Full width at half maximum (FWHM) is the width of an LED emission band measured at the two points where intensity drops to half the peak. In SWIR food sorting, FWHM decides how precisely the illuminator probes the absorption features that separate good product from defects. Pick the wrong bandwidth and even a correct center wavelength fails to sort.
The Absorption Bands That Do the Sorting
Sorting physics is spectral. Water absorbs strongly near 970, 1200, 1450 and 1900 nm, so moisture gradients in grains, nuts and baked goods appear as reflectance dips at these bands. Fat and oil carry signatures around 1200 nm and 1725-1760 nm; protein and starch absorb at other near-infrared positions. A SWIR LED centered on one of these features illuminates the product, and an InGaAs camera images the reflected band.
If the LED band is too wide, emission spills across both the absorption dip and the adjacent reflectance plateau. Contrast between moist and dry material drops, the signal-to-noise ratio falls, target and background spectral signatures blur together, and classification thresholds become unstable. If the band is too narrow, it probes only the center of broad molecular features and adds little discriminating power.
The detector matters as much as the source. InGaAs arrays carry their best sensitivity through roughly 900-1700 nm, so LED bands at 1200 and 1450 nm sit exactly where the camera responds strongest; a well-chosen FWHM concentrates flux inside that responsive window instead of wasting it at the edges.
How Wide Is Right
Typical high-quality SWIR LEDs achieve FWHM of 40-90 nm. For moisture-sensitive sorting at 1200 or 1450 nm, 50-80 nm is the balanced zone: wide enough to cover the molecular feature, narrow enough to keep the surrounding plateau clean. Below 30 nm the contrast gain is small and flux falls, because spectral purity trades against efficiency in InGaAsP materials.
| Sorting task | Center wavelength | Target feature | Suggested FWHM | Note |
|---|---|---|---|---|
| Moisture in grains | 1450 nm | water peak | 50-80 nm | strongest dip, best SNR |
| Moisture, wider field | 1200 nm | water peak | 50-80 nm | weaker dip than 1450 nm |
| Oil and fat detection | 1725-1760 nm | fat/oil band | 50-80 nm | verify camera response |
| Multispectral heads | 1050/1200/1450/1650 nm | multiple bands | ≤50 nm | channels must not overlap |
Matching the Camera Filter
Many sorters place narrow band-pass filters in front of the InGaAs camera to reject ambient light. The passband must transmit the full LED band at operating temperature, not just at room conditions. SWIR LED peak wavelength shifts with junction temperature, and FWHM widens slightly as Tj rises, so leave margin at both passband edges. A filter cut too tight transmits only the band center and starves the sensor; a cut too loose admits ambient light from hall lighting or neighboring halogen lamps on the line. In multispectral heads running several LED bands, FWHM must stay narrow enough that adjacent channels do not overlap; once bands merge, the classifier cannot separate the channels and the extra optics add nothing.
What to Specify With the Supplier
Request the spectral plot at rated current and at the real operating junction temperature, not only the 25°C datasheet curve, and verify FWHM at both conditions. Confirm bin-to-bin consistency of peak wavelength and FWHM across shipments, and ask for drift data over life: FWHM creep shifts the effective measurement window and silently re-tunes the filter match. Narrow-binned sources with documented spectral bandwidth certificates keep discrimination thresholds repeatable across high-speed, 24/7 conveyor installations with minimal tuning between batches.
FAQ
Q: Why not buy the narrowest FWHM available?
Below 30 nm the sorting contrast gain is marginal for broad water and fat features, while radiometric flux falls in InGaAsP dies. You pay efficiency for spectral purity the application cannot use.
Q: How does junction temperature affect filter design?
Peak wavelength shifts and FWHM widens as Tj climbs. Define the filter passband around the hot spectrum with margin at both edges, then verify the overlap at maximum load and end-of-life.
Q: Does FWHM matter more than center wavelength?
Both matter, as one decision. The center wavelength sets which absorption feature you probe; FWHM sets how cleanly you probe it. Specify them together against the target spectrum.
