Hi

There have been several threads on the SYPD-1 and it’s cousin in a different package the RPD-1. Both are designed targeted
as phase detectors rather than mixers. While you can never say “always” the best performance generally is to drive both
RF ports hard from matched sources. Terminate the IF port in a high impedance (> 5K ohms) and run a low noise op amp after
it. My preference is to run a positive gain amp. As others will caution - this risks blowing the amp in the case of a power supply
failure. I’m careful with power supplies ( = have not blown one out that way in 30 years).

With something like a gain of 10 in the op amp stage, you will almost clip / will clip the output running unlocked. Obviously if it
clips, drop the gain or increase the supply voltage. The open loop part lets you calibrate the phase slope of the detector. You
need that to evaluate the data you get.

For phase noise, you normally run a single mixer in phase quadrature. Easy way to do this is to feed DC out of the op amp into
the control port on one of your sources. You likely will need gain / attenuation in this path as well as an offset voltage.

Once in quadrature, you feed the output of the op amp into a sound card or spectrum analyzer. The noise you read out is (after
a bit of math based on the calibration) the phase noise of the sources. You can get well into the  160’s (and maybe further)
with a setup like this.

Bob

Hoi Jerry,

On Tue, 18 Dec 2018 23:18:42 -0800
jerry jerry@hanler.com wrote:

I guess you are looking more into phase noise than stability?

All phase detectors exhibit non-linearity in some form or other.
If you get all the way to +/-176° with good linearity, then be
happy. If you need more than that, then you have to either use
two phase detectors that are driven with a reference that is 90°
apart, or you have to digitize the signal and do the same in digital.

There is a decent alternative: Use two transformers and 4 2N2222 to
do your own mixer. The quality of it, according to NIST[1] is pretty
good. I assume that any other NPN transistor of similar make-up will
also do and lead to same/similar performance.

You might also want to have a look at Enrico's Mixer Tutorial[2]

Jürg Kögel did a redesign of the Riley DMTD earlier this year,
which looks very nice. And the measurements I've seen are very
impressive (it's on par with the best state of the art I am aware of).
So you might want to contact him and see whether he has some board left.

		Attila Kinali

[1] "Residual PM Noise Evaluation of Radio Frequecny Mixers"
by Barnes, Hati, Nelson, Howe, 2011
https://doi.org/10.1109/FCS.2011.5977868
http://time.kinali.ch/mixers/residual_pm_noise_evaluation_of_radio_frequency_mixers-2011-barnes_hati_nelson_howe-05977868.pdf

[2] "Tutorial on the double-balanced mixer"
by Enrico Rubiola, 2006
http://arxiv.org/abs/physics/0608211

It is upon moral qualities that a society is ultimately founded. All
the prosperity and technological sophistication in the world is of no
use without that foundation.
-- Miss Matheson, The Diamond Age, Neal Stephenson

Thanks to all that replied, lots of good info which will take me months to sort through, especially the link to Jürg Kögel looks helpful as well as a note from Bert.

Attila, Yes, more phase noise than long term though this is really an exercise to get a better understanding of oscillators and measurement techniques.

The 2n2222 mixer sounds like another H-mode, I’ll have to look at it.  It probably uses self-wound, micro-sized bobbins which will be a challenge for me :-].

Bob, I think you were the person who pointed me to the loose PLL a while back.  The circuit I am building used a gain of 300ish (30k/100) and I expect that to really clip when unlocked.  I have a 10-turn variable in the circuit to bring it close initially.

Another question or if you can point me to the method, if you are using quadrature signals, is the measurement then the geometric sum? Or do you mix them back together?  I am using a 3457a on the PD after an amp right now and I would think you would need either another meter, which I have, or a scanner, again I have both.

My N2PK VNA uses two 24bit, low noise ADCs (LTC2410) for the same purpose more or less. It has effectively two DC receivers with two DDS. I can throw away the DDS and mixers and just read the DC voltage.  I have a second N2PK board I never finished as I planned to use another master oscillator on it with lower PN, but I needed a couple of parts and my tremor made SMD soldering difficult.  This was one of the drivers for my interest in phase noise. Now that I think about it, I can repurpose the second board, bypassing the MC1496 mixers. They are filtered to be close to DC within a few hz and has -135dB noise floor.  I can even change the input filter to use the 220 ohm/.0046uf R/C recommended.

I wish there was a way to use or modify one of my 8568B Spectrum analyzers to do this.

Regards,

Jerry

Hi

If indeed the only purpose here is to do a phase lock, then this is being way over complicated.
There is no need for anything past a simple mixer followed by the loop amplifier. There are always
going to be minor bits to keep RF out of here or there, but they very much depend on exactly
how you are doing things.

