Choosing an infrared window material for a thermal sensor enclosure

Your thermopile or pyroelectric temperature sensor is filtered to see longwave infrared, and the protective housing needs a window in front of the detector. Choose that window as if visible clarity were the goal and you will ship a sensor that reads its own enclosure instead of the target. This guide gives you the passband logic, the materials that actually transmit above 5.5 µm, the Fresnel arithmetic that decides whether a coating is optional or mandatory, and a qualification procedure for the finished housing. The worked numbers assume the common longwave-pass cut-on at 5.5 µm and a detector window spanning 5–14 µm; if your part filters differently, run the same steps with your band. By the end you can write the window line on the drawing: material, coating, thickness, tilt and the transmission you will accept.

Match the window to the passband, not to your eye

A non-contact temperature thermopile sees only the band its filter lets through. The common design is a longwave-pass filter with the cut-on at 5.5 µm, and the detector's can window is coated for roughly 5–14 µm. The target's emission sits inside that band. Wien's displacement law puts the emission peak at:

At a 25 °C ambient (298 K) the peak is 9.7 µm, a 100 °C target peaks at 7.8 µm, a 150 °C target at 6.9 µm. Nearly all of the signal lives between 6 and 14 µm, where ordinary glass is completely opaque. Two rules follow. First, the window must transmit the full 5.5–14 µm band close to uniformly: a material that fades across the band re-weights the passband and shows up like an emissivity error. Second, visible clarity tells you nothing. The correct window can be fully opaque to your eye, and a crystal-clear one can be optically black in the band.

The four materials that pass 5.5 to 14 µm

Four material families pass a longwave-pass band properly, the same families used in LWIR thermal lenses; thin HDPE earns a compromise row. The transmission column is the Fresnel estimate for uncoated surfaces, and coatings change it, which is the next section's subject.

Chalcogenide glasses transmit the whole band and some compositions reach 25 µm, which is why moulded chalcogenide has become the default for volume-built housings. Everything else on the shelf fails the band: soda-lime glass, acrylic and polycarbonate stop transmitting between about 2.5 and 3 µm, fused silica reaches roughly 3.5 µm, sapphire fades near 5 µm. Each falls short of the 5.5 µm cut-on, so the sensor receives no target energy at all and settles on the only thing it can see, the window's inner face at housing temperature.

Fresnel loss makes the coating the main decision

At normal incidence, each air-to-material surface reflects a fraction set by the refractive index:

Germanium's index at 10 µm is about 4.0, so each bare face reflects (3/5)^2 = 0.36 and the window passes (1 − 0.36)^2 = 0.41 of the band. Silicon loses about 30% per face and passes 0.49; chalcogenide about 21% per face and 0.60–0.64; zinc selenide 17% per face and 0.69. A broadband anti-reflection coating on both faces lifts two-face transmission above 0.9, with about 0.95 typical across the passband. The reflected light does not vanish either: each uncoated face throws a third or more back toward the detector, and a flat window sitting parallel to the can returns it as a ghost. Tilt or wedge the window so reflections land off the chip.

Worked example: 1 mm germanium window on a 100 °C target

Inputs. The detector filter is longwave-pass with a 5.5 µm cut-on and its can window is coated across the longwave band (detector data). The housing window is 1 mm germanium with two flat faces; germanium's index at 10 µm is about 4.0 (material data). The sensor body sits at 25 °C (298 K); the check targets are 50 °C (323 K), 100 °C (373 K) and 150 °C (423 K). The signal model: a thermopile's output scales with the band radiance contrast between target and body, and carrying that contrast through the fourth-power law reproduces the sign and scale of the bias — an estimate, with the sensor's own calibration curve deciding the exact reading.

Rule in words. The window multiplies the target-minus-body contrast by its transmission τ, and the sensor reports the temperature whose contrast equals the transmitted one.

Arithmetic. Per face, R = (3/5)^2 = 0.36, so bare τ = (1 − 0.36)^2 = 0.41. Contrast at 100 °C: 373^4 − 298^4 = 1.94 × 10^10 − 0.79 × 10^10 = 1.15 × 10^10 (relative units). Bare window: transmitted contrast = 0.41 × 1.15 × 10^10 = 4.7 × 10^9; the reading solves (298 K)^4 + 4.7 × 10^9 = 1.26 × 10^10, whose fourth root is 335 K, about 62 °C — a 38 °C cold bias. Coated to τ = 0.95: transmitted contrast = 1.09 × 10^10; reading = (0.79 × 10^10 + 1.09 × 10^10)^0.25 = (1.88 × 10^10)^0.25 = 370 K, about 97 °C — a 3 °C bias.

