My memory is I had arrived at the design around Christmas, I was board as I was up with the family away from the shack. John, G4BAO commented my home made F5XG designed patch antenna wasn't circular and it would need two feed-points. DJ7GP has developed a design that apparently is, but I didn't have the dimensions. I thought I would have a go at simulating one. I struggled with this in Octave until Prof. Robert Watson suggested I try the student version of CST and to check the "Axial Ratio". A measure of the circularness.
It took me an evening to learn how to use CST and a few more to learn to use it properly. The student version of CST is limited to 10000 polygons and that turned out to be a challenge. Much time was spent getting the simulation to fit and I was (and still am) doubtful of the fidelity of the results. I am sure the full version with mesh optimisation will be better, but it's not cheap.
The CST simulations implied John was right about the F5XG design so I started again, with an idea based on the old K3TZ patch for AO40, and then put copper water pipe in the middle. The result was awful but after some fettling I could make it well matched with a poor axial ratio or I could make it not very well matched and circularly polarised.
Speaking to Paul M0EYT about this he offered to make a prototype so we could see if it really did work and put to be the doubts about the accuracy of the simulations. This was a heroic effort by Paul as he had no copper sheet in stock. He did have some copper pipe and somehow turned it back into sheet to make the prototype. Paul ended up making lots of prototypes.
Meanwhile, I looked at other solutions, the elliptical patch, the notched patch, the tabbed patch, patch with horn, patch with choke ring. etc. All these patches are approximations of the ellipse. The idea is to have two distinct resonances. You could probably make one rectangular if you fed in in the right place. I understand you can also do the same thing with a tuning capacitor if you get it just right, but I couldn't.
To address the match I experimented in CST with tuning screws, dielectric loading and so on. The best result was with an ellipse and a small coaxial PTFE tuning disc.
Paul made one later on and I am using that now with fair success. The problem is, ellipses are not easy to make, at least not intentionally.
The square patch with the corners cut off and an offset feed was easiest to make with simple tools so that design was settled on. I had assumed most people would be making these in sheds, not with laser cutting and CNC milling Paul made some more prototypes and tested them which verified we were on the right track.
After Christmas, with Remco's help back to basics on the tuning. I was getting better at using the student version of CST's variables which allow you to paramaterise a design and then do some hill climbing optimisations. The correct resonances show up as a loop on a Smith chart, tricky to optimise but the ability to adjust things in 0.1mm steps while doing something else or overnight was extremely useful. Even with the 10000 polygon restriction and no mesh optimisation possible because of that limitation, each run takes several minutes, but far fewer minutes on my i7 desktop than my i5 laptop. I think the feedpoint only moved a couple of mm from the pre-Christmas design.
Paul made another prototype with the revised dimensions and it worked. Remco made a batch for his local club and others reproduced with fair but sometimes variable results, an indication the dimensions are critical.
Remco suggested writing it up. So far the work had carried on on Twitter and IRC. He produced a nice paper with a few tweaks from Paul and myself and it can be found at https://uhf-satcom.com/blog/patch_antenna
Even now, nobody has actually tested one on an antenna range but there are quite a few in use and they seem to work well enough for the simple design. The compromises are:
Squint
I have always been aware of it, but as the patch over illuminates a standard dish it doesn't really matter all that much. It is caused by the asymmetrical fields.
Over Illumination
The over-illumination loss is the main failing of the feed. I simulates additional directors, but the gain wasn't worth the complexity and I quickly ran into simulation size issues.
Poor Match
For the return loss (VSWR). Low enough is good enough. This is not HF where resonance matters, if the return loss is 10dB (VSWR 3:1) and patch is not getting hot it must be radiating. The poor match means 10% of the power is reflected and lost, but almost 90% of the energy is radiated or absorbed in the copper.
Axial Ratio
Polarisation mismatch loss between CP and LP as a function of axial ratio.
The patch axial ratio is not perfect and get worse if the resonances are not right. The loss with the fees properly made should be small, under a dB. A well matched but linear rather than circularly polarised patch loses 50% of the power to cross polarisation.
10 GHz
This is all about 2.4 GHz performance. 10 GHz performance depends on your lens. and how well it matches the dish.
Mike