A machine vision light is a current-driven device, and a lot of measurement trouble blamed on optics or code starts at the drive electronics. If the brightness at the part wobbles between frames, thresholds shift, edges move, and caliper results wander with nothing changed in the program. This page puts numbers on why LED output follows forward current rather than terminal voltage, works the arithmetic from millivolts at the junction to pixels in the image, and gives you the decision rule: where constant-current drive is required, where a constant rail supply is genuinely fine, and how to size a controller for a light you already have.
The decision: regulate current, not voltage
Drive the light with a controlled current whenever the image feeds a decision you have to defend: caliper and edge measurement, defect grading, code grading, colour comparison. Keep constant-voltage drive for lights that regulate themselves internally and for presence checks where a ±10% brightness swing cannot change the outcome.
Three numbers carry the whole argument:
- LED flux is close to proportional to forward current across the usable dimming range.
- The voltage across an LED is a poor proxy for that current: the junction's I-V curve is so steep that a change of about 1% in the voltage across the chip moves current — and with it light output — by around 10%.
- Downstream, lighting-controller application notes quote a matching rule: a 10% drop in light level shifts a threshold-based caliper measurement by about 0.5% of its value.
A constant-current controller closes its feedback loop on the current itself and adjusts its output voltage to whatever the junction demands at that instant. Rail sag, cable resistance, connector heating and the forward-voltage spread between two lights of the same type are absorbed inside the loop instead of appearing as brightness. Regulation is the main reason a dedicated light controller earns its place in a measurement cell; the five-reasons overview of lighting controllers lists the others, strobe timing first among them.
Why brightness follows current
In an LED the photon rate is set by the current through the junction; the forward voltage is the dependent variable. The diode equation makes the consequence quantitative:
Here n is the ideality factor (roughly 2-5 for power LEDs), R_s is the series and spreading resistance (a few tenths of an ohm on a 1 W chip), and Φ is the flux, which tracks current within a few percent over the usable range. Because the slope r is a fraction of an ohm, millivolts of extra voltage move the current by whole percentages — that steepness is what makes voltage a bad control knob.
A constant-voltage supply has no such loop: it holds the terminal voltage and lets the junction take whatever current that voltage implies. Any disturbance in the path lands directly on the current, and the current is what sets brightness. For how the current loop holds its setpoint against rail movement, how a constant-current driver regulates its output walks through the control stage in detail.
Worked example: 25 mV at the junction becomes 5 pixels in the image
Electrical side. Take a typical 1 W white LED chip at 350 mA with about 3.1 V across it. With n ≈ 4 and R_s ≈ 0.4 Ω:
r = (4 × 0.026 V) / 0.350 A + 0.4 Ω ≈ 0.30 Ω + 0.40 Ω ≈ 0.7 Ω
A 10% brightness change therefore needs:
ΔV = 0.10 × 0.350 A × 0.7 Ω ≈ 0.0245 V ≈ 25 mV
That is 25 mV / 3100 mV ≈ 0.8% — call it 1% — of the forward voltage. This is where the 1% voltage to 10% light rule comes from; it is junction physics, not marketing.
Optical side. Suppose the camera sees 40 mm across 2,000 pixels: 40 mm / 2000 px = 20 µm/px. The part is a Ø20.000 mm shaft with a ±0.05 mm tolerance. Let the light level fall 10% — a warming light on a sagging rail will do exactly that. The grey-level threshold crossing moves inward and the measured diameter shifts by about 0.5%: 20 mm × 0.005 = 0.100 mm = 100 µm, which at 20 µm/px is 100 µm / 20 µm/px = 5 px. One 10% brightness error consumes the entire ±50 µm band (100 µm total) with no code change anywhere.
Supply side. A standard industrial rail is commonly specified ±5%; at 24 V that is ±1.2 V of variation before load transients and cable drop. Tens of millivolts at the junction are enough to move light by double-digit percent, so the junction must not see the rail directly: something between them has to regulate current.
Decision: if your tolerance band is thinner than about 0.5% of the feature you measure, regulated current is not optional.
