xor mixer

Building a phase comparator?

How about borrowing technology from radios:

Mix both signals to a lower frequency and use fast CMOS switches to do the final mixing. Google for 'Tayloe mixer'.

The first mixing does not ned to be DC accurate.

That's interesting. I want really low low-frequency phase noise, which the dual front-end mixers might trash. That's harder to think about.

I did find the HFA3101, a pure, fast Gilbert array.

Intersil's HFA3101, with 10GHz transistors.

From a project, early 2015, we evaluated...

AD831 AD834 ADL5391 LT5560 ...Jim Thompson

One thing that I think I want to avoid is AM sensitivity, which is why I may not want a linear mixer. The A and B signals will be fast differential PECL, so we'd get AM rejection by using an XOR gate or an over-driven Gilbert multiplier.

A Gilbert will have six transistors worth of base current noise, whereas a diff pair (after an XOR gate) would have three or maybe two.

IC mixers generally don't show their internal schematic, so it's hard to tell how they will actually behave.

Diode ring mixers are handicapped by their low output voltage and internal impedance Johnson noise, so have noise floors around -150 dBc/Hz.

There are literally hundreds of times as many academic papers about Gilbert cell multipliers, than there are data sheets of parts you can actually buy.

The Gilbert topology seems to be a favorite for lectures and thesis papers.

PMB2335 is a little weird but it does use 25 GHz transistors.

Re: diode rings

Check out the Mini Circuits MPD series phase detectors. You get about 1V/radian and a nanovolt or two 1-Hz noise, so you're down in 170 dBc territory.

Cheers

Phil Hobbs

That's nice. It's almost worth shorting out all the upper transistors and using the lower diff pair after my XOR gate.

That's impressive. Using two detectors and some correlation math is the next step.

All 100k ECL uses differential-pair inputs; could you use 100EP40 ?

I'd prefer a Gilbert-cell type for that, and it's gonna benefit from a thermostat enclosure. Gate mixers get AM from power and thermal sources, Gilbert cells are (theoretically) balanced but still have gain/temperature modulation.

IC transistors aren't especially fast, I'd think that hand-matching of discretes can accomplish 'balance' better than sorting through arrays at IC test time. In any case, if balance fails, it won't hurt the DC output, mainly just add some 'around 150 MHz'.

I was considering using a 10EP08 XOR gate, or, more extreme, an NBSG86A. That last one has 40 ps edges.

IC diff pairs or Gilberts are available with 10 GHz transistors. Base current shot noise will be a problem with a diff pair or a Gilbert cell.

Maybe I could use the XOR gate ECL output directly, but that's scary. Diode current steering maybe?

This is the sort of thing people spend a lifetime specializing in. I have 5 weeks maybe.

Why will shot noise (broadband) be a problem for the narrow bandpass your 'DC' signal occupies?

Noise is noise. If I have a transistor steering, say, 10 mA, its base current might be 100 uA. The shot noise of the base current will be 6 pA/rthz. That's pretty small, about 185 dB down in 1 Hz, but it's not zero. More transistors, like in a Gilbert cell, would be worse.

I wonder if finite Ft is equivalent to lower beta from a shot noise standpoint. Probably.

Analog Devices seems to have quite a few specialised chips for this kind of work. The unspecialised AD835 is still around, but the ECL high and low logic levels would show up in the output from that.

The definition of fT is, beta(freq) = 1. The dominant pole (at f3 = fT / beta(DC)) would be the traditional rolloff point (or roll-up, if your circuit is pushing harder at higher frequencies, like in a negative feedback loop), which would give you your necessarily finite bandwidth.

If you're worried about current noise, why not use FETs? ;-)

Tim

Six transistor cell will bump it up by a factor of 1.73 (square root of three); not a killer, starting from -185 dB. I'm thinking your power supply ripple is more important.

I always thought the noise figure of a diode ring mixer came about purely because of the conversion loss and the thermal noise in the various source resistances.

If you're driving it hard from low-noise sources is it really that noisy? Where does the noise come from?

Waiting to be eddicated...

It's perhaps a pity you have ECL inputs. At least on-chip, for frequency dividers, fast CMOS tends to provide better phase noise than bipolar, provided your power rail is very quiet. I don't know if this is true also of off-the-shelf logic chips though. If you get the opportunity to compare fast CMOS against the ECL in an experiment, that would be interesting.

Chris

The video resistance of a Schottky diode depends on the conduction angle among other things. Conversion loss is the sum of reflection and dissipation. Reflection doesn't introduce noise, but dissipation does.

Cheers

Phil Hobbs