An indirect time-of-flight (iToF) sensor hides its most important numbers behind two datasheet fields: modulation frequency range and ambient-light behaviour. Get those two right and the depth map is usable; get them wrong and no amount of later tuning recovers it. This guide gives you the arithmetic a sensor choice needs: how far a modulation frequency sees before phase wraps, how depth error scales with phase noise, what sunlight does to the measurement, and which sensor features — BSI pixels, pixel-level ambient cancellation, 940 nm illumination — keep that accuracy alive in daylight. The worked examples use current VGA-class sensors as the reference; the method transfers to any iToF part.
What is an indirect time-of-flight sensor actually measuring?
An indirect time-of-flight sensor does not time a single photon's trip; it measures the phase shift between a modulated flood of infrared light at 850 nm or 940 nm and the echo that returns, then converts that phase into a distance for every pixel in the array.
Each pixel integrates charge in at least four windows, phase-stepped against the emitted waveform, and the samples combine into one phase and one amplitude. That is why iToF depth noise, motion artifacts and ambient-light problems all trace back to charge accounting, not to timing circuits. Wikipedia's time-of-flight camera overview describes the same principle from the timing side: each pixel measures the time the light has taken to travel from the illumination unit to the object and back.
Because the windows are sampled in sequence, fast motion between them skews depth. The guide on in-sensor four-phase capture covers that failure mode and the capture schemes that reduce it.
What does modulation frequency set on an iToF sensor?
Modulation frequency sets the unambiguous range, the distance window in which one phase value maps to exactly one distance: d_unamb = c/(2·f_mod), which gives 1.5 m at 100 MHz, 3.0 m at 50 MHz and 7.5 m at 20 MHz, using c = 3×10⁸ m/s.
At 100 MHz the modulation wavelength is 3 m and every distance beyond the 1.5 m window folds back inside it: a part at 2.2 m reports 0.7 m. The fix is multi-frequency capture. A sensor whose modulation range runs from a few megahertz into the 100–120 MHz class measures coarse phase at a low frequency to resolve the wrap, then fine phase at a high frequency for precision. At 10 MHz the window is 15 m; at 2 MHz it is 75 m.
| Modulation frequency | λ_mod = c/f | Unambiguous range c/(2f) |
|---|---|---|
| 2 MHz | 150 m | 75 m |
| 10 MHz | 30 m | 15 m |
| 20 MHz | 15 m | 7.5 m |
| 50 MHz | 6 m | 3.0 m |
| 100 MHz | 3 m | 1.5 m |
| 120 MHz | 2.5 m | 1.25 m |
The capture-side consequences — what the wrap does to a moving scene and how to fix the range before ordering — are covered in choosing a time-of-flight unambiguous range for depth capture.
How much depth precision does a higher modulation frequency buy?
For the same phase noise, depth error scales as 1/f_mod: a 1° phase error is about 20.8 mm of depth error at 20 MHz and about 4.2 mm at 100 MHz, so moving up in frequency buys precision without touching optics, exposure or lighting.
| f_mod | Depth error per 1° of phase noise |
|---|---|
| 20 MHz | 20.8 mm |
| 30 MHz | 13.9 mm |
| 50 MHz | 8.3 mm |
| 100 MHz | 4.2 mm |
| 120 MHz | 3.5 mm |
The mapping assumes an equal signal-to-noise ratio at both frequencies; modulation contrast falls somewhat as frequency rises, so real gains land slightly below the ratio, and multipath plus temperature drift stack on top of it. Read the depth-accuracy plot at your working distance and check it against this ratio before committing.
Why does bright ambient light ruin iToF depth data?
The photodiode cannot tell your VCSEL's photons from the sun's: daylight adds an unmodulated DC photocurrent that injects shot noise growing with the square root of its level and, past a point, fills the pixel's charge wells so the phase measurement collapses entirely.