What else moves the light besides the supply
Temperature works on both terms. Forward voltage falls by about 2 mV per °C per chip on typical white and blue LEDs, so under constant voltage a warming light draws more current, heats further and can walk away from its operating point. Efficiency falls at the same time: flux drops roughly 10-15% between a 25 °C and a 60 °C junction on typical white chips. Constant-current drive pins the current so only the efficiency term remains, and strobing keeps the junction near ambient so that term largely disappears too.
Aging reduces flux over operating hours — the L70 lifetime trade-off and why strobe mode extends LED life quantifies it. No drive mode reverses aging; current control simply stops the supply from adding drift on top of it, and strobe cuts the hours of actual emission dramatically.
Everything downstream of the electronics matters too: ambient light leaking past the diffuser — the ambient-light handling for pick-and-place vision covers that fight — and camera-side automation quietly masking the drift until it saturates, which the auto-exposure cap and brightness-drift articles untangle.
Constant voltage vs constant current, disturbance by disturbance
| Disturbance | Constant-voltage drive | Constant-current drive |
|---|---|---|
| Rail tolerance ±5% (±1.2 V on a 24 V rail) | Current follows the rail; brightness moves double-digit percent | Current held at setpoint |
| Junction warms 35 °C | Vf falls ~2 mV/°C, current rises, output drifts unpredictably | Current pinned; only the 10-15% efficiency term remains |
| Cable and connector resistance | Steals volts from the junction unpredictably | Absorbed within the controller's compliance voltage |
| Swap in a second light of the same type | Brightness differs with Vf spread | Same setpoint, near-identical flux |
| LEDs age | Nothing compensates | Nothing compensates: strobe and closed loop are the remedies |
The pattern is consistent: constant voltage passes every upstream disturbance straight to the junction; constant current stops it at the controller. Which mode fits follows from what the light actually is, covered next. For the same comparison on general lighting hardware, this constant current versus constant voltage selection guide runs the decision for ordinary LED drivers.
Where a fixed rail voltage is the right answer
Many production lights are sold with a stated rail voltage as their input — 24 V is the norm in machine vision — and carry their own regulation on board. For those, a clean supply rail is the correct feed and a constant-current source behind that input is wrong: a CC loop that sees less current than its setpoint will keep raising its output voltage to force the current, and the light's input stage takes the stress. Match the supply type to the light's stated input; how driver ratings separate constant-current from constant-voltage parts decodes the labels.
Presence checks, rough vision-guided picking with generous grip offsets, and indicator on/off verification tolerate a few percent of brightness wobble because the decision threshold sits far from the edge; the ambient-light article above is the playbook there. And if the process forces continuous light in a drifting environment, closed-loop lighting control — the camera reading a reference patch and trimming the light — recovers repeatability from the software side.
Sizing the controller for your light
Work the numbers from the light's label. Example: a light rated for a 24 V input at 700 mA draws 24 V × 0.7 A = 16.8 W. Add roughly 30% headroom for strobe pulses, cable losses and aging: 16.8 W × 1.3 ≈ 22 W per channel. Confirm the channel's current range brackets the light's 700 mA and that its compliance voltage covers the chain's forward voltage plus a couple of volts for cable drop.
For a typical multi-channel bench, the LS-CTL-0180 provides 4 constant-current channels at 120 W.
- Output
- Constant current
- Channels
- 4 channels
- Control
- Digital
- Feature
- Constant current
Long bars and overdrive strobing need more per channel: the LS-CTL-0054 is rated 240 W per channel and 320 W total.
- Output
- Constant current
- Channels
- 2 channels
- Control
- Analog
- Feature
- Constant current
Where the light's input is a rail with its own driver, the controller must deliver constant voltage: the LS-CTL-0487 offers both output modes on 2 channels, 240 W, RS-232.
- Output
- Constant current, Constant voltage 24 V
- Channels
- 2 channels
- Control
- Digital
On the load side, a 372 × 28 mm white diffused bar light such as the LS-BAR-1276 is a representative rail-input unit for the sizing above, and small fields take a spot unit such as the LS-SPT-0133 (4 W) instead. If the load is still open, the bar lights and spot lights categories filter by size and colour, and the lights finder narrows by wavelength.
