LCD Backlight PWM and Camera Banding: Test the Capture Setup Before Tuning

LCD backlight PWM camera banding is often a capture artifact before it is a display defect. A display that shows horizontal bands in a phone video is often reported as a flicker defect. The camera may be sampling a PWM backlight with an exposure that is not synchronized to the dimming period, or it may be beating the panel refresh against its own frame rate. Before changing the backlight, the capture setup has to be understood. This guide separates PWM banding from display scan artifacts and sets out the camera, luminance and thermal records that make the comparison valid.

How does LCD backlight PWM camera banding depend on exposure?

Banding depends on exposure relative to the PWM period.

A rolling shutter exposes rows at different times, so a PWM backlight unrelated to the exposure produces bright and dark bands across the frame.

A PWM backlight switches on and off at a fixed frequency. A camera exposes each row for a finite time. If the exposure is short relative to the PWM period, different rows sample different parts of the on/off cycle, and the frame shows bands. If the exposure covers several PWM periods, the modulation averages out and the bands disappear.

The band pattern moves when the PWM frequency, the camera frame rate or the exposure changes. That movement is a clue: a static band that stays in the same place across recordings is more likely to be a panel or optical artifact, while a band that scrolls is more likely to be a beat between the camera and the light source.

The duty cycle changes the average light but not the switching frequency; a short duty at a high frequency can still band if the exposure is shorter than the period. The driver’s PWM frequency and duty cycle are the starting point. The camera’s exposure and shutter type determine how those are sampled. Both belong in the test record.

What changes between rolling and global shutters?

A global shutter exposes all rows at once.

A global-shutter camera samples the frame at once, so PWM appears as overall brightness variation rather than spatial bands, though flicker still affects exposure.

A rolling shutter’s readout time can be longer than the exposure, which means the top and bottom rows sample different PWM cycles even at a moderate exposure. A rolling shutter reads rows sequentially. The time between the first and last row is the readout time, and during that time the backlight may change. The result is a spatial gradient or a set of bands. A global shutter exposes the whole sensor simultaneously, so every pixel sees the same part of the PWM cycle. The frame may be uniformly brighter or darker, but it does not show a horizontal band from the PWM.

Both shutter types can still show display scan artifacts. A panel that refreshes at a rate close to the camera frame rate can produce a moving band even with a constant backlight. The shutter type changes how that artifact appears, not whether it exists.

Record the shutter type and the readout time. A rolling-shutter capture with a long readout is more sensitive to PWM than a global-shutter capture with a short exposure.

Which PWM settings are supported by the actual driver?

The driver defines usable PWM frequency and minimum pulse.

The backlight driver’s datasheet and registers determine PWM range, minimum pulse and hybrid dimming; a frequency outside that range may not be achievable.

A backlight driver has a limited range of PWM frequencies and a minimum pulse width. Setting a frequency outside the range may be ignored, clamped or produce an unstable output. The minimum pulse determines the lowest duty cycle that can be reproduced, which matters at low brightness.

The LP8556 backlight driver datasheet is an example of a device with its own dimming and fault behaviour, and the LP8860-Q1 datasheet documents a different automotive driver with its own dimming and diagnostics; the actual frequency and duty limits come from that document and from the LED string. A driver that uses hybrid dimming may switch between PWM and analog control at a crossover, which can change the banding behaviour.

A driver that accepts a frequency command may round it to the nearest achievable divider, so the measured frequency should be recorded rather than the requested value. Record the requested frequency, the actual measured frequency and the duty cycle. A setting that is accepted by the API is not necessarily the setting the hardware produces.

How should brightness be held constant during comparison?

Hold measured luminance constant, not duty percentage.

Two backlights at the same duty can emit different luminance; compare at measured luminance and record the setting that produced it.

Banding depends on the modulation depth and the average light level. If two panels are compared at the same duty cycle but different luminance, the difference in banding may come from the luminance, not from the PWM design. Measure the luminance with a photometer and set each panel to the same value before recording.

Ambient light also changes the camera’s exposure and the perceived contrast. Use a controlled room, or at least record the ambient level. If the product has automatic brightness control, disable it or fix the ambient condition.

Record the luminance, the duty cycle and the PWM frequency for each capture. A comparison without all three is not reproducible.