Range check at τ = 0.95. The 50 °C target reads 49 °C and the 150 °C target reads 146 °C, so the residual stays between 1 and 4 °C across a 50–150 °C span. The bare window is also nonlinear — its 50 °C target reads only 36 °C — so no single offset trim repairs it.

Decision. Reject an uncoated germanium window: τ = 0.41 costs 38 °C at a 100 °C target and drifts worse through the range. Accept a two-face AR-coated window and write the threshold on the drawing: two-face transmission ≥ 0.90 across the band, targeting 0.95, then trim once through the finished housing.

Fit the window, then qualify the assembly

Three mechanical details protect the reading. Tilt the window 3–5°, or order it wedged, so surface reflections walk off the detector; the same move the protective window guide for vision lenses uses works unchanged here. Keep the window a few millimetres clear of the can and seal it with a compliant gasket — germanium is stiff but brittle, and a hard-clamped plate cracks at the flange. Keep conducted heat away from germanium, whose transmission falls as the window warms; that is why the table steers warm or moulded housings to chalcogenide. Temperature also moves infrared optics themselves: once you graduate from a bare thermopile to a thermal camera, the infrared lens athermalization guide works through the focus drift.

Then qualify through the housing, not the bare part. Aim at a matte-painted plate with a bonded thermocouple, read bare and housed at two or three temperatures across your range, and log the offsets as the housing's transmission signature. Keep the window clean in service, because a film of dust or oil is itself an uncoated optical layer; the inspection-cell viewing-glass cleaning guide covers the safe method.

Why a motion-detector Fresnel lens is the wrong optic

The faceted Fresnel sheets sold for occupancy detection are thin polyethylene and solve a different problem: they segment a very wide field into alternating zones so a pyroelectric element triggers when anything moves anywhere in the room. Temperature measurement wants the opposite geometry, a defined cone aimed at the target and a transmission you can state. The faceted sheet delivers a field tens of degrees wide with no defined spot, adds scattering and uncontrolled absorption, and gives you nothing to qualify against. For measurement at more than a few tens of centimetres, use a real LWIR optic or a thermal camera, which is where this goes next.

When the band is near-IR, or the sensor needs a real optic

Not every IR sensor is longwave. A photodiode filtered for 900–1700 nm works straight through ordinary glass, which only stops near 2.5–3 µm, so the window becomes a mechanical choice and the optics matter only if you shape the beam. The SWIR vs NIR vs visible guide sorts out which band your detector actually covers, and the SWIR wavelength selection guide shows how the band follows the material. Threaded band-pass filters for near-IR ambient rejection, such as the 850 nm parts under lens accessories, mount on the lens front. For imaging at those wavelengths the lens must be SWIR-corrected; these two bracket the choice.

The LS-SWL-0229 covers 900–1700 nm at 35 mm and F1.4 on a 1" format; the LS-SWL-0246 spans 400–1700 nm at 25 mm and F1.8 on 1/2", so one window and lens can also carry visible alignment light.

All 2 share: Mount C-mount; Focus Manual; Operating temperature -10 to 50 °C; Storage temperature -20 to 60 °C. Values as printed in each product's data sheet; open a part for its drawing and full specifications.

If the task is a defined spot at distance, move from the bare thermopile to a thermal camera behind one of the LWIR thermal lenses. A 17 mm, F1.2 LWIR thermal lens such as the LS-SWL-0006 covers a 1/1.2" sensor with an 11–16 mm image circle and turns the reading into an image you can aim; the thermal camera working distance and spot size guide does the spot arithmetic. Coating and passband still decide contrast, for cameras as for thermopiles — the lens coating and transmission guide shows why a mismatched coating reads as lost signal.

Checklist before you order the window

• Passband on the drawing: two-face transmission ≥ 0.90 across the full band, targeting 0.95. A single peak value near 10.6 µm hides band-edge falloff, where coatings usually fade first.
• Material matched to temperature: germanium for near-ambient housings, chalcogenide for warm or moulded ones, silicon only if the passband stops by 8 µm, zinc selenide when visible shares the aperture.
• Thickness 1–2 mm for a small housing: brittleness and gasket load set it, not optics.
• Tilt or wedge 3–5° so ghost reflections walk off the detector.
• Sealed assembly: gasket hardness, compression and the cable entry belong on the same drawing as the window.
• Qualification plan: bare-versus-housed offsets at two or three known temperatures, recorded against the housing serial.

If the housing also carries a camera-grade LWIR or SWIR optic, shortlist it with the machine vision lens finder or start the lens configurator at the lens step. The lens guides hub collects the rest of the material and coating background, and our contact page is open when you want the window numbers checked before you order.