Report Details Which Wavelengths Six Key Industries Require For Infrared Filters

Hangzhou, Zhejiang, China – October 28, 2026

Where IR Filters Go: Six Industries and the Wavelengths They Depend On

Every IR system has one component that decides whether the rest of it works. Ask an engineer what limits their infrared system and you will usually hear about the detector, the lens, or the algorithm. Ask them what happens when the filter is specified wrong and the answer changes. The detector still runs, the lens still focuses, the algorithm still produces a number. The number is just wrong, and nothing on the screen tells you so.

An optical filter does one job: it decides which photons reach the detector. That sounds passive. In practice it is the single point where the physics of the application gets converted into an engineering requirement. A gas sensor that needs to see 4.26 micrometres does not need a better detector if its filter passes 4.30. A thermal camera that needs 8 to 14 micrometres does not benefit from a faster processor if the coating leaks at 3 micrometres.

This article works through six industries where infrared filters are load-bearing: gas detection, medical diagnostics, security and surveillance, industrial temperature measurement, optical communications, and consumer electronics. For each one the question is the same set of three: which band, why that band specifically, and what goes wrong when the filter is specified carelessly.

The bands are not arbitrary. They are the intersection of three physical constraints, and understanding that intersection is what separates a filter that works from one that happens to work in the lab.

Why wavelength decides the application

Three things have to line up before an infrared system can function at all: the target has to emit or absorb something in the band, the atmosphere has to let that band through, and the materials available have to be transparent there. Where all three overlap is where instruments get built.

The first constraint is molecular. Most gases of industrial interest are heteronuclear molecules, and their chemical bonds vibrate and rotate at frequencies that correspond to mid-wave and long-wave infrared. Carbon dioxide has a strong asymmetric stretch at 4.26 micrometres. Carbon monoxide absorbs at 4.67. Methane has a fundamental band centred near 3.3 and a weaker overtone structure around 7 to 8. Sulphur hexafluoride, an exceptionally strong absorber, peaks near 10.55.

The exceptions matter as much as the rule. Homonuclear diatomic gases such as nitrogen, oxygen, and hydrogen have no permanent dipole moment, which means their vibrational modes are not infrared active. They cannot be measured with an infrared absorption method at all, no matter how narrow the filter. This is why hydrogen leak detection relies on thermal conductivity, catalytic beads, or electrochemical cells rather than NDIR.

The second constraint is atmospheric transmission. Water vapour and carbon dioxide carve absorption bands out of the spectrum, leaving windows at roughly 1 to 3, 3 to 5, and 8 to 14 micrometres. A system working at 6.3 micrometres, where water absorbs strongly, will lose most of its signal over any outdoor path length. The window structure is not a nuisance to be worked around; it is the reason band selection exists as a discipline.

The third constraint is material transparency, and it is stricter than most people expect. Silicon transmits from about 1.2 to 7 micrometres and then absorbs hard. Germanium covers 2 to 14 but becomes opaque below 2. Zinc selenide spans 0.6 to 21 and is the widest option, at the cost of being soft enough to scratch with a fingernail. Sapphire is tough and broadband in the near infrared but cuts off around 5.5. Fused silica stops at roughly 2.5.

There is a fourth constraint that sits downstream: the detector itself. Indium gallium arsenide responds from 0.9 to 1.7 micrometres. Mercury cadmium telluride can be tuned across a wide range by adjusting composition. Microbolometer arrays peak in the long wave. A filter and a detector that disagree about the band will produce a system that measures nothing, however good each component is on its own.

When people ask why a particular filter costs what it does, the answer usually traces back to how tightly these four constraints had to be squeezed. A 20 nanometre bandpass in the near infrared is a routine part. A 0.22 nanometre passband that has to hold its centre to within 0.05 nanometres across temperature is a different category of object entirely.

1. Gas detection and emissions monitoring

This is the application where filter specification is least forgiving, because the entire measurement is an absorption measurement. Non-dispersive infrared sensors work by comparing the light reaching the detector through a gas-filled path against a reference. If the filter passband is 20 nanometres wide where the absorption line is 60 nanometres wide, a large fraction of the transmitted energy never interacts with the gas, and the signal-to-noise ratio collapses before the electronics have done anything.

