I think I've figured out a good scheme, which may be the right way to build the C-band antenna. It's a bit different than you guys have been thinking.
You've been thinking about how to accomplish the phasing of the RF signal, but ignoring the physical distribution of RF. Physical distribution of C-band RF signal to 35 or so elements is nontrivial. It could be done with splitters and cables and connectors, or printed splitters and microstrip lines, but with any of these schemes there will be amplitude mismatches and various phase shifts associated with just the distribution. If we distribute RF, we'll have to compensate for those amplitude and phase variations. Can be done of course.
How about this. Instead of distributing RF, we distribute a much lower reference frequency which is multiplied up by a PLL at each antenna element. A PLL is one chip these days, so it certainly doesn't cost much in dollars weight or power. We've already got a PA and phase control stuff at each element after all. So maybe we distribute 100 MHz or something easy. It is only gonna be used as a reference, so we don't need to match amplitudes. We design the reference distribution network for high isolation between the outputs, so the individual element circuits have least chance to couple. We don't need to match phases of the distributed reference because we're gonna adjust the phase of each element under software control anyway. (We'll build a calibration table which the software will use as a term it adds into the total phase shift it specifies for each antenna element.)
At each element we have a PLL to generate the RF frequency, followed by a digitally controlled analog phase shifter chip, and a digitally controlled attenuator chip (to adjust out variations in the phase shifter chip vs control input), and a balanced mixer to generate the 180 degree phase shift of BPSK. (If the phase shift chip has a 180 degree input, we can leave out the balanced mixer.) Next, of course, the signal goes to the local PA.
This entire circuit fits on a small board smaller than the antenna element (ie patch) itself, and bolts to the back of the antenna element.
This design is highly modular. The only things distributed are power, reference, data-to-be-modulated, and data to control the individual elements. The controls are on/off, phase, and gain.
It is also possible to move the digital phase shifter to BEFORE the PLL, at which point it may be possible to remove the digital attenuator entirely. You would no longer need it to compensate for the changing attenuation of the phase shifter vs control input, as long as the PLL could handle the range of possible amplitudes. (On the other hand you might still want the attenuator as an easy way to balance the gain and/or power output level of the PAs. With the digitally controlled phase shifter operating at a lower frequency it would likely be more accurate and have less amplitude variation anyway.
This scheme works no matter what the geometry of the array. Software just generates the phases, and sends 'em down some bus.
The thing I have described easily makes unfiltered BPSK or even QPSK. Somebody was talking about possible filtering. That's harder. I hope we don't need filtering. Distributed filtering is not an easy thing.
What do you think?