Ed,
Thanks for the many good advice.
I've tried to incorporate as much as possible, updated schematic can be
found here: http://athome.kaashoek.com/time-nuts/PNA/SSPNA.JPG

For audio into the PC I'm using a professional balanced microphone to
USB input with a noise level of -130dBc/Hz and no spurs.
Using a 7805 and 1500uF capacitors I tried to create a solid reference
instead of the buffer op amp output and that did make a difference.
Further the input of the first opamp has been change to have identical
resistors at both the + and - input to reduce common mode signals.
None of the capacitors (5.6pF) I tried to reduce high frequency gain
improved the results. Most of the time it got worse.
Removal of a ceramic capacitors eliminated the microphony.

I've added a switch to select between 0dB and 20dB gain so I can
calibrate the level by offsetting the DUT frequency while keeping the
drive to the mixer.
To calibrate the effective noise BW of the FFT I create a test signal
combining a -70dBm 10.001MHz signal with a -90dBm/Hz noise signal.
The 20dB power ratio was confirmed using a calibrated spectrum analyzer.
The FFT length and sample rate at the PC where then changed till the PC
FFT showed the same power ratio.

Noise level at 1kHz is now -150dBc/Hz and -155dBc/Hz at 10kHz which is
the spec of the DOCXO used so no need to go any lower.
Also the 50Hz spurs and its harmonics are greatly reduced.

I'll have to invest in better coax cables as the current cable seem to
leak RF.
Erik.

Erik, that looks pretty nice, with good stout bypassing. I'd recommend
adding some ceramic caps bypassing for RF, just in case. Also, I believe
the standard three-terminal regulators have no limit on capacitive
loading, so you probably don't need R1. The standard 7805 can put out a
lot of current (>1A), so I'd suggest using the 78L05 low power version
(100 mA or so). That will make it a little safer, fault-wise. Especially
when building and experimenting, it's good to have some limiting
appropriate to the circuit loads. If you should say, inadvertently short
the circuit common to ground, a way bigger than needed regulator will
supply lots of current, and possibly crash the +12V if it can't keep up.
Finally, whenever these regulators have large output capacitors, it's
good to have a reverse protection diode from output to input, in case
the input voltage is suddenly dumped - another thing that can easily happen.

So, with good DC and LF regulation, and plenty of audio and RF
bypassing, the common line should be very solid. I don't think the very
low frequency regulator noise or DC drift will cause any grief, since
the opamp's CMRR should take care of it. I looked up the OP-27, and it
shows over 120 dB CMRR up to a few kHz, so the CM to input signal
conversion is less than 1 uV/V. You should check all the part specs to
be sure if it will be OK. The PSRR is much weaker, so the cleanliness of
the +12V supply should be considered too.

I see that you've changed the LNA circuit (to inverting) and feed to the
audio (ground-referenced - nice). Also, it appears that 10X voltage gain
is sufficient, instead of the 100X previous. This should greatly reduce
the grief. One thing that's different about the inverting mode versus
the previous non-inverting, is that now you have a definite DC/LF
termination/load resistance (the 100 ohms into the summing node) on the
mixer IF port, while the non-inverting mode can provide very high input
R, so you can have more flexibility on termination R choice, from "open"
on down to whatever you want. I don't know what the ideal is, but as in
previous discussions, it doesn't need to be 50 ohms as in RF work, but
can be much higher for the audio, to get a bigger voltage signal. For
inverting mode, the input R must be all or part of the termination R,
and "open" is not an option.

I'm wondering what the PC USB audio box is. Could you please let us know
make, model, etc? The "balanced microphone" description sounds like it's
600 ohm input R.

Ed

Erik, it sounds like you have it nearly finished and working - congrats.
You should be able to greatly reduce those pesky line-interference
spurs. The first weak link may be the audio signal path, so it's time to
let the audio box help, if you haven't already optimized it. In the last
design round, you now have the audio box driven from a coax line with
its shield connected to your local instrumentation ground, which is
earth ground there, and the minus end of the 12V supply, right?

How is the other end hooked up to the audio box? I looked at some info
on the unit, but it doesn't say anything about hookup details - a
"normal" user (not us) would probably just plug in a mic and go. If it
is a balanced mic connection, then it is at least some form of
differential input, and you had been using it this way before, with the
audio common (low side) connected to your floating signal common. That's
the way you should connect the audio box still, even though the cable is
now earth ground referenced at the the instrument box. If it's still set
up this way anyway, then OK. If you have connected the cable shield to
the ground at the PC or USB audio box, try removing it. The audio box
input should have some amount of LF CMR, which should help in this part.
You can experiment with different arrangements and see what works best.
The next thing is to try adding common-mode choking with various methods

• I'll explain later.

From the very beginning, I think, you have considered the benefits of
galvanic isolation of the DBM/PD mixer ports, which is good.
Unfortunately, powering up the OCXOs and getting their signals connected
leads to some inevitable ground loops anyway, so you can't always take
full advantage of built-in isolation, but at least you have it. Try to
use it when you can.

The rest of the line and low frequency ground loop issues are hard to
predict, depending on the overall powering and operational setups, but
some general methods can be applied.

Now back to the audio output path, the first thing is to see if there's
any point to improving its CMR - if the interference is getting in
elsewhere, it doesn't matter (yet) how good it is. You can experiment
with the circuit hookup and operation. For instance, in your latest
post, the first chart shows the operating noise floor and line spurs
present "without the DUT connected." This can describe a range of
possibilities, from the RF connector being unplugged, to the DUT being
totally removed from the experiment, power-wise, ground-wise, and so on.
In your setup, you have a number of items hooked up, each contributing
to the ground loop situation.

You can apply the process of elimination to gradually improve it. The
first and easiest step is to shut everything down except the PC and
audio analyzer setup, and assess the spurs. If there's little change,
then the interference is what it is, due to the environment of the
setup. If the spurs go away, then your analyzer circuit is amplifying
them from the signal chain, or from the power supplies. Then you would
short the input of the LNA, power it up and look at the spurs again, and
so on.

This narrows things down as you go, with various tests to see what
things make the biggest difference. If the spurs are relatively
unchanged with power down, then you would unhook things until you're
left with just the analyzer circuit hanging from the cable to the audio
box - you systematically delete connections to power supplies, grounds,
signals etc, until it shows clean. Then you can add back and think about
the possible ways the interference gets in.

Once you figure out the major interference sources, you'll know what and
where to make improvements. Keep in mind that within your own project
circuit, you have lots of options, since you're building it, while on
most of the other stuff, you'll likely be limited to external fixes like
CM chokes (prayer beads) and grounding tricks.

Next time I'll talk about some of the methods in detail.

Ed