The standard bands are well established. CO2 at 4.26 micrometres is the most common, used in indoor air quality monitors, capnography, and refrigerant leak detection. CO at 4.67 is used in combustion and safety monitoring. Methane at 3.3 is used in natural gas detection, though the 7 to 8 micrometre band is also used when uncooled detectors are preferred.

Refrigerants have become their own subfield. Fluorinated refrigerants absorb in the 8 to 9 micrometre region. R32 shows absorption features around 2.2, 2.4, 3.3, and 9.1 micrometres. Propane, increasingly common in small charge systems, absorbs near 3.4 and 6.8. A single leak detector that has to cover several refrigerant types needs either multiple filters or a band wide enough to catch more than one species, and both choices have costs.

Optical gas imaging takes the same physics in a different direction. Instead of a single detector and a beam path, an OGI camera images a scene through a narrowband filter and looks for the difference between background radiance and gas absorption. The critical detail here is that visibility depends on the temperature difference between the gas cloud and the background, not on concentration alone. If the background is warmer than the gas, the plume appears dark. If the gas is warmer, it appears bright. When the two are close in temperature, the plume disappears entirely regardless of how much gas is present.

That last point is where a lot of field confusion comes from. An operator who sees nothing on screen concludes there is no leak. The correct reading is that this particular camera, at this particular angle, against this particular background, cannot resolve the difference. Change the viewing angle so the background becomes sky or a cold structure, and the same leak becomes obvious.

The practical specification points for gas detection filters come down to four. Centre wavelength tolerance, because a shift of a few nanometres moves the passband off the absorption line. Blocking depth, because out-of-band energy raises the baseline and buries the absorption signal. Temperature coefficient of the centre wavelength, because a sensor installed in an outdoor cabinet will see a 60 degree swing between summer and winter. And batch consistency, because a sensor that reads 400 ppm in the lab and 460 ppm in the field has a calibration problem that is really a filter problem.

2. Medical diagnostics and health monitoring

Medical applications use infrared and near-infrared filters in four distinct ways, and they are often confused with each other.

Fever screening is the simplest. A thermal camera operating in the 8 to 14 micrometre window measures skin surface temperature, which correlates imperfectly with core temperature. The filter requirement is a clean longwave pass with strong rejection of shorter wavelengths, because a screening station is usually full of visible lighting, display screens, and windows.

Pulse oximetry is the most widely misunderstood. A pulse oximeter uses two LEDs, typically at 660 and 940 nanometres, and exploits the fact that oxygenated and deoxygenated haemoglobin have opposite absorption behaviour at those two wavelengths. Oxygenated haemoglobin absorbs more infrared and passes more red; deoxygenated haemoglobin does the reverse. The ratio of pulsatile to baseline absorption at each wavelength gives a ratio-of-ratios that maps to oxygen saturation through an empirical calibration curve.

What matters for filter design is that the LEDs already have moderately narrow emission, and the measurement separates the two wavelengths by switching them on and off in sequence rather than by optical filtering. So a filter is not what makes the two-wavelength measurement work. Its role is rejecting ambient light, and that role becomes critical in reflectance-mode and wearable devices, where the detector faces outward into a room full of sunlight and LED lighting, and where the signal is far weaker than in a transmission fingertip clip.

Fluorescence detection is where filters carry the most weight. In a fluorescence assay, the excitation light is orders of magnitude stronger than the emitted fluorescence. A filter that leaks even a fraction of a percent of the excitation wavelength into the emission channel will swamp the signal entirely, because the leakage is far brighter than the thing being measured. Emission filters in these systems are specified at optical densities of six or higher, and the number is not decorative: it sets the detection limit of the entire instrument.

Endoscopy adds a spatial dimension. Narrow band imaging uses specific wavelengths, commonly around 415 and 540 nanometres, because those penetrate tissue to different depths and therefore highlight superficial versus deeper vascular structures. The filter has to switch cleanly and repeatably, since the clinical value comes from comparing the two images.

Across all four, the recurring requirement is not bandwidth but out-of-band rejection, and the second recurring requirement is repeatability. A diagnostic device that drifts between units creates a calibration burden that no algorithm can absorb.

