[time-nuts] SR620 Floor Noise

John Miles jmiles at pop.net
Thu Sep 22 15:28:58 UTC 2011


> My measurement technique is from Stanford's own application note.
> 
> The two 10 MHz are fed to Ch A and B.  The measurement is a time interval
> gated at 1 kHz from the SR620's ref out.
> 
> So we make 1000 measurements per second.
> 
> So the one shot 25 ps SR620 spec is reduced by square root of 1000.  So
the
> theoretical floor noise is 0.8 ps per measurement.

But to borrow one of Bruce's colorful idioms, what you are measuring is not
ADEV, but "SRDEV."  :)   SRDEV may agree well with ADEV for some noise
slopes, but it will not be good for absolute readings in the general case.
And an 0.8 ps floor is still nowhere near low enough to measure most
amplifiers.

You might try configuring it to average 10 readings per second via that 1
kHz gate, or perhaps 100.  100 should get you most of the way to 1-ps/sec
without excessive harm to the ADEV transfer function.    From what I've
seen, ADEV is not substantially biased if the TI averaging process takes
less than 10% of the tau-zero interval... but I haven't seen that kind of
broad assertion in print anywhere, and wouldn't be able to defend it if
challenged.   Averaging over 100% of the tau-zero interval is definitely not
ideal if your goal is to measure ADEV.

> I also have plenty of 3048A phase noise test sets and I also use them to
> make ADEV measurements.

That's the right thing to use to measure amplifier phase stability, then.
What I described in the other post *is* the 3048A's 2-port measurement
scheme, except they use an FFT analyzer instead of a voltmeter. 

Think about what happens at the block-diagram level when you use a counter
to characterize an amplifier.  You are essentially using a comparator with
high open-loop gain, which has to make zero-crossing decisions that will
vary with slew rate and other factors, and expecting it to outperform an
amplifier with no such constraints.  Sure, TI averaging can help under the
right conditions, but it's not enough.

(A DMTD with an offset oscillator could be another thing to think about, but
I've had no personal experience with those.  They don't strike me as the
optimal solution for 2-port measurements, for a couple of reasons.)

-- john, KE5FX





More information about the time-nuts mailing list