[time-nuts] TimePod, cross-correlation fun and measurements

[Magnus Danielson]

Fellow time-nuts,

As you know, I spent last week in Boulder. NIST T&F seminar, fellow 
time-nuts and much fun.

I also picked up my TimePod out of the hands of no other than John Miles 
himself. Being sleep deprived from the travel, progress have been slow, 
but I have now come to the point where I got a bunch of measurements 
done. This has been greatly aided by the excellent cables and adapters 
provided by Tom Knox.

I did a comparison of my OSA8601-02 SN1314 and OSA8600-1 SN817. Some 
material to let you know what these are, see:
http://rubidium.dyndns.org/~magnus/time/oscilloquartz/8601/

Essentially the 8601 is a different interconnection mounting than the 
8600. I tend to refer to them as 8600s as they are essentially the same. 
However, my 8601 is an older generation than my 8600. We will come back 
to that.

I have made three essential measurements:

OSA8600-OSA8601 pair (one as input and one as reference)
OSA8601 with dual reference (OSA8600 and HP5065A)
OSA8600 with dual reference (OSA8601 and HP5065A)

With the first measurement, I get the sum of the noise-levels.
With the second and third, the noise of the other sources 
cross-correlates out to some degree and the individuals noise should 
remain. It works to some degree. Let's see how the noise-floor 
measurements look:

http://rubidium.dyndns.org/~magnus/time/timelab/OSA8600_HP5065A_OSA8600_20120618_1.png

The green trace is the sum of them both. This is best seen from 10 Hz 
and above.

The blue trace is the OSA8601-02 AT-cut BVA.
The purple trace is the OSA8600-1 AT cut BVA.

The overall plot shows the typical f^-3 noise, some f^-1 noise and the 
white noise. The white noise is around -155 dBc or better and is 
essentially flat from 1 kHz to 100 kHz, except for the 20 kHz bump. This 
bump I do not know the source of.

In the 7 Hz to 500 Hz region, the f^-1 noise of the 8600 dominates the 
sum (green), where as below 7 Hz the 8601 f^-3 noise dominates.

Considering that their Q values should be roughly the same, and hence 
their break-point, it is fair to assume that the white noise of the 8600 
buffer amp is worse than the 8601 buffer amp, but the amplifier core of 
the 8600 is better than the 8601 amplifier core.

Doing a Hadamard analysis with linear drift:
http://rubidium.dyndns.org/~magnus/time/timelab/OSA8600_HP5065A_OSA8600_20120618_2.png
and without linear drift:
http://rubidium.dyndns.org/~magnus/time/timelab/OSA8600_HP5065A_OSA8600_20120618_6.png

Shows kind of OK values, with the flat part of ADEV around 3-5 E-13.
It's obvious how the 50 Hz breaks through, and that spurious was clearly 
visible in the phase-noise plots too. To maintain that low I kept lights 
off in the lab (with me walking around the disaster area with few visual 
cueues). Obviously I need to work on that. The ADEV is clearly 
school-book with a 9E-14/tau slope and then level out at about 4E-13 and 
then go steep up at tau or tau square depending on the linear drift 
being removed or not. This prooves that the frequency drift is not 
linear, and that Hadamard is not able to remove the effect.

Looking at the phase difference:
http://rubidium.dyndns.org/~magnus/time/timelab/OSA8600_HP5065A_OSA8600_20120618_3.png

and frequency difference:
http://rubidium.dyndns.org/~magnus/time/timelab/OSA8600_HP5065A_OSA8600_20120618_4.png

Reveal that this is not matching up very well with the linear drift 
assumption, but there is clearly a systematic effect in here. A better 
frequency drift model should be applied, and it can be assumed that it 
pollutes the xDEV data, especially in the long-term.

Finally a look at TDEV
http://rubidium.dyndns.org/~magnus/time/timelab/OSA8600_HP5065A_OSA8600_20120618_5.png

For short times, the time-stability can be as low as below 20 fs RMS. It 
is clear that 50 Hz noise is a threat for that performance. At 1 s they 
all remain below 300 fs. 1 ns stability is in the range of 100-400 s or so.

Actually, there is many improvements in how these measures are being 
done, but it is a nice example of how cross correlation allows one to 
measure below onces oscillators noise. The HP5065A is even noiser.

I could for instance not trim the OSA8601 for optimum frequency, so the 
measurements of the OSA8600 may suffer from tracking issues. This could 
be the explanation that the f^-3 noises isn't as good as they should be.

Also, the OSA8600s wasn't separated physically or power-wise.

It's pretty good numbers in there. -121.9 dBc and -147 dBc for 1 and 10 
Hz is pretty amazing numbers, for a home lab.

Suggestions for improvements is welcome. The long measurements was for 1 
hour.

Now, nobody donating a H-maser or two?

Cheers,
Magnus