3. Security and surveillance

Surveillance is where infrared filters became a mass-market component, and where the design trade-offs are most visible to the end user.

The daytime problem is colour accuracy. Silicon image sensors respond to near-infrared light up to about 1.1 micrometres, well beyond what the human eye sees. If that light reaches the sensor, foliage turns white and black fabrics turn purple. The solution is an infrared-cut filter, a shortwave-pass coating that transmits visible light and blocks beyond roughly 650 to 700 nanometres. At night, that same filter becomes a liability, because it also blocks the 850 or 940 nanometre illumination the camera depends on. Hence the mechanical filter switcher: the filter moves out of the optical path when the scene goes dark.

The choice between 850 and 940 nanometres is a classic engineering trade. At 850, sensor quantum efficiency is higher and illumination LEDs are cheaper and more efficient, but the light carries a faint red glow visible to the human eye, which matters for covert installations and for residential products where a red ring around the camera looks intrusive. At 940, the glow is essentially invisible, but quantum efficiency drops, LEDs cost more, and the same illumination power produces less signal. The filter has to be matched to whichever band the illuminator uses, with enough out-of-band blocking that ambient light at other wavelengths does not compete.

Shortwave infrared opens a different capability. In the 0.9 to 1.7 micrometre range, silicon-based sensors are blind, but the atmosphere scatters long wavelengths less than short ones, so SWIR imaging penetrates haze, smoke, and dust better than visible light. This is why SWIR cameras are used for long-range surveillance and for inspecting through enclosures. The catch is that 1.4 micrometres sits in a strong water absorption band, so any outdoor path longer than a few tens of metres has to account for it.

For the filter itself, the specification that gets overlooked is environmental durability. A dome camera sits in direct sun, rain, salt fog, and temperature cycling for a decade. A coating that passes incoming inspection and fails after two years of thermal cycling is not a cosmetic problem; it changes the spectral response of a camera that nobody will re-calibrate.

4. Industrial temperature measurement

Thermal imaging and spot pyrometry both depend on a filter that defines the measurement band, and the band choice follows directly from Wien’s displacement law: the peak wavelength of blackbody emission scales inversely with absolute temperature. A surface at 300 kelvin radiates most strongly near 10 micrometres; the same surface at 1000 kelvin peaks near 3.

That is why room-temperature targets are measured in the 8 to 14 micrometre longwave window and hot targets such as furnace interiors, molten metal, and exhaust components are measured in the 3 to 5 micrometre midwave window. Using a longwave camera on a 1200 degree target produces saturation, not measurement. Using a midwave camera on a 30 degree target produces a signal barely above the detector noise floor.

The second decision is bandwidth. A narrowband filter improves the effective emissivity assumption, because emissivity varies with wavelength and a narrow band makes the variation negligible across the passband. It also reduces the influence of reflected background radiation. The cost is signal: a narrower passband admits less energy, which matters most for low-temperature targets where the total radiated power is already small.

Two-colour, or ratio, pyrometry takes a different route to the same problem. It measures at two bands and uses the ratio of the two signals, which cancels the emissivity term if the emissivity is similar at both wavelengths. This dramatically reduces sensitivity to emissivity error, but it introduces a new requirement on the optics: the two filters have to track each other over temperature. If one filter’s passband drifts by two nanometres per degree and the other by half a nanometre, the ratio slowly loses meaning, and the instrument will report a plausible but wrong temperature.

This is a failure mode that shows up late. Nothing breaks, no alarm triggers, and the reading stays inside its advertised range while becoming progressively less true. Systems that monitor a process for years are especially exposed, because there is no moment when anyone re-verifies the reference.

5. Optical communications

Telecom filters live at the opposite end of the precision spectrum from industrial sensors, and the contrast is instructive. Where a gas detection filter might be specified to two nanometres, a dense wavelength division multiplexing filter is specified to five hundredths of a nanometre.

