Why prism multi-sensor cameras need prism-optimized lenses

A prism camera earns its keep by splitting one optical path into blue, green, red and often near-infrared channels that stay pixel-registered frame after frame — no Bayer interpolation, no false color on edges, no motion-dependent mismatch. The catch sits behind the lens mount: a thick glass block coated with dichroic filters that only behave when rays arrive close to normal. Fit an ordinary single-sensor lens and the corners of each channel shade, one band drifts soft, and the channels stop registering. This guide gives you the geometry that decides it, the two chromatic aberrations that matter most, and the datasheet numbers to demand before you order.

What does the prism change between the lens and the sensors?

The prism inserts a thick glass block with dichroic filter coatings on its internal surfaces between the rear element and the sensors, so every ray your lens delivers crosses several coated surfaces whose transmission depends on the angle at which it arrives.

A single-sensor lens never has to care what happens behind its rear element; a prism camera makes that space part of the optics. Dichroic coatings reflect one waveband and pass the rest, so the block folds the bands onto separate sensors that stay aligned regardless of motion or viewing angle — the advantage over a Bayer sensor covered in prism 3-sensor color cameras versus Bayer. The same advantage is the constraint: your lens now feeds a wavelength splitter, and everything about its output fan — ray angle, cone spread, chromatic content — lands on the splitting surfaces. Prism models sit in the area scan cameras category; the block behind the mount is what separates them.

What causes the color shading a standard lens leaves behind?

Dichroic filters separate wavelengths by optical interference, and an interference coating's transmission edge moves as the incidence angle grows, so a lens that fans rays across a wide angle range gives each channel a field-dependent transmission you see as color gradients.

The edge is steep by design — that steepness is what gives a prism camera its clean color separation — so a small angle change flips transmission noticeably. The corners of the field carry the steepest ray angles, which is why shading concentrates there and why each channel shades differently: the blue-reflecting surface, the red-reflecting surface and any NIR surface each see the same fan of angles against a different edge. On a uniform gray target this reads as a green or magenta wash toward the corners. White balance will not remove it, because the error is per channel and per image position, not a global gain.

How far should the exit pupil sit from the prism?

Work the geometry backward from a shift budget: keeping the cutoff shift near 0.3 nm at 500 nm means holding the chief ray within about 5° in air, which for a 2/3-inch sensor (5.5 mm half-diagonal) puts the exit pupil about 63 mm from the sensor.

The exit pupil is the image of the aperture stop formed behind the lens, and every chief ray leaves it headed straight for its field point; the exit pupil article works through the geometry. Its position is paraxial and often virtual — it can sit well behind the last element, with no mechanical reference on the barrel — so treat it as a datasheet number, not something you can see. Prism-optimized designs push that virtual pupil far from the prism; ordinary designs leave it short because single-sensor imaging gives it no reason to matter.

Here h is the sensor half-diagonal, d_pupil the exit-pupil-to-sensor distance, n_prism about 1.5 for the prism glass, and n_eff 1.7-2.0 the effective index of a visible dichroic stack. The last line is the standard interference-filter estimate, and the shift it predicts grows with the square of the sine of the angle, which is why the corners suffer first.

Worked example, 2/3-inch camera: h = 5.5 mm, n_prism = 1.5, n_eff = 1.8, cutoff edge at 500 nm.

• Short-pupil standard lens, d_pupil = 15 mm: θ_corner = atan(5.5/15) = 20.1°. In glass, arcsin(0.344/1.5) = 13.2°. Then Δλ/λ = 1 − sqrt(1 − 0.0161) = 0.81%, so the edge moves 500 nm × 0.0081 = 4 nm.
• Prism-optimized lens, d_pupil = 60 mm: θ_corner = atan(5.5/60) = 5.2°. In glass, arcsin(0.091/1.5) = 3.5°. Then Δλ/λ = 0.056%, so the same edge moves about 0.3 nm.

The ratio is (sin 20.1°/sin 5.2°)² ≈ 14: the long pupil leaves a corner error a flat-field calibration can absorb, where the short pupil's error reshapes the channel crossover itself. Turn the budget into a distance with d_pupil ≥ h / tan(5°): 5.5/0.0875 = 63 mm for 2/3-inch, 8.0/0.0875 = 91 mm for 1-inch sensors. Ask for the printed exit pupil distance and compare it against those classes.

Does stopping down or a telecentric design fix the angle problem?

Stopping down only narrows each ray cone — to roughly ±atan(1/(2N)), about ±3.6° at f/8 — while the cone's center line stays at the corner's chief-ray angle, so it softens the shading without removing it; exit pupil distance is what removes it.

At f/2.8 the cone alone spreads ±10.1° around the chief ray; at f/8 it is ±3.6°, but in the worked example the center still sits at 20°. Stopping down also collides with diffraction: the Airy diameter is 2.44λN, so 2.44 × 0.55 µm × 8 = 10.7 µm, about three pixels across a 3.45 µm pitch. Slower apertures do tame classical aberrations, as covered in optical aberrations and stopping down, but the prism problem is geometric and only the exit pupil solves geometry.

