Pulsed IR illumination timing for low-light image capture

You are running a small serial camera at night from batteries, lighting the scene with a 940 nm infrared LED, and capturing one 8-bit raw frame every 0.5 s. The camera needs 150-200 ms of shutter time in low light, then tens of milliseconds to move the bytes over a 921600 baud line. The question is whether the LED must stay on through all of that. It must not. This guide gives the pulse timing, the transfer arithmetic for 60 × 80 and 128 × 96 raw frames, the duty-cycle and battery math, and the checks that prove the light was off exactly when it should be. The method carries over to any snapshot camera: the exposure window sets the on-time, and the readout runs in the dark.

The decision: pulse the LED, switch it off at the first data byte

A snapshot camera works in a fixed order: the sensor integrates, the finished frame is written into the module's memory, and only then is it read out to the wire. Readout is a memory copy, not more exposure. So yes, you can switch the infrared light off before the transfer, and you never have to guess where the boundary sits. The moment the first data byte lands in your host's receive buffer, the frame is already in memory and the light is no longer needed.

The timing for this case: LED on just before you send the capture command, hold it through the shutter window, LED off at the first received byte. Worst-case on-time is 210 ms in a 500 ms cycle, a 42% duty cycle, and the 52 ms transfer at 60 × 80 runs entirely with the light off. The split between command, exposure and readout is the same one behind trigger call versus exposure time.

What happens in each window

The shutter window runs from command received to frame in memory; in low light it sits at the top of the 150-200 ms range for this class of camera. Inside it the sensor integrates the scene and the module applies gain and contrast to the raw data. Nothing about your scene shortens it. Exposure and image processing happen once per capture command, inside this window, not in the background between captures.

The transfer window is pure data movement: memory to UART to host at 921600 baud, 8 data bits, no parity, one stop bit, so 10 bits per byte. Headers and acknowledgements add a few dozen bytes, under 2 ms at this rate. Nothing in this window needs light: the photons have already been counted and stored.

Step by step: pulse the light around one capture

1. Frame the scene with the LED continuously on, then lock exposure and gain. Note the settings and leave auto-exposure off.
2. Put the LED behind a switch you control from the host: a logic-level MOSFET on a battery rig, or a strobe channel on a light controller in a powered station.
3. At the cycle start, switch the LED on and wait 10 ms. The diode itself is lit in well under a microsecond; the 10 ms covers driver settling and start-up margin.
4. Send the capture command.
5. Hold the LED on through the shutter window.
6. In the host's receive interrupt, switch the LED off on the first data byte.
7. Let the transfer finish in the dark; the arithmetic below sizes it.
8. Sleep the host and camera until the next 500 ms tick, then repeat from step 3.

Set the host's frame timeout above the worst-case total, 333 ms at 128 × 96, or a slow frame will read as a fault; software trigger image timeout covers the failure modes of one-shot capture. If the LED is on but the frame arrives black, the light is firing outside the exposure window, and strobe light on but camera image is black walks that diagnosis.

The arithmetic: one 940 nm frame every 500 ms

Inputs, each with its source:

• Shutter window: 150-200 ms, the low-light range for this class of serial camera.
• Line rate: 921600 baud with 8N1 framing, so 10 bits per byte and 92,160 bytes per second.
• Payloads: 60 × 80 = 4,800 bytes and 128 × 96 = 12,288 bytes of 8-bit raw.
• Cycle time: 500 ms. Guard after LED switch-on: 10 ms.
• LED current: 300 mA, a mid-size infrared illuminator.

The rules in words: transfer time is payload divided by line rate over ten; duty cycle is on-time over cycle time; average current is duty cycle times LED current.

Worked through:

• Throughput: 921,600 / 10 = 92,160 bytes per second.
• 60 × 80: 4,800 / 92,160 = 0.052 s, about 52 ms.
• 128 × 96: 12,288 / 92,160 = 0.133 s, about 133 ms.
• Worst-case on-time: 10 + 200 = 210 ms.
• Duty cycle: 210 / 500 = 42%, or 32% at the 150 ms end of the window.
• Average LED current: 0.42 × 300 mA = 126 mA.
• LED run time gain: 1 / 0.42 = 2.4x, and 1 / 0.32 = 3.1x at the short end.
• Headroom, worst case at 128 × 96: 210 ms on + 133 ms transfer = 343 ms, leaving 157 ms before the next pulse is due.