Coarse wavelength division multiplexing uses 20 nanometre channel spacing across 1270 to 1610 nanometres, with passbands of roughly 13 to 15 nanometres. Dense wavelength division multiplexing uses 100 or 200 gigahertz spacing, which corresponds to 0.8 or 1.6 nanometres, following the ITU-T G.694.1 grid across the C-band from roughly 1526 to 1565 nanometres. A 100 gigahertz filter typically has a 0.22 nanometre passband at 0.5 decibel and a centre wavelength accuracy of plus or minus 0.05 nanometres.

The performance numbers are equally demanding. Adjacent channel isolation needs to exceed 30 decibels, non-adjacent 45. Return loss above 45 decibels. Polarisation dependent loss under 0.1 decibel. Wavelength thermal stability on the order of 0.001 to 0.002 nanometres per degree Celsius, which over an 80 degree operating range means the passband moves less than a fifth of its own width.

None of this is achievable without a specific manufacturing approach. Telecom filters are built as multi-cavity thin film stacks, often dozens of layers, deposited with sub-nanometre thickness control and monitored in situ by optical methods rather than by time. They are then qualified to Telcordia GR-1221 and GR-1209, which subject them to damp heat, temperature cycling, mechanical shock, and vibration before they are allowed into a network.

There is one detail worth noting for anyone who works on both telecom and sensing optics. These filters are used at a defined angle of incidence, often 1.8 degrees in a three-port coupler or 13.5 degrees in a compact CWDM module. The angle is not incidental; it is part of the design, because tilting the filter separates the transmitted and reflected beams that make the device work. Design the coating for normal incidence and the passband lands in the wrong place.

6. Consumer electronics and 3D sensing

Consumer devices are where filter volumes reach nine figures per year and where cost, not performance, is the governing constraint. The two main categories are colour correction and depth sensing.

Every smartphone camera module contains an infrared-cut filter directly above the sensor. Without it, images in sunlight look washed out and dark synthetic fabrics take on a magenta cast, because the silicon sensor sees near-infrared that human vision does not. The part is small, cheap, and manufactured in wafer-level processes where hundreds of filters are coated on one substrate and then diced. At that scale, the interesting specification is not peak performance but uniformity across the wafer and consistency between wafers.

Depth sensing uses a different geometry. A time-of-flight or structured-light module projects near-infrared light, commonly at 940 nanometres, and images the returning signal with a sensor behind a narrowband filter. The filter’s job is to pass the projector wavelength and reject everything else, because the returning signal is extremely weak and the scene is usually bathed in sunlight that contains substantial energy at that wavelength even in the near-infrared tail.

The design tension here mirrors the surveillance case. A 940 nanometre projector is chosen partly because it is invisible, which helps in social contexts, and partly because it sits in a water absorption band where the sun’s contribution is lower. But 940 is also where silicon quantum efficiency is falling and where filter coatings have to work harder to achieve the same blocking. Moving to 850 improves signal and lowers cost, at the price of a faint visible glow.

For suppliers, the consumer segment enforces a discipline that industrial segments do not: the part must be identical across millions of units, produced on a schedule measured in weeks, and priced so that it disappears into a bill of materials. A perfect filter that costs a dollar more than the alternative does not win this market.

What all six have in common

Strip away the application detail and four requirements recur in every section above.

The first is centre wavelength accuracy, but the interesting part is the tolerance relative to bandwidth. A two nanometre tolerance on a ten nanometre passband is a twenty percent error budget. The same tolerance on a two hundred nanometre passband is negligible. This is why the same coating house can be perfectly competent for one application and inadequate for another: the requirement is not the number, it is the ratio.

The second is blocking depth and blocking range, specified together. An optical density of four means one part in ten thousand transmission, which is a ninety-nine point nine nine percent rejection. That number is only meaningful when paired with a wavelength range and an angle of incidence. A datasheet that says OD4 without saying from where to where, and measured at what angle, has told you nothing that can be verified.

The third is environmental durability, and it is where incoming inspection and field performance diverge most. Environmental test standards such as GB/T 26331 or ISO 9211-3 define test methods and conditions, but they do not set an acceptable limit for how much the spectral curve may shift. That limit has to be agreed in a technical annex. A supplier who reports a pass on the environmental test and does not report the spectral drift has answered a different question than the one that matters.