The design that does is image-side telecentricity: a lens telecentric in image space forms its exit pupil at infinity, so every chief ray runs parallel to the axis and strikes the prism near normal across the whole field. That is precisely the geometry a prism-optimized lens builds in on purpose; the costs are size, working distance and price, weighed in entocentric versus telecentric lenses.

Why do the chromatic aberrations hit a three-sensor camera harder?

Because the prism delivers each band to its own sensor, longitudinal chromatic aberration now decides whether all channels can ever be sharp at once, and lateral chromatic aberration decides whether the factory's sub-pixel registration survives your lens.

Longitudinal aberration is the axial spread: glass bends blue harder than red, so the planes of best focus separate along the axis. A Bayer sensor averages that spread into one image; a prism camera cannot, because each band lands on its own sensor. The lens must be achromatized across every band the camera splits — through the NIR where a channel sits near 850 nm — or one channel is always the soft one.

Lateral aberration is the registration killer: each wavelength images at slightly different magnification, so the bands drift apart toward the edges. Put a number on it: 0.1% lateral CA displaces one band 5.5 µm against another at the corner of a 2/3-inch sensor, about 1.6 pixels on a 3.45 µm pitch — enough to erase an alignment done to a fraction of a pixel. Current 12-25 MP machine-vision sensors run 1.6-3.5 µm pitches, so the tolerance only tightens as resolution climbs. Applications that live on band-to-band agreement — measurement, precision color, RGB and NIR fusion — need that figure in writing, by the same discipline that pushes gauging toward telecentric lenses.

Which lens specifications should you check before ordering?

Demand five numbers from the lens datasheet, and treat any one of them as missing if it is not printed: image circle against your sensor diagonal, exit pupil distance, the wavelength range of the chromatic correction, lateral chromatic aberration, and the spectral range the transmission is quoted over.

• Image circle at or above the sensor diagonal: 11 mm for 2/3-inch, 16 mm for 1-inch; the sensor format explainer shows how format, circle and mount pair up.
• Exit pupil distance at or above the class from the worked example: about 63 mm for 2/3-inch, 91 mm for 1-inch.
• Chromatic correction stated across every band the camera splits, visible through NIR where an NIR channel exists; a visible-only achromat leaves that channel soft.
• Lateral chromatic aberration specified in percent or micrometers at the field edge, not left blank.
• Resolution class matched to the pixel pitch, because a 2-3 µm pitch punishes a lens resolved only for 5 µm pixels.

Focal length and coverage come first, as in how to choose a machine vision lens focal length; these prism constraints go on top. Filter the machine vision lenses by mount, image circle and resolution, run the lens finder, or start the lens configurator at the lens step.

When can a standard lens still work on a prism camera?

A standard lens stays usable when the application only inspects geometry in a central region, tolerates a few percent of channel shading after calibration, and never measures color — roughly, when you could tolerate a Bayer image but want the registration.

Run the central-region numbers on the short-pupil lens from the worked example: a ROI spanning half the 2/3-inch diagonal puts its edge rays at atan(2.75/15) = 10.4°, about 1 nm of cutoff shift at 500 nm — visible on a color target, small enough to calibrate out for geometry.

Many prism cameras provide a per-channel flat-field or shading calibration; run it with the actual lens fitted, at the working aperture, and rerun it whenever the lens or aperture changes. What calibration cannot restore is sharpness in a band the lens was never corrected for, or registration once lateral aberration has displaced the bands. So the honest split: standard lens for presence, orientation and defect-geometry checks that read one band; prism-specified lens for color measurement, band-to-band measurement, or any RGB+NIR camera.

How do you verify the pairing before it goes on the line?

Three bench tests separate a matched pair from a compromise in under an hour: a uniform white shot per channel for shading, a slanted high-contrast edge per channel for focus and registration, and both repeated at the production working distance and aperture.

For shading, shoot the uniform field and read each channel's gradient center to corner; a workable acceptance line is a few percent edge-to-center per channel after focus is set at the working aperture. For registration, measure the channel-to-channel offset on the slanted edge and hold it to about 0.3 pixels, the range in which factory sub-pixel alignment stays meaningful, and check that each channel's best focus falls inside the shared focal depth. Repeat after any lens or aperture change — the shading calibration and these tests move together.

Which lens do you actually order?

Order the lens family your prism camera's maker pairs with the camera, confirm its exit pupil distance and chromatic range against the numbers above — 63 mm class for 2/3-inch, 91 mm for 1-inch, correction spanning every split band — then buy through the lens category with mount and format filters.

If a datasheet will not state exit pupil distance or full-band chromatic correction, that silence is the answer: a lens designed for prism work carries both numbers because they are its reason to exist. Send the camera model, sensor format, working distance and bands through the lens configurator or to expert support, and have the pairing checked against the prism before you order.