When does display scan timing contribute a separate artifact?

Display refresh can beat with the camera frame rate.

A panel refresh not synchronized to the camera can produce moving bands even with constant backlight, so separate the two before blaming PWM.

A display refreshes at its own rate, and a camera captures at another. If the two are close but not equal, the difference produces a slow beat that appears as a moving band or a rolling shutter artifact. The band may look similar to PWM banding, but it comes from the panel scan rather than the backlight.

A band that moves slowly across the screen is often a beat; a band that stays fixed while the camera moves is more likely to be a panel or optical artifact. Separate the sources by holding the backlight constant and changing only the panel refresh, or by holding the refresh and changing only the PWM. A band that moves with the refresh is a scan artifact; a band that changes with the PWM is a backlight artifact. A band that depends on both may be a beat between the two.

Record the panel refresh, the camera frame rate and the exposure. The relationship between them explains the motion of the band.

Which camera settings belong in the test matrix?

Fix exposure, gain, shutter and white balance.

Record camera model, shutter type, exposure, gain, frame rate and white balance, and keep them fixed across the comparison.

The test matrix should list the camera model and lens, the shutter type, the exposure time, the gain, the frame rate and the white balance. It should also record the distance, the angle and the lighting. A change in any of these can change the banding pattern or the visibility of the artifact.

A fixed white balance matters because automatic white balance can shift the colour of the bands and make two captures look different. For a comparison between two panels or two PWM settings, only one variable should change. If the camera settings change between captures, the result cannot be attributed to the display.

Record the settings in the file metadata or in a companion note. A video without its settings is an anecdote, not evidence that can be checked again.

How should thermal and dimming trade-offs be recorded?

Thermal change shifts LED output and dimming behaviour.

As the backlight heats, LED efficiency and driver behaviour change, so a cold-start result may differ after the product warms up.

A backlight that is stable at room temperature may show different banding when the LED string heats. The forward voltage changes, the driver’s operating point shifts and the dimming linearity can change. A product that is judged at cold start may fail after a long run.

Record the temperature of the panel and the driver, because the driver may heat differently from the panel. Record the panel temperature at the time of capture, the warm-up time and the brightness setting. Repeat the capture after the product has reached thermal equilibrium. If the banding changes with temperature, the result is a thermal design issue as well as a PWM issue.

The industrial LCD lifespan and backlight MTBF guide covers the long-term side of the same thermal question.

What evidence distinguishes a camera artifact from a panel defect?

Optical measurement separates the display from the camera.

A photodiode can measure the backlight waveform directly; if it is stable while the camera shows bands, the artifact is in the capture setup.

A photodiode or fast optical sensor can measure the light output without a rolling shutter. If the waveform is stable and the camera shows bands, the display is not modulating in the way the camera suggests. If the waveform shows the same modulation, the backlight is the source.

Compare the photodiode waveform and the camera frame at the same moment; a modulation that appears in one but not the other locates the artifact. The optical measurement should be taken at the same luminance and the same PWM setting as the camera capture. Record the sensor model, its bandwidth and the oscilloscope settings. A slow sensor can average the modulation and hide it.

The table below is a template for the capture record. It is not a set of limits; the values come from the display, the driver and the camera documentation.

PWM banding capture record

Workload or state	Pixel or buffer format	Measured traffic or energy	Constraint	Evidence source
Full-white screen	Panel native format	Backlight at target luminance	Fixed ambient light	Photometer and camera
Dimmed screen	Same format	Low duty cycle	Driver minimum pulse	Driver read-back
PWM frequency sweep	Same format	Modulation at each setting	Camera exposure fixed	Camera and photodiode
Panel refresh sweep	Same format	Scan beat with camera	Backlight constant	Camera recording
Thermal soak	Same format	Output after warm-up	Panel temperature recorded	Thermocouple and photometer

The IEEE 1789 standard addresses the relationship between modulation frequency, depth and flicker perception in LED lighting; it does not define a camera-capture method, but it explains why frequency and duty matter. The ST AN4861 LTDC application note covers the display controller timing for specific STM32 families. The LVGL 9.2 display porting guide documents the display update path that sets the refresh behaviour.