[time-nuts] Phase Detectors/Mixers for DMTD and PN measurements

[Magnus Danielson]

Hi Gerhard,

On 2019-12-12 09:17, Gerhard Hoffmann wrote:
>
> Am 12.12.19 um 07:19 schrieb Bruce Griffiths:
>> Cross Correlation methods are commonly used with analog mixers.
>> Most of the high end commercial offerings use it (Holzworth, Anapico,
>> Rhohde & Schwarz, etc.)
>
>
> "High End"  is a contradiction to "Commonly Used" somehow.
>
>
> In the published timenuts world, aka NIST or the usual suspects, there
> is only
>
> deafening silence. I have seen the marketing blurbs and have even
> limited access
>
> to the really good stuff.
>
>
> Yes, there is the all-digital solution of Andrew Holme based on a
>
> Spartan-6 demo board, some ADCs and Verilog but that's it. (And I like
> it!)
>
>
> In the published analog area, there is pure joy when a 2N2222 mixer
> pushes
>
> the state of the art a few dB at 5 MHz.

You should realize that NIST is using cross-correlations all the time to
combat noise in mixers and reference oscillators. They publish a lot on
things which really is about this, I have even published with them on
the interferometric cross-correlator approach to phase-noise
measurements and to handle noise even better.

They do however not have a lot of time to write elaborate things on how
they do things, as most of the time, that is already published as such
and nothing really new have happen. NIST isn't a sales-department, so
their job is to do academic research and reference measurements, pushing
the limits in various ways.

I have a draft article lying around, maybe I should revisit it and work
on that, because the interferometric cross-correlator is really try to
manage the noise of the mixers and reference oscillator noise
back-propagating, while avoiding the anti-correlated noise of the
power-splitter.

In classical cross-correlators, the resistor creating isolation between
the ports also causes anti-correlated noise effectively subtracting
twice the noise energy of its temperature. This causes the thermal noise
as measured go through zero and end up an -300 K rather than 300 K, as
one views the absolute function, one sees a dip going to zero and up
again. For all practical purposes, one has replaced the actual DUT noise
with the noise of the power-splitter. A higher DUT noise causes the
apparent noise floor to look like it is below 300 K and you have
extremely nice properties. In such cancellation mode, other oddities
shows up making the measurement not very stable, it is actually very
messy. Removing the resistor causes all kinds of havoc since now the
noise sources is not isolated. Cooling the power divider in liquid
helium just removed that effect, but then uncovered a whole range of the
other effects.

The interferometric approach to cross-correlators (I was inspired by the
interferometric phase noise measurements that Enrico Rubiola written
about) avoids the resistor in the power-splitter, and instead aims to
create isolation by moving the noise to the imaginary axis while the DUT
noise remains on the real axis. This vector approach to cross-correlator
measurments was not used, but inspired by Enrico Rubiola as he explained
on a napkin in a wine-infused evening in Potsdam that you can observe
the shift of sign on the real axis rather than just taking the absolute
function. I then questioned the isolation just to figure out how to
solve it differently.

Since then I also lectured on the noise from mixers bouncing of the DUT,
and how that can be interferometicly isolated just the same as the
direct leakage.

Anyway, so while you might feel that NIST isn't writing much about
cross-correlations, I see a lot of it when I look at what they publish
in phase-noise. It's just a standard technique, it's there and they are
fighting the many issues of it, and mixer noise is indeed one of them.

I am greatful for NIST being able to experimentally verify the
interferometic cross-correlator, it does work. In their realization of
it, it was however hard to maintain lock. We have to work on improving
the realization, it is as simple as that.

So, mixer noise does go into cross-correlator techniques and overcome
that. The mixer noise is just bundled up with the reference oscillator
noise to become the "channel noise".

If I had more time and resources, I would work more on this than I can.

Cheers,
Magnus