then how to isolate it. Until then I
now know I need to pull that cable for precision measurements, but that
is not a satisfactory solution as I want that log-data as an intact series.

I send this note, to also indicate the importance of monitoring the data
in different views, that phase-noise measurement may be giving a
significant clue as to what goes on, and then the phase (and/or
frequency) view to further locate it. Amplitude noise and pure spectrum
analysis may have been additional methods to also consider, just to
illustrate that one needs a wide selection of tools to locate the root
cause of a disturbance and mitigate it. I've heard people say "I'm only
interested in stability", well, that may be the end goal of the
measurement, but to get there and to improve the setup other measurement
discipilnes may be necessary.

Thanks goes to Bob Camp and John Miles as discussion partners, and to
Ole Petter Rønningen who helped me a lot with the EFOS10 maser.

I will continue my measurement campaign, and see if there is more things
to clean up. I will also investigate more on the serial link.

My initial measurement campaign is to just establish roughly the noise
of various sources and for various measurements, in order to be prepared
for a second round allowing for more qualitative measurements with the
best sources. This also raises the awareness of those sources having
issues, so that I will for instance work harder to look at my 5065s, put
them all into operational state and see what is the main limitations in
their performance.

The main point being, it's fun to be back in a more functional lab again. :)

Cheers,
Magnus