The problem as it presented itself
The customer arrived with a working prototype and a production problem: the LEDs that passed the bench test were not the ones that survived the line. Nothing in the data was out of spec — the failures were spread across batches rather than concentrated in one lot, which ruled out a supplier problem and pointed at the process instead.
Three measurements separated the causes. First, forward voltage at a fixed low current across a sample of each batch, which showed whether the die set was consistent. Second, case temperature under the real drive condition, which showed whether the thermal path matched the assumption in the design. Third, the reflow profile logged on the line, which showed whether the assembly process stayed inside the qualified window.
The first two came back clean. The profile log showed the peak temperature sitting close enough to the limit that normal line-to-line variation pushed parts past it. This is the most common way a qualified component becomes an unqualified assembly, and it is invisible on the datasheet.
- Measure the process, not just the part. A component that meets its specification can still be assembled outside its qualification.
- Log the profile, do not assume it. Reflow profiles drift as boards and fixtures change.
- Look at failure distribution first. Failures spread across batches point at the process; failures concentrated in one batch point at the supply.
Once the profile was corrected and re-qualified, the field return rate dropped to the level the component datasheet predicts. The component never changed — only the process around it.
Modern travel documents are engineered to be read by machines — but only under the right light. This project shows how 850nm infrared LED arrays raised first-pass passport reading at automated border gates to 99.7%, while adding invisible anti-spoofing illumination for face capture.
Application Background
An airport operator expanding its automated border control program found that gate throughput was capped not by matching algorithms but by the document scan itself. Under the gates’ original white LED lighting:
- Glossy polycarbonate data pages threw specular hot spots that occluded the machine-readable zone (MRZ), forcing retry loops
- Visible lighting produced pupil artifacts in face capture and caused discomfort for travelers moving through the enclosed gate corridor
- Printed or screen-based document spoofing was difficult to detect without a second, invisible illumination channel
- Failed first-pass reads routed travelers to staffed counters — at 8.3% of traffic during peak hours, the queue erased the gates’ business case
Engineering Challenge
Passport readability is fundamentally a wavelength problem. Document inks and substrates interact with infrared very differently than with visible light:
- At 850nm, the MRZ ink stays dark while the substrate brightens — contrast improves exactly where OCR needs it, and specular glare from the glossy laminate drops sharply
- Silicon camera sensors hold their best quantum efficiency at 850nm, so the 850nm/940nm trade favors 850: 940nm eliminates the faint red glow but costs sensor response, weakening the link budget at the same drive current
- Wavelength binning of ±5nm was required across all gates so OCR thresholds stay identical gate-to-gate — an un-binned mix shifts contrast unpredictably
- The face-capture channel needed pulsed IR synchronized to camera frames, doubling as a spoofing check: printed photos and screens respond differently to IR than live skin
Solution & Key Components
Queendom supplied binned 850nm illumination arrays for the document scan window and the face-capture channel:
| Parameter | Value | Design Note |
|---|---|---|
| Peak wavelength | 850 nm, ±5 nm binned | Uniform OCR contrast across all gates |
| Illumination uniformity | > 85% across 300 mm window | Flood optics, mixed wide/narrow emitters |
| Drive mode | Frame-synchronized pulsed | Captures only when the camera integrates |
| Visible red tail | Minimal — near-covert to travelers | No glare or pupil artifacts in face channel |
| Duty design | 24/7 rated, 50,000 h (L70) | Withstands continuous gate operation |
| Platform | High-power IR + compact DIP variants | Scales from kiosks to full-height gates |
Integration notes for gate and kiosk builders: the document channel runs continuously during the scan window, while the face channel fires in sync with each camera frame — the same array doubles as the spoof-detection light source, so no separate IR flash unit is needed.
Results
After the retrofit, retry loops at the document window nearly disappeared, and the synchronized IR face channel began flagging screen-and-print spoofing attempts that had previously passed. The operator extended the illumination design to its self-service bag-drop kiosks.
Recommended Products
Frequently Asked Questions
Why 850nm instead of 940nm for document scanning?
850nm sits where silicon sensors are most sensitive, so the illumination achieves target irradiance at lower drive current — and document inks still show the contrast reversal that OCR needs. 940nm is fully invisible but wastes sensor response; it’s the right choice only when absolute visual concealment outranks link budget.
Does the faint red glow of 850nm bother travelers?
Barely — at the drive levels used in document windows the residual visible tail is a dim deep-red hint, not glare. Where even that is unacceptable, the face-capture channel can move to 940nm because it works at closer range with a stronger return signal.
Why does wavelength binning matter across gates?
OCR thresholds are tuned to a specific contrast signature. If one gate’s array peaks at 845nm and another’s at 860nm, the same passport reads differently — binning to ±5nm keeps every gate’s illumination inside the tuning window, which is why our arrays ship sorted, not mixed-bin.