It starts with how two eyes judge depth. Then it follows a top view of the stack, from the backlight at the top down to a viewer at the bottom. Every ray is traced through its lens, so the strips each eye receives are computed, not drawn by hand.
A printed lenticular layer behind the LCD gives each eye a slightly different background, so the depth is real binocular 3D. It is baked into the plate at the factory, not rendered per viewer.
Order from the front: LCD panel, 3D plate, backlight. A touch overlay on the front glass cannot steer the plate's rays.
The box is an interlaced print under the lenses. No eye tracking, so everyone inside the zone sees it at the same time.
Outside it the views wrap. The cone in this drawing is wider so it fits on screen.
Stand too close (touch distance) and the eyes stop sampling the strips. You see the lens lines instead.
Download the video (MP4, 1080p, 44 s, 35 MB) for presentations.
The drawing is a simplified top view, not to scale. The number of lenses and views, the lens shape and the cone width are illustrative. The real 3D cone is 24–29°.