- Size
- 200-400 mm
- Type
- Diffused
- Voltage
- 24 V
- Size
- Under 50 mm
- Voltage
- 24 V
Set-up checks before ordering
- Input type first: a bare LED chain takes constant current set in mA; a stated rail voltage input with an internal driver takes a clean supply at that voltage. Every other choice follows from this answer.
- Power headroom: channel rating of at least 1.3 × light power (a 16.8 W light needs a 22 W or larger channel).
- Current range brackets the light's rated current; compliance voltage covers forward voltage plus cable drop.
- Channel count matches the light's segments: white is one channel, RGB or RGBW is one per colour.
- Wiring and connectors: the controller-to-light cable and adapter guide covers pinouts and current per pin.
- Strobe intent: for pulses shorter than a millisecond or overdrive plans, confirm the strobe feature and set the duty cycle per the strobe duty-cycle matching guide, and budget the rise time into the exposure window.
When the usual advice does not apply
Overdrive strobing bends the proportionality: at several times rated current, flux per pulse no longer scales linearly and pulses droop within a burst — the current-droop diagnosis and the hidden-strobe technique show what to expect and how to keep the flash invisible on the line. The constant-current loop is still what holds the per-pulse amplitude repeatable there.
If brightness still wanders on a properly current-driven system, the electronics are no longer the suspect: work through camera auto exposure and gain hunting, mains flicker coupling into the sensor, and ambient light. The frame-to-frame brightness flicker fix is the fastest differential test.
The decision, case by case
- Measurement or graded image, light with a bare LED input: constant current, setpoint in mA, logged in the recipe.
- Light with a stated rail voltage input and internal driver: clean constant-voltage supply, and no CC source behind it.
- Presence checks with wide thresholds: either mode; spend the effort on ambient control instead.
- Continuous light in a drifting environment: closed-loop control recovers what the drive cannot.
- Short strobe pulses: constant current plus rated duty, with rise time and droop budgeted.
Read the rated current and power off your light's label, answer the input-type question, and the rest is sizing. The controller finder filters by channels, power and strobe, the configurator's controller step walks the same questions in order, and if you send us the light's ratings through the contact page, we will match the channel and check the headroom with you.
Questions
Can I run a 24 V machine vision light straight from a switching power supply?
For a light with its own internal driver, yes: keep the rail inside its stated tolerance and keep cable runs short. For a bare LED light, no — the rail's tolerance, ripple and cable drop all reach the junction, where roughly 1% of extra voltage means roughly 10% of extra brightness. For any measurement task, put a constant-current controller between the supply and the light and set the value in milliamps.
How much does LED brightness drift with temperature?
Two separate effects. Forward voltage falls about 2 mV per °C per chip, and flux efficiency drops roughly 10-15% between a 25 °C and a 60 °C junction on typical white LEDs. Constant-current drive pins the voltage effect; pulsing keeps the junction near ambient and mostly removes the efficiency term as well. A light left on continuously in a warm cell reads dimmer at the end of a shift than at the start.
Does PWM brightness control fix constant-voltage drift?
No. PWM changes how long the LED is on, not how hard each pulse drives it. If the current per pulse has drifted because the rail moved, every pulse is proportionally wrong and the duty cycle just rescales the error. Repeatability comes from regulating the current amplitude first; PWM on top of a constant-current source then gives clean, repeatable brightness steps in the image.
What controller power do I need for my light?
Take the light's rated power and add about 30% headroom for strobe pulses, cable losses and aging: a 16.8 W light (24 V, 700 mA) sits comfortably on a 25 W or larger channel. Also confirm the channel's current range includes the light's rated current and that its compliance voltage covers the forward voltage plus a couple of volts for cable drop.
Is constant-current drive still right for short strobe pulses?
Yes, and it matters more there. Overdrive strobing pushes several times the rated current for microseconds, and light output per pulse follows the current amplitude closely. A voltage source cannot hold that amplitude against a rising junction temperature within a burst. Keep peak current and duty cycle inside the light's strobe rating, and expect a few percent of droop across the burst even then.




