Finally got around to doing some optical simulation after thinking about what exactly the light is doing. For starters, I think it's easier to start from the eye and cast the rays backwards. That already made me realise that I actually want the image on the retina side of the eyeball lens. If the image is formed there, then you'd see it. Working backwards better allows me to see what rays -- friend or foe -- will hit the retina.

The reason why the image in the test rig looked tiny is because I was trying to grab the rays I'd see at >1m. I'd imagine that the image would look the same size if my eye was [insert system focal length] away too. I might be able to fix it just by bringing the fresnel lens in the decollimator closer.

Anyway, this is the simulation I set up to see if I could actually get a solution with the 30mm beamsplitter, and the short answer is "not really".

With these rays, I can finally understand why I was getting some mystery mirrored image on the bottom of my vision when I angled the right angled prism to reflect stuff 90 degrees off axis. 

• That bright section of rays is what I'd see through the F50 fresnel lens. The slightly darker section is the full 32 degrees I'm aiming for. 
• The length of the lens is the same as the length of the beamsplitter cube (or 30mm right angled prism)
• Therefore, anything that misses the lens would've hit the front wall of the glass.
• Due to the angle of light, these rays would've been totally internally reflected.

I've got the eye relief at 24mm and I can't have relected rays going past that point, so the beamsplitter reflective plane would have to move back. The rays in question would likely need a 40mm lens, thus a 40mm beamsplitter (or 45mm beamsplitter plane). I'm currently unsure as to if there's actually enough space infront of the head to acommodate that within the 300mm cylinder.

I'm not annoyed by the reflections... yet... but 25 degree FOV isn't exactly ideal. Every way I slice it, it feels small.