One IR LED not lighting: test board illuminators the phone camera way

You cover the light sensor on a day/night board camera, hear the mechanical click of the IR-cut filter, and the preview goes dark — but the two on-board IR LEDs that should take over show nothing to your eye. Through a phone camera, one glows faint purple and the other stays black. That combination is a common failure on compact night-vision modules, and it has already told you most of what you need: the filter assembly works, the driver has power, and the problem is local to one LED or its solder joint.

This guide walks the diagnosis in order: what the click proves, how to see near-infrared with a phone, how to read which LEDs light, and how a multimeter pins the failure to one branch. You will finish knowing whether to reflow a joint, change one LED, or replace the illuminator.

What the click proves

The click is the IR-cut filter assembly: a small motor or solenoid slides a piece of glass out of the optical path so the sensor can see near-infrared. That mechanism and the LED illuminator are separate circuits that merely share the same night-mode decision. On several modules of this class the filter motor is even switched from the camera's indicator-LED control line, which is exactly why the filter can click away while the illuminator sits dead.

The click plus a clean color image in daylight clears the whole sensor path. The problem lives in the illuminator chain: the LEDs themselves, the current-limiting driver, or the trace between them. The full story of what the moving filter does to color and near-infrared response is in color camera near-infrared imaging and IR-cut filter options.

Check for a night-mode jumper before anything else. A jumper left open kills both LEDs at once, so your single dark LED is not a jumper problem — but a jumper set to force-on removes the light-sensor threshold from every test that follows.

See near-infrared: the phone camera test

Your eye sees 850 nm only as a faint deep-red dot from a powerful source, and 940 nm not at all. A silicon sensor answers from roughly 400 nm to about 1100 nm, so a phone camera — whose IR-cut filter always leaks a little — shows a running 850 nm LED as a bright purple-white spot on screen. At 940 nm the silicon response and the filter both work against you: many phones show it dimly, some not at all. The two wavelengths trade exactly this way — at the same drive current an 850 nm LED delivers more optical power, while 940 nm buys total invisibility to people. The wider band trade-offs are covered in SWIR vs NIR vs visible imaging: which band to inspect with.

Run the test like this:

1. Darken the room. Daylight swamps the faint spots on the phone screen.
2. Power the module, cover the light sensor — or set the night-mode jumper to force on — and confirm the click.
3. Open the phone's camera app. If the rear camera shows nothing, switch to the front camera; its IR-cut filter is usually weaker.
4. Hold the phone 10-20 cm from the board and read the screen, not the board.
5. Record one frame with both LEDs in it: which glow, how bright, and whether the brightness is steady or flickering.

Read the pattern before you open the toolbox

One lit LED beside one dark LED also settles the wiring topology. If the two LEDs sat in series, a single open LED would break the loop and both would go dark; only parallel branches, each with its own current path, can fail one at a time. Your board is parallel, so the dark branch is isolated by construction.

Confirm it with a multimeter

Unpowered, diode-test mode across each LED. A healthy IR LED reads about 0.9-1.2 V forward and open (OL) in reverse. Open in both directions means the die has failed; a near-zero reading both ways means it is shorted.

Powered, DC voltage across each LED, night mode forced. This reading says where the branch is broken. The arithmetic behind it:

Take a typical small module: a 5 V rail feeds each LED through its own 33 Ω resistor, and the LED forward voltage is 1.3 V at working current.

• Current per LED: (5 - 1.3) / 33 = 3.7 / 33 ≈ 0.11 A, about 110 mA.
• Across the glowing LED the meter reads about 1.3 V; its resistor drops the remaining 3.7 V.
• Across the dark LED you read close to the full 5 V: no current flows in that branch, the resistor drops nothing, and the whole rail appears across the open LED.

The rail-voltage reading is the conviction. Everything upstream of the LED is intact, so the failure is the die or its solder joint. A reading of 0 V across the dark LED instead means the die is shorted or power never reaches that branch — follow the trace back toward the resistor and the driver.

