[time-nuts] HP3561 and Phase Noise calibration

[alan bain]

It seems from the list archives that people are familiar with the 3561
and its use in the 3048 PN measuring setup.  The PN calibration
depends a lot on sweeping out paths with periodic white noise and
making a single FFT measurement (the noise comes from 3561 source into
the 11848A interface, a path is set up with the various internal
switches and the noise measured at the 3561 output). With the source
being periodic there are no issues of windowing, or need for averaging
to get usable results.

I happened to have a 3563 to hand and thought it would be interesting
to make it do the same job (as it has two channels it also can do
cross correlation).  In doing this I ran into a problem with the
calibration e.g. for CALLO.The 3562/3 seem to have dropped the
periodic random noise, leaving only long period pseudo random noise
which is much less suitable.

I vaguely thought it should be possible to emulate this using the
arbitrary source and indeed it is at full bandwidth uploading some
suitable white noise data, BUT for reduced spans the 3562/3 continues
to sample at 256kHz but repeats each point from the DAC multiple
times; so in effect this is upsampling followed by a FIR with all
coeffs 1. Trivial calculation shows this extra low pass filtering
gives rise to about a 2db loss in a 4kHz sweep which is what I
observed (two separate sweeps, one 100kHz span, one 4kHz with same
Y-axis).

https://imgur.com/QWGResS

Now I can work around this by applying suitable HF boost to the
frequencies in my noise sample (which is just an inverse fft of a load
of magnitude one coefficients with random phase, with suitable
conjugate symmetry to ensure a real output) but I need to use a
different noise sample for each span which is painful, or maybe
simpler - correct in software.  But I get a distinct feeling I'm
missing something as I cannot see why HP would drop their favourite
measurement technique unless it had been replaced by something better.

Looking in the 3561 the pseudo random noise is just obtained by
dumping an EPROM looks like a 27128 or similar into the DAC, with a
counter clocked at ESPx4 (no precise ESP definition other than at
100kHz bandwidth it is 256kHz, so I presume it drops as the bandwidth
drops).  Without a dump of U701 I cannot be sure, but I doubt it
repeats each value 4 times, so it is likely oversampling to move the
knee up out the way.

Does anyone know what happened to the PERN "periodic random noise"
source in the 3561,  since it clearly wasn't the cost of an EPROM and
counter.  Have HP assumed that with two channels it is easier to use a
swept sine chirp to measure transfer functions?

Alan