The fourth is batch consistency, and it is the hardest to assess from a datasheet because datasheets describe a typical part. What matters in production is the distribution: how wide the centre wavelength scatters across twenty consecutive batches, and how long it takes a coating chamber to return to its baseline after cleaning. These are process statistics, not product specifications, and asking for them is a reliable way to distinguish a manufacturer from an assembler.

How we cover these bands

Multi IR designs and produces infrared filters across the bands described above, from 850 nanometres in the near infrared through 1653 nanometres in the shortwave infrared, and from 3 to 14 micrometres in the mid and long wave.

Centre wavelength tolerance can be held to within plus or minus 2 nanometres on our narrowband products, with out-of-band blocking specified at OD4 or better and average in-band transmission above 91 percent in the infrared. Substrates include germanium, silicon, zinc selenide, sapphire, and chalcogenide glass, selected according to the transmission window and mechanical environment of each application.

The capability we consider most relevant to the six industries above is not any single coating design but the consistency of the process. We hold more than ten thousand infrared component types in stock, and we are the lead drafting unit for China’s industry standard on infrared optical filters. That work involves writing the test methods and acceptance criteria that downstream customers and regulators both rely on, which keeps the specification language close to how the parts are actually measured.

If you are working on a system in one of these bands and are not certain whether your current filter is the limiting factor, the useful first step is usually a measured spectral curve at your actual angle of incidence and operating temperature, compared against the curve you specified. The gap between those two curves explains most field failures we are asked about.

Every application in this article ultimately reduces to the same question: does the filter pass what the physics requires and reject what the environment provides. The industries differ, the wavelengths differ, and the tolerance tightens or loosens by orders of magnitude, but the question does not change.

Frequently asked questions

What is the difference between a bandpass filter and a longpass filter? A bandpass filter transmits a defined range and blocks on both sides, which is what gas detection and fluorescence require. A longpass filter transmits everything above a cut-on wavelength and blocks below it, which is what thermal imaging and infrared-cut applications require when the goal is simply to exclude shorter wavelengths.

Why does centre wavelength shift with temperature? Two reasons act together: the refractive index of the coating materials changes with temperature, and the physical thickness of the layers expands. Both effects alter the optical path length through each layer, which shifts the interference condition and therefore the transmitted wavelength. The magnitude depends on the materials and the design, typically in the range of 0.1 to 0.3 nanometres per degree Celsius for infrared coatings, and it can be reduced by design but not eliminated.

What does OD4 actually mean? Optical density four means a transmission of 10 to the minus four, or 0.01 percent. It is a logarithmic scale, so OD4 is ten times better blocking than OD3 and one hundred times better than OD2. It is meaningless without a stated wavelength range and angle, because blocking varies strongly across the spectrum and with incidence angle.

Can one filter serve multiple gases? Only when their absorption bands overlap closely enough that a single passband covers both without also admitting interfering species. Methane and other light hydrocarbons share the 3.3 micrometre region, which is why a hydrocarbon filter can be broad. Carbon dioxide at 4.26 and carbon monoxide at 4.67 are far enough apart that one filter cannot serve both without becoming so wide that selectivity is lost.

Which substrate should I choose? Start from the transmission window, then from the mechanical environment. Germanium covers 2 to 14 micrometres but is heavy and its refractive index changes strongly with temperature. Silicon is lighter and more thermally stable but opaque beyond 7. Zinc selenide is the broadest option and the softest. Chalcogenide glass can be moulded into aspheres at low cost but lags germanium in refractive index temperature coefficient, long-wave transmission, and large-aperture capability. There is no universally correct answer, only the best fit to a specific set of constraints.

About Us

Founded in 2007, Hangzhou MULTI IR Technology Co., Ltd. is an optoelectronic technology enterprise integrating R&D, production, and sales. Its products are widely applied in aerospace, medical care, AR/VR, display imaging, photography, and other fields, steadily holding the position of the world’s largest spot supplier of optical components.

Media Contact
Company Name: HANGZHOU MULTI IR TECHNOLOGY CO., LTD.
Contact Person: Media Relations
Email: Send Email
Country: China
Website: https://www.miroptech.com/

 

Press Release Distributed by ABNewswire.com

To view the original version on ABNewswire visit: Report Details Which Wavelengths Six Key Industries Require For Infrared Filters