The resistor also explains why these joints crack. 0.11 A squared through 33 Ω is about 0.4 W dissipated in a small surface-mount part, so the board heats and cools every night cycle and the solder works itself apart.

The fix, by verdict

Open LED — rail voltage across it, open on the diode test. Inspect under magnification first. A cracked joint at a pad reflows in seconds with a fine tip and flux; if the joints are sound, replace the LED. Match wavelength, package and viewing angle to the surviving part, and replace both LEDs of a pair when they light one scene: a new LED beside an aged one leaves a brightness step.

Dead branch — 0 V across the dark LED. Check continuity from the rail to the branch resistor. If the driver is the casualty, rebuilding it as a constant-current source is a small job; a constant-current LED driver calculator for IR illuminators does the arithmetic for the current the LEDs were designed around.

Module beyond repair. Swap the whole camera board — board level camera selection for embedded vision machines covers what to check — or move the illumination off the sensor board entirely. A separate infrared light is easier to service than a pair of LEDs on a flex cable:

The infrared units in that selection run 730 nm to 856 nm. Drive one through a controller sized for its current; LED strobe duty cycle: matching lights and controllers covers the arithmetic.

When the usual answer is wrong

No phone shows anything, both LEDs dark. Try a second phone and the front camera before condemning the LEDs: rear cameras with strong IR-cut filters hide 850 nm, and 940 nm hides from nearly every phone. A mono board-level camera with no IR-cut filter makes a dependable bench viewer for 850 nm.

No click. Then the filter is stuck, and the daylight image should carry a pink cast. That is a different repair, and the LEDs may be perfectly healthy.

Everything cold at the pads. Reseat every flex connector first: the sensor can run while the illuminator pins of the same cable are open. Camera GPIO wiring: preventing damage and triaging power problems covers the triage order for board power issues.

Both LEDs glow but the preview stays black. The LEDs are innocent. Chase exposure and the timing between light and frame — the failure path in strobe light on but camera image is black: fix the timing.

A bench camera that sees both bands

A phone is a pass/fail tool, not a measurement. If you test modules in batches or need to file numbers, a short-wave infrared camera images both common wavelengths with the same sensitivity and turns the glow into data:

The LS-IRC-0029 is a 0.3 MP short-wave infrared area scan camera, GigE, mono, with 15 μm × 15 μm pixels and a C-mount.

For more pixels on a small board:

The MV-CI013-GS-NN is a 1.3 MP SWIR camera with an IMX990 sensor, GigE, and a fanless build.

One caution before you add a lens: glass corrected for the visible band loses transmission and shifts focus out in the short-wave infrared, so read SWIR lens and UV lens basics: what changes outside visible light first. Both cameras sit in the infrared cameras section, and the camera finder filters by interface and sensor format.

Checks before you order parts

• Confirm the click with the sensor covered: the filter assembly works.
• Set any night-mode jumper to force on, so the light sensor is out of the test.
• Photograph both LEDs in one phone frame in a dark room; note which glow and any flicker.
• Diode-test each LED unpowered: 0.9-1.2 V forward with open in reverse is healthy.
• Measure DC across each LED powered: 1.2-1.6 V is healthy, near the rail means open, 0 V means short or dead branch.
• Reseat the flex connectors before condemning any component.
• Note the LED wavelength, package and viewing angle from the board or its data sheet before ordering.

Keep the next one alive

Heat is what kills these boards. The classic reliability rule of thumb — service life roughly halves for every 10 °C rise in junction temperature — argues for running the LEDs at 80% or less of rated current wherever the scene allows, and for giving the board somewhere to shed heat. Strain-relieve the flex cable so vibration cannot work the joints loose. After any repair, put the module through a machine vision lighting stability test before deployment so the fix is proven, not assumed.

If the diagnosis sends you to a replacement illuminator rather than a reflow, start at the light step below: choose 850 nm unless the glow must be invisible to people, size the lit area, and the walk-through adds a matching controller. The design side continues across the lighting topic page.