Decision: pulse. The ceiling is cycle time minus transfer time, 500 − 133 = 367 ms here, and your 210 ms sits well under it. Only two changes break that: a shutter window that must grow past about 350 ms at this cycle, or a cycle that must shrink below about 300 ms. Past either point the duty cycle climbs above 70% and the switching saves too little to justify the parts.

Choosing the LED, the beam and the switch

Wavelength first. A 940 nm LED shows no visible glow at all, which is why unattended night capture uses it; the cost is that silicon sensors respond less strongly at 940 nm than at 850 nm, so the exposure window stays long. If covertness does not matter, an 850 nm LED gives a stronger signal and often a shorter window, at the price of a faint deep-red glow visible at close range. The spot lights finder lists both wavelengths.

Beam next: a spot from the spot lights range for a target at several metres, a flood for a wide near scene. The pattern and power budget for wide-area infrared coverage is in infrared flood illumination for stereo depth imaging; the long-range version of the same exposure budget is in long distance camera recording for triggered infrared capture.

Current: stay at the continuous rating. A pulse of this length at this duty cycle is continuous operation by any LED standard; overdrive currents belong to microsecond and millisecond flashes at 10% duty or less, and LED strobe duty cycle sets out that boundary. A high-current infrared source is invisible and provokes no blink reflex: never look into it at close range, keep to the rated current, and spread the beam rather than focusing it.

For a mains-powered station, the same timing runs on shop parts. The LS-SPT-0133 is a 4 W spot light. The LS-CTL-0070 is a 4-channel, 200 W digital control box with a strobe feature:

If the station grows past two lights, the LS-CTL-0365 is a digital controller with 4 channels for 16-24 V lights. The two side by side:

All 2 share: Input voltage 24 V DC; External trigger voltage 5-24 V. Values as printed in each product's data sheet; open a part for its drawing and full specifications.

On the camera side, a sensor built for the short-wave infrared band collects far more of a 940 nm scene than a visible-sensor camera with its filter removed. The LS-IRC-0029 is a 0.3 MP short-wave infrared area scan camera with 15 μm × 15 μm pixels, a global shutter and a C-mount:

Pixels that large hold about four times the area of a 7.4 μm visible-sensor pixel, so at the same f-number the pulse shortens or the LED current drops. Browse infrared cameras if that trade is worth making.

Checks that prove the timing

• Image equality. Capture the same static scene with continuous light and with the pulse; the mean pixel value should match within a few counts.
• Rolling-cut test. From the safe off-point, pull the light-off moment earlier in 25 ms steps. The first dark band or brightness gradient across the frame marks where integration really ends on your unit; add 25 ms and keep it.
• Average current. A multimeter in average mode across ten cycles should read duty × LED current, 126 mA for the 300 mA example.
• Timeout. Host frame timeout at least 1.5× the worst-case total.
• Soak. After 30 minutes of cycling, the image mean and the current reading should be unchanged.

When pulsing is the wrong answer

• Streaming cameras. Exposure overlaps readout, so the first-byte cue is invalid. Keep the light on through the exposure of the frame you keep.
• Unlocked auto-exposure. A pulsed scene can make the gain hunt from frame to frame, and the first frame after any change arrives dark; first camera frame dark separates the two causes.
• Short cycles. At a 250 ms cycle with a 200 ms window the duty cycle is 84% and the switching saves almost nothing. Stay continuous and shorten the window instead, with 850 nm or larger pixels.
• Moving subjects. A 150-200 ms exposure smears anything that moves; exposure time versus shutter speed covers how to get there.
• Compensating with current. Shortening the pulse and raising the current invites start-up droop, covered in LED strobe current droop, and pushes the junction temperature. Measure usable light as in LED strobe optical rise time before you trust a hotter setting.

Checklist before you wire it

• Lock exposure and gain under continuous light.
• 10 ms guard between LED on and the capture command.
• LED off in the receive interrupt, on the first data byte.
• On-time plus transfer below the cycle time, checked at the largest frame size.
• LED at its continuous current rating, never an overdrive figure.
• Rolling-cut test passed, with margin.
• Average current over ten cycles matches the duty-cycle prediction.
• Host frame timeout comfortably above the worst-case total.

The vision configurator sizes the rest of a station, light first, and the controller step picks the channels. Bring your frame format and cycle time to the contact page and we will check the arithmetic against your camera.