VHF discrete transistor mixer, matching + how to?

[Yansi]

Hello,

Let me introduce to my problem: I am trying to make a HF frequency synthesizer "the old way", as a proof of concept. I know there are better ways to generate clean few tens of MHz these days, but I value more the knowledge I get from this and the experience.

I have a two VHF oscillators near 100MHz, that are buffered (to achieve near 50 ohm outputs), then low pass filtered using a 5 (or 7) pole cheby filter. The characteristic impedance of the filter is 50ohm, but I may change it to higher impedance (say up to 200ohm)  for better match if required.

What I am missing now is a simple way to mix the two VHF signals and to get the intermediate frequency of 0-30MHz. What is the easiest way to mix these (using discrete components), with reasonable linearity not to make unwanted harmonics at the output? How to match the mixer to the outputs of the VCO filters?

I can only think of a passive diode mixer or active transistor mixer, for example using a NPN, feeding the signals from base and emitter. Is this a suitable solution?  If yes, how to design the circuit and match the three ports to 50 ohm (or whatever more suitable impedance)?

Principle schematic:


Thank you for help,
Y.

[Yansi]

I did find myself a good dose of reading: https://www.nxp.com/docs/en/application-note/AN238.pdf

However I do not understand much of it fully yet.

What I find useful is their claim that by injecting the LO into the base, one can achieve better stability easily and that it is easier to implement - which I find correct, after reading through it. However injecting LO into the base still leaves me with only two ports matched while the LO being "just injected" into the base through 1pF capacitor which they have "found suitable".

To start with, I need the admittance/impedance values for the desired transistor. Which sounds really nice on the paper "First  obtain  admittance  parameters  at  the  desired frequency. " however am not sure where to get these for general purpose transistors, like BC547 and such. I know that RF transistors come with s-parameter tables, however these are not precise enough at the low end of frequencies.

How to get the Y/Z values in general?

[T3sl4co1l]

Well, you're obviously going to get a lot of crosstalk with that circuit -- what consequences are there, if LO goes up the RF port, and such?  What about mixing products (because the B-E junction is a diode, it's doing the mixing on the low side already -- the collector output is just the amplified current from this process) -- is there a suitable impedance for them to terminate in?  Is it better with a low impedance?  (It might -- more signal current at harmonics and products?)

And how much gain and NF do you need?  Do you need to conserve LO power, or is this pretty much anything goes?  If you don't have many restrictions, you might as well just do it and be done; make up the gain in the IF stage, and you'll get what you get.

If you're looking for an optimized design, you'll need to instrument this anyway -- reflectance bridges, matching networks, maybe diplexing filters, stuff like that -- to figure out what the impedances are at each frequency, and to measure overall gain and noise.  A starting point is about all you can do on paper -- I don't know of any large-signal average-AC analysis methods that you'd do by hand.

Tim

[Yansi]

Crosstalk should not be big issue, the RF port is in fact a second LO.  (Note: I need a PLLVCO, however it is not possible to make a VCO with a range of 500k-30MHz, hence two VHF LOs mixed to get the desired difference signal of 500k-30M)

The overall paper principle is below. What I have now is two buffered oscillators, with 5 or 7 pole low pass filters after them. I can get like ~0dBm signal out of them, or more if required.

Now I am seeking a solution to mix the two oscillators to get the difference frequency. I need both inputs to the mixer to be matched, for the LO filter to see a correct load impedance. I am stuck here and do not know how to move forward.

I have found ways to design the mixer, I even could calculate the small signal matching networks by hand, if I could have those bloody Y/Z/s parameters of any of those general purpose transistors. Otherwise I am stuck using guesswork and hours of breadboarding, trying to knit a whip out of horseshit. 

There are a few options:
1) find the Y/Z/s parameters required
2) smh calculate guesstimate the Y/Z/s parameters
3) find a different circuit that would not require unobtainium parameters as design input
4) abandon this project yet again due to lack of component data and/or knowledge

Large signal AC analysis would be nice, however I have tried once. Conclusion: There is none (to my best knowledge) free SW that would actually work (large signal AC analysis). There is only plenty of paid, professional software, I have no access to. So screw me, no solution to that.

How much gain? I have no specific criteria for that. If too low gain, I just will need to slap an amp or two in there. The only requirement is a discrete solution, low distortion (so the 5/7 pole filters after the LOs will be actually usefull - the LO signals are pretty clean).

y.

[PA0PBZ]

Not sure what levels of input signal you have but a double balanced diode mixer like a SBL1 is hard to beat for simplicity and performance, and it's 50 ohm. The only thing is that it has a conversion loss of about 6 dB, but like you mentioned an amp will solve that.

[Yansi]

Yeah, SBL1 is also hard to beat on price.    So, definitely a no to a SBL1. I have one or two laying here, but for this application, really please nope.