The “why” is pretty simple - any TimeNut category source is not going very far when in phase lock.
They will be quiet and there is no modulation to worry about. If you are out past 90 degrees, you
aren’t locked ….

====

The “typical” way to play with the baseband I/Q stuff is to digitize both and then do the math in the CPU …
Yes, that makes a lot of assumptions about the end goal.

Bob

I had once some leftover space on a home-etched test board and put

one of these BJT mixers in a corner; just for fun, I had no application
in mind

so they collected some dust.

Transformers are 1:4 Macom MABAES0061 or Pulse Engineering CX2047 or
sth. like that.

avail. at DigiKey for $1.50 . Transistors are BFR93 since they happened
to be on the table.

I checked the PD just now since it was mentioned.

Optimum return loss seems to occur at 9 dBm, the sweep goes from 2 to
100 MHz.

10 MHz is the sweet spot at this level. I did not calibrate the VNA.

Cheers, Gerhard

Am 19.12.18 um 16:57 schrieb Attila Kinali:

NIST used off the shelf RF transformers in their mixers based  on diode connected (collector connected directly to base) 2N2222. The transformer impedance ratios (!:5 IIRC) are specified in the papers on the mixer performance.
Suitable transformers are made by Minicircuits and others.

Bruce

Thanks again.  I have some MCL T1-6T transformers sitting here so maybe they will work.

Regards,

Jerry

Gerhard, would there be any advantage to matching the transistors?  I have a pretty accurate curve tracer.

thanks,

Jerry

Hi,

matching in the sense of selecting 4 equal ones won't help much with BJTs.

Take them from the same tape, that's enough. Also, the transformer ratio

does not matter much. Once VBE = 0.7V, the transistors switch on. The
authors

write that thy use the BE junction as diodes, but that is not true. The
transistors

are operated as switches. That's probably the reason why they are so good.

A different transformer ratio only shifts the level where the 0.7V Vbe
is reached;

a dB more or less, not a big thing.

The power required for optimum return loss seems to be quite precise.

If you do not have enough for Vbe=0.7V, the switches will sit there
idle, giving

bad return loss; having too much pumps a lot of current through the ring and

the mismatch will be to the short circuit side.

You could also drive them somewhat into saturation which would make

them slow. That's why I used a fast-ish transistor from the start.

In my case, 8 or 10 dBm made quite a difference from the 9dBm for
optimum BW.

I have some MCL T1-6 and could check them over the weekend.

If you have more signal power, it's probably wise to split it to more PDs

and add the IF outputs.  This here seems to be an interesting Wilkinson:

<   https://www.digikey.de/products/de?keywords=1465-1815-1-nd    >

So it all boils down to el cheapo coils and Schnaps or beta-blockers for
soldering.

In the case of JFETs as switches, your curve tracer could probably help
a lot.

JFETs are all individuals, see Ic over Vgs for IF3601 / IF3602

<
https://www.flickr.com/photos/137684711@N07/37321004540/in/album-72157662535945536/
   >

The IF3601 is a large geometry JFET with huge gm, that can deliver 0.3
nV/rtHz voltage noise.

Four of them in parallel yield 160 pV/rt Hz; I could verify that but
they were not

completely stable for inductive signal source impedance. If you want to
simply parallel them,

one of them might happily sink 1A while the other is still completely
cut-off.

They also seem to need more drain current than promised and the IF3602 duals

are not pairs in the closer sense. But no opposite outliers, OK.  
(avail. @ Mouser)

My newest creation uses a bootstrapped cascode to effectively remove the
LARGE

input capacitance; that also seems to make it unconditionally stable, at
least in simulation.

The new boards are spending this week in the Leipzig, DE customs office
since someone

does not believe that PCBWAY in China can deliver 10 proto boards for
$5.  :-[

Having such an amplifier after a ring mixer is probably overkill; a RF
Schottky diode

may have easily a ohmic 50 Ohms component in series and two of them
active in a ring

mixer guarantee 1.3 nV/rtHz thermal noise alone. (and that is NOT the
half-thermal slope

resistance of the diode effect itself)

High power type3 ring mixers have additional resistors in series to the
diodes to create

some back-bias. Those also create noise and so it should not come as a
surprise that

high power mixers may be kings of IP3 but lose big time at the noise end
of the scale.

<
http://home.deib.polimi.it/svelto/didattica/materiale_didattico/materiale%20didattico_MRF/appnote/wj_Mixers_part_2.pdf
   >

regards, Gerhard

Am 20.12.18 um 03:16 schrieb Jerry Hancock: