Due to the frustrations trying to find a suitable receiver/transmitter chip, I decided to spend some time trying to figure out how to do a mixer from (sort of) scratch. I've spent a lot of time learning about mixers.
I have come up with a design using a fully differential amplifier (FDA) and a MOSFET ring mixer. I have simulated it in spice and have been getting pretty good results. I also did a design in kicad and did a basic layout; I believe a full transceiver could be done in less than 1 square inch of board space. I did a 6th order Chebychev filter with LC components which takes up most of the space (you need 5 or 6 filters for a full transceiver), but a 4th order is probably good enough for transmit, certainly better than the switched capacitor filters that most on-chip solutions probably use. And I started with 150 ohm impedance; decreasing that using better MOSFETs is possible and will reduce the inductor sizes.
I like this better because the individual chips are more available and easier to find in the temperature ranges we need. FDAs are plentiful at high speeds; they are commonly used to drive high-speed ADCs. Enough are available at extended temperatures to make me comfortable. MOSFETs are fairly plentiful, and of course the passive components are easy to get. It looks to use less power, too, but that's hard to tell. I would guess this is what most single-chip solutions do, but they don't tell.
I still have questions. I have read in a couple of places that ring mixers perform better with transformer interfaces. We can't do that at baseband, of course. But I can't find anywhere that says how it performs better, or why. Just a few vague statements.
On board layout, I've never done that. Others have done it for me. It brings up its own set of questions. Can you put vias on pads? How wide should the traces be? Lots of little details.
Anyway, that's what I've been playing with.
-corey