Three things happen as ambient rises. Shot noise on the DC background scatters depth even when nothing saturates. DC charge consumes full-well capacity, so the modulated signal clips and pixels drop out or report constant depth. Residual offset biases phase into a distance error that indoor calibration cannot see. The scale is stark: full sun delivers about 100,000 lux at ground level against roughly 500 lux on a lit factory floor, two hundred times.
A narrowband filter is the first defence. Silicon integrates sunlight across roughly 600 nm of spectrum, and a 10 nm FWHM filter at the VCSEL line passes about 1.7% of it; at 940 nm the water-vapour dip trims the passband further. Indoors the picture inverts: mains lighting flickers at 50/60 Hz, slow next to a microsecond-scale ToF exposure, so it behaves as harmless DC. That is why iToF works in an unlit warehouse and struggles on a sunny loading dock. Cutting exposure against glare costs frame rate, and the ToF exposure and frame-rate trade-off article quantifies that limit.
Which sensor features keep iToF working in daylight?
Two features matter most for daylight: a pixel-level ambient-cancellation circuit (marketed under names such as pixel drive) that subtracts the sun's DC component before it saturates the pixel, and 940 nm illumination, where atmospheric water vapour trims the sun's own light by several-fold.
The cancellation circuit samples the ambient component separately from the modulated charge and removes it, keeping the correlation alive where an unaided pixel would clip. When comparing sensors, ask for the documented test conditions behind that feature — irradiance, working distance, target reflectance — then hold them against your dock door or press shop, not against a demo scene.
On wavelength, the pair to weigh: at 940 nm the solar water-vapour dip cuts ground-level irradiance several-fold, while silicon responds more strongly at 850 nm. The indoor-versus-outdoor wavelength choice follows from that pair.
How much does pixel size and BSI change iToF performance?
Pixel pitch sets the sensitivity-versus-resolution trade: first-generation iToF pixels ran about 10 µm, current BSI parts sit near 5 µm on a 1/4.5-type optical format, and the newer parts keep most of their sensitivity because back-side illumination moves the wiring away from the light path.
An iToF pixel is a charge bucket with a stopwatch: full-well capacity decides how much DC it survives, and quantum efficiency decides how much modulated signal each photon leaves. Front-side designs lose pixel area to the wiring stack above the photodiode; BSI flips the wafer so light reaches the photosensitive layer first, raising the effective fill factor. More signal per photon means shorter exposure for the same depth noise, and ambient charge scales with exposure, so a 2× sensitivity gain halves the sunlight problem too.
Pitch also decides optics. A 5 µm-pitch VGA (640 × 480) array is about 3.2 × 2.4 mm, small enough for compact, low-cost lenses. Keep the review-meeting distinction straight: pixel count is spatial resolution in pixels; depth precision is millimetres, governed by the formula in the section above. Frame rate adds motion capability — current VGA-class parts reach the 50–60 fps class — but only while exposure at your ambient level allows it; the depth-camera frame-rate limit article shows how exposure caps fps.
When is an iToF sensor the wrong choice?
Step away from iToF when the unambiguous window your exposure budget supports is smaller than the working distance, when direct sun dominates and no filter plus cancellation survives it, or when the part is dark or specular — structured light, laser profiling and passive stereo then hold accuracy better.
The decisions fall out quickly:
- Indoor cell up to 2–3 m with matte parts: iToF. Active, compact, no external triggering.
- Bench distances of 0.3–1 m, dark or shiny parts, dents and volume checks: projected structured light, which encodes absolute position without phase wrap.
- Long outdoor ranges: passive stereo with a wide baseline; the long-range stereo depth accuracy guide shows how baseline sets the error.
- Sub-millimetre cross-sections of a known part: a laser profiler.
The industrial 3D camera selection guide walks the full stereo, ToF and structured-light decision with the same trade-offs. Where those limits bite, projected structured-light cameras and laser-galvo stereo cameras are the ones doing the job:
- Type
- Stereo / structured light
- Type
- Stereo / structