First try at printing a gradient-index lens based on a variable wall-thickness gyroid lattice with the lattice surface meshes calculated using MATLAB code, then stitched and repaired, imported into Fusion 360, smoothed and then exported as STL into Prusaslicer and printed in PLA on the MK3S+ in the Prusa Enclosure.
Now I need to do another version that uses a hyperbolic secant parameter varying inversely with the radius from the cylinder axis, to form a cylindrical analogue of a biconvex focussing lens for microwave frequencies around 10-24 GHz, then print that using Nanoe's Zetamix Epsilon RF dielectric filament. PLA first though, I need to be sure it's 100% correct before I commit to using all that expensive RF filament.
@g4dbn
"... a hyperbolic secant parameter varying inversely with the radius from the cylinder axis, to form a cylindrical analogue of a biconvex focussing lens..."
I'll try to say it as fast as I can to improve my spoken English skills !
More seriously, it looks impressive.
An hyperfrequency lens, if I understand correctly ?
Had a look at the Zetamix filament. I wonder what's the cost of such a highly technical product ! (Stratospheric, I guess !)
@g4dbn Is there a potential effect of the surface condition of the final printing on losses or efficiency ?
Fuse filament printing is not very smooth.
Just curious...
@F4JWJ Any features smaller than about 0.1 wavelength have very small effects on the bulk properties of the lenses. Most of the rough surface features are less then 0.5 mm, so the surface is probably OK to at least 30 GHz and may be OK at 47 GHz. The unit cell size of this print it too large to be used as a practical lens, it is best to keep the unit cell to about 0.1 wavelength, so 3mm at 10 GHz or 1.2 mm at 24 GHz.
@F4JWJ The surface finish on conductive materials like waveguide is much more important to prevent scattering and excitation of higher modes that can cause attenuation and phase cancellation, but even then, surface features less than 0.05 wavelength are unlikely to have significant effects unless the length of the guide is hundreds of wavelengths and the roughness is spread out over the whole run of guide.
@g4dbn Thanks a lot.
Seems you are quite knowledgeable on the subject !
@F4JWJ Did you see my Youtube videos about the UV resin printed lenses? https://youtu.be/3YMRfw0uWlw and https://youtu.be/_4vv3N5UpW0
@g4dbn Very interesting, indeed !
Nice precision machining, in other videos I checked randomly.
Soviet spies were quite creative...
@g4dbn BTW, I didn't see any test or real life measurements of the lenses (nor of the spy bug or the mechanical coaxial LPF)
Did I miss something ?
@F4JWJ No, I got side-tracked on to other projects, but I'll be doing those over the next few months
@g4dbn I'm pretty sure Aimée won't let you forget to do these videos ! 
@F4JWJ List price of the Zetamix Epsilon range is between 275€ and 350€ per kg. That example lens would need around 80g of the ε4.5 material, so around 24€ plus tax and carriage. The material is supplied at 500g reels.
@gorplop The unit cell size determines the frequency range where it will work. It's best to use about 0.2 wavelengths, but a little more is OK. You also have to take into account the relative permittivity of the materials, so it ends up around maybe 1/10 of a wavelength or even smaller. At 30 GHz, that means a unit cell of around 1 mm for best results, which is getting close to the limits of precision of the resin process. There are some exciting ceramic sintering processes being developed.
@gorplop To get close to 1 THz, you start hitting material resonances. Even PTFE has a nasty suck-out at about 1.5 THz, then goes terrible above 8 THz. I've never done anything above 0.28 THz, but if I can get hold of devices, I'll have a go around 410 GHz. That's pretty serious stuff, and making measurements is pretty much impossible. When you get to 280 GHz, the waveguide sizes are sub-millimetre, and getting a fine polished finish is a huge challenge. 47 GHz is the end of my comfort zone
@jmorris I'm having a few nightmares with material distortion at the moment, but I'm thinking of printing a large Luneburg lens in segments so I can test it more thoroughly. I'm trying to do 100 things at once though, and just wasted four hours trying to fix my old RasPi 3 running dump1090 for an ADSB feed to an aggregator to use in aircraft scatter communications. Turns out the image was SO ancient (wheezy), it was nearly impossible to upgrade. However, it is now shiny and lovely again. Phew!
@g4dbn wow that's impressive! Back then in university our teachers were not confident about such things and in 2023 it's 3D printable.
BTW, Maybe calculations can be done internally in OpenSCAD? Something like a factory of lenses, based on input parameters and it just generates an .stl, voila. I've also seen gyroid as an infill pattern option in Creality slicer, can it be related?