I was in a bit of a hurry to get out this morning when I wrote my original post above, I've since spent some time looking into this in a little more detail.
The power splitting is achieved with a cascade of 35 ohm 0.25 w/l transformers taking 50 ohms down to 25 ohms, and then paralleling up the next two 50 ohms.
At each antenna, there are two feedpoints phased 90 degrees apart for circular polarisation each with an approximate impedance of ~25 ohms. These are transformed up to about 100 ohms with a 0.25 w/l transformer of about ~50 ohms. The resulting two 100 ohm feedlines are paralleled up to make 50 ohms and feed back into the remaining cascaded power splitters.
The additional 0.25 w/l delay in two of the 50 ohm sections is because two of the patches are being fed at different corners to the other two, an phase needs to be maintained on the wavefronts emitted from each antenna or you'll definitely not get the desired directionality and gain (think about the constructive interference pattern we're after to achive gain).
On reflection, I had missed in my haste the four parasitic copper director squares in front of the patches separated by the foam which provide more gain and directivity. The foam is there to maintain mechanical distance for the desired beam pattern. A side effect of the parasitic directors is that there will be a significant lowering of the feed impedance of the directly driven patches.
Seeing there are technically two sets of antennas would you expect the tuning is slightly different between them? I can't tell from the images but I would think one set is for TX and one set is for RX being they are 100 MHz apart with Inmarsat. If Dave gets bored and gives us precise measurements of the various transmission lines etc we would know for certain.
Not on this one board, this is a simple single L band antenna CP parch array with nothing more than power splitters, delay lines and matching sections, there is no diplexer. Now there may be another antenna on the device, or the operation is half duplex or TDM like GSM. There is also only one feedpoint if you look, if it is full duplex then the diplexer is not on this board.
FWIW, back in the early noughties I used to design antennas for L and S band for space applications, including patch arrays on PCB like this one, the techniques here are exactly what I used then.