//EDIT: Just looked, they are for couple bucks a pop at ebay - interesting. :-0 I bought myself those for about $25 a piece years back... But still no. Too nice component for such project.

I'd like to attempt a discrete solution, not an integrated one. Even though it is merely impossible, at least it will bring in some new knowledge and experience.

[PA0PBZ]

Well, if you want it discrete it's not rocket science to make your own DBM, after all it's just 4 diodes and 2 coils 

[Yansi]

But it is rocket science to make it work, as far as I know

1) matched diodes (not really that required)
2) transformers suitable for 100 MHz (definitely required, not sure if I have suitable cores, maybe)
3) high LO drive required (guesstimate of +5 to +10dBm)
3) impedance matching impossible, wideband termination required instead (diplexer).

You tell me, what is easier, I have not much positive experience with making custom DBMs. Every one I built sucked. (LO power and my patience)

But sure, it is one way to go, but would need characterization beforehand (conversion loss, LO power, IP3) - that is a bit problematic due to my limited equipment. Doh!!! 

[KJDS]

A dual gate FET mixer will be the easiest way to make progress.

[rfeecs]

How about MC1496?
https://www.onsemi.cn/PowerSolutions/document/MC1496-D.PDF
edit:  Probably doesn't go high enough in frequency.  I guess there are higher frequency Gilbert cell type parts NE602, NE612, HFA3101, but a bit more expensive.

Your single transistor mixer is basically just a single diode mixer with amplification.  You can use the SPICE model with LTSPICE and the .NET directive to get S-parameters, Y-parameters, etc.  Of course these are small signal parameters, not what you really need, but it is a starting point.

[ferdieCX]

You could use a BF254, the datasheet has the Y parameters.
If you want it to be discrete, it is also quite easy to build a mixer using a differential stage with 3 transistors

[T3sl4co1l]

Okay, so with buffers and no particular dynamic range issues, you can use any mixer you like, even a single diode.

For a single transistor, you might instead just use a series resistor from each filter (which still need to be terminated, mind) to the transistor base, and not worry about coupling into the low impedance emitter.

Single balanced mixers are easy, you only need the one transformer (a few ferrite beads will do, by the way).  Use a constant-resistance filter on the IF port (or, a parallel terminator, and use an L-input type filter).

As for admittance or what have you -- not tabulated?  Go back to the old fashioned hybrid-pi model.  Tack on capacitances, and set the gain to a single-pole roll-off (this reasonably models recombination in BJTs, but keep in mind a diffusion model is more accurate for MOSFETs -- they don't tank at some frequency, they just roll off, gradually getting "heavier").  Crank the numbers, and out comes your values.

Conversely, whenever you see s-params, or h or y or whatever, you can mentally convert the curves, into impedances, into equivalent RLC parameters.  This is reasonable to do on RF transistors in the low frequency range (say, below a gig), but gets iffy at high frequencies (where higher order structure of the device itself matters, and is therefore harder to model).

But mind that you need to do cycle averages, because you're doing large signal, not small signal, analysis here.  You might assign a bias point and do small-signal around that, but the bias point will shift due to applied AC (say if you're driving more than 50mVpp into the base), and further shift the averages.  Umm, I would guess the impedance will generally rise, while shifting more capacitive (spending more time at less bias also reduces the average capacitance), but who knows.

You can always plug the values into SPICE and run a transient sim -- you have to create your own time-domain instruments to do the AC steady-state measurements, but it's a far sight easier doing that with arbitrary functions, than building those instruments IRL.

Tim

[phenol]

Dual gate mosfet, as already suggested ... it is widely used as a mixer in portable vhf/uhf radios and lends itself to broadband matching.
or NE612...
or schmick DDS. synthesizing a clean reference is quite a challenge in it’s own right...

[Wolfgang]

If it has to be as retro as you described, the pilgrim fathers of mixer technology used a dual gate MOSFET called the 40673 (60ies and 70ies). Its fairly foolproof,
has gain instead of loss, little coupling between the gates, protection diodes so you dont kill it instantly, an acceptable noise figure up to VHF, ...

Older ARRL handbooks are a good source of example circuits. The 40673 is available on eBay or some electronics dealers for ca. 2€.
A BF960 would also do the job.

The mixer you described works, but has lousy properties in almost all aspects.

Much fun experimenting !

[chris_leyson]

Old school, use a diplexer on the output and dump anything you don't want into a resitive load.  DBM would be wideband transformers and Schottky diodes, I think Murata still sell some of the wideband TOKO transformers and the HLMP-2800 series is still available from Broadcomm. What ever you use as a "switch" you have to switch it fast for minimum noise figure and that implies a high LO drive power. You can use transistors in your mixer instead of diodes thus reducing LO power, Adret 7100 signal generator is one example and Rockwell/Collins HF2050 is another example of transitor mixers.