Polygon scanner
Traces reflection from every facet of a rotating regular polygon.
Open in the canvas →In the real world
A rotating polygon scanner is a prism of flat mirror facets cut around a wheel, spun continuously by a motor. Each facet sweeps the beam through one line; as it passes out of the beam the next facet picks it up at the start of the next line. The idea is old enough to be everywhere without being noticed — it is the mechanism inside laser printers, supermarket barcode scanners, many LiDAR heads, and the line-scanning laser processing systems used for high-throughput marking and ablation.
Its advantage over a galvo mirror is that the motion never reverses. A galvo has to decelerate, stop and accelerate back at the end of every line, and the settling that follows is what limits how fast it can scan. A polygon turns one way at constant speed, so there is no turnaround to wait for and the line rate is set purely by how fast the motor spins and how many facets it carries:
What you pay for that speed is pupil walk. A galvo pivots about its own face, so the beam leaves from roughly the same place and only the angle changes. A polygon facet is offset from the rotation axis, so as the wheel turns the reflection point slides bodily along the facet and the beam translates as well as tilting. Scan lenses for polygon systems are designed around that moving pupil, and facets are made generously larger than the beam so it has room to walk.
The other cost is the gap between facets. For part of every rotation the beam straddles the edge between two facets and is split in two, each half leaving at a completely different angle. Nothing useful can be done with that light, so the source is gated off across the transition — the scanner's duty cycle is the fraction of each facet period that survives. A wider beam eats more of the facet and leaves less duty, which is the trade behind the large wheels in high-power line-scanning heads.
Because every facet is cut and mounted separately, real wheels also carry facet-to-facet angular errors. A facet tilted a fraction of a milliradian out of plane puts its line slightly above or below the others, and since the error repeats once per revolution it shows up as periodic banding in the scanned image — the reason precision systems either specify pyramidal error tightly or correct it actively.
In OpticalSetup
The component is a regular polygon centred on its rotation axis, and the vertices that draw it are the same vertices that get traced: every facet you can see is a real mirror surface, so there is no separate abstract scan angle that could disagree with the picture. Each facet reflects by the ordinary vector law of reflection used by the plain mirror, which means the 2× angle doubling and the pupil walk are not written into the model — they simply come out of turning the geometry.
Rotation runs the wheel continuously at a set RPM, or holds a static phase so you can step through a facet by hand. The facet rate readout gives the physical lines per second at all times, even when playback is slowing the visible motion down for inspection.
The usable scan window is an ideal synchronized blanker: a centred fraction of each facet period during which the facets reflect, with the hub drawn green. Outside it the facets absorb, the hub turns amber, and no outgoing ray remains — the modelled equivalent of gating the source across a facet transition.
The wheel is opaque, so a facet reflectivity below 100% loses the remainder to the coating rather than transmitting it. That is deliberate: a solid metal wheel has no way to pass light, and letting it through would produce spurious reflections off the inside faces of the far facets.
The scan window is not derived from your beam. It is a fraction of the facet period centred on the facet, and the component has no knowledge of what is illuminating it, so a window left wider than the geometry supports will show the beam splitting across two facets while the hub still reads open. That split is real behaviour — it is what the blanking exists to hide — but choosing the window to suit the beam is left to you. Oblique incidence tightens it further and asymmetrically: the footprint on the facet is the beam width divided by the cosine of the incidence angle, and that angle grows on one side of the sweep and shrinks on the other, so the clean window is both narrower than the facet ratio suggests and not centred on the facet.
Blanking is an ideal switch synchronized to the facet, not a model of how any particular controller drives a source. Every facet is perfect and identical: there is no pyramidal or facet-to-facet angular error, so none of the periodic line banding that characterizes real wheels appears, and no bearing wobble, windage, or timing jitter. There is no f-theta or telecentric scan lens — put an ordinary lens after the wheel and the focus moves as f·tan θ, with the pincushion that implies. Nothing here predicts a diffraction-limited spot size, and the second scan axis that turns lines into an area is out of the plane and not modelled.