[time-nuts] Sub Pico Second Phase logger
bruce.griffiths at xtra.co.nz
Sat Dec 27 02:46:33 UTC 2008
> I have been following this thread from the start & learned a lot from it.
> Thanks to you and Bruce for the insights.
> I thought I would take a look at a simple, passive approach, just to
> see what happens. I also limited my attention to stable 5 or 10 MHz
> sources & observation windows of <100s. I used a Mini-circuits
> SYPD-1 PD with a PSCQ-2-10.5 90 deg hybrid plus an ATM
> P1506 phase shifter (delta=0.32ns). There are lots of other items
> which will work just as well, this stuff happens to be on hand.
> The SYPD-1 also needs a diplexer to terminate the high order
> products & low pass filter the desired output; this is a cobble
> together item from the parts box. I used a PC logging program
> with my HP 3478A to take DC readings from the diplexer output.
The use of a diplexer perpetuates a common fallacy.
You get significantly lower noise with a capacitive IF port termination.
This is true even in the microwave region when reflecting the sum
frequency back into the mixer reduces the noise.
At low frequencies this has little effect on the Rf and LO port VSWR.
The capacitive termination limits the IF port bandwidth to a few tens of
kHz and a 50 ohm termination is more useful if an IF bandwidth of
several MHz is desired.
> Thus armed, I first checked the DVM logging with no RF in to see
> if it was stable enough to yield a useable noise floor. This first result
> was good enough. Reading 720 points over 100 seconds; the data
> was limited to -1, 0 or +1 uV.
> Then I applied 10MHz @ to the 90 deg hybrid & connected one
> hybrid output to the SYPD-1 LO & the other hybrid output to the
> ATM P1506. The ATM output then drives the SYPD-1 RF port.
> I measured both SYPD-1 inputs to be +6.4dBm. Now I was able
> to set the ATM phase to zero the SYPD-1 output & verify it was
> stable over the 100s time I was interested in. The 3478A readings
> ranged from -1 to +2uV during this time with STDDEV of 0.6uV
> -again, good enough for me.
Can you produce ADEV and MDEV plots for tau up to 1000sec or more?
This would then give us some idea of the noise floor of your setup.
SDEV by itself can be very misleading as for some noise processes it
The RPD and MPD series are better, in that they allow low frequency
isolation of the RF, LO and IF port grounds.
This can be critical for low offset frequency measurements.
> Finally, I used a HP 105B to get a quiet 5MHz source & a 1MHz
> ext. reference for my HP3336C sythesizer. All this to give me 2
> "nearly" identical. phase locked sources to measure the calibration
> factor of the SYPD-1/Diplexer output. The 3336C was set to
> 5,000,000.005Hz & +6.4dBm driving the SYPD-1 LO port. The
> 105B 5MHz output was attenuated to +6.4dBm driving the RF
> port. The resulting 5mHz data was logged & analyzed at the zero
> crossings (max gain). The result is a phase sensitivity of 1.22uRad/uV.
> The + & - zero crossings were identical and the 1.22uRad/uV factor
> is linear (+/- 2 %) up to +/- 20 % of FS. FS = 1.023V
> If my math is about right, @10MHz, 1uRad = 16fs; so a noise floor
> of +/- 2uV (+/-2.5uRad) equals +/-40fs. I expect that, if 2 sources
> are near zero phase, then it should only require <10s to estimate
> the actual phase error within 1ps, or less. Anyone agree?
> I await & hope for constructive comments and/or corrections.
> Pete Rawson
At 10MHz 1 radian = 16ns
1 millirad = 16 ps.
1urad = 16fs.
It would be more useful if you logged the DVM readings for an hour or
more so that ADEV, MDEV plots can be produced.
Eyeballing the data doesnt always deliver the full story.
If one compares a couple of oscillators one has to be very careful to
ensure that inadvertent injection locking isn't occuring.
This can happen with almost imperceptible coupling if the oscillators
are very close in frequency.
You need to use very high reverse isolation ( ideally 150dB or more)
buffer amplifiers to ensure this doesnt occur.
The first test you made only indicates the system noise floor.
When the mixer/phase detector is driven by 2 different oscillators the
noise level will be much higher that the noise level of the mixer system
In fact practically all OCXOs will/should be much noisier.
If the noise is too low this is an indication of injection locking.
One thing to watch out for is the measurement system noise bandwidth.
You need to know this as ADEV, MDEV etc depend on the actual noise
When you use a DVM it usually has built in low pass filtering as well as
its inherent sinc response due to the integration process.
It can be useful to use external passive components to accurately define
the noise response.
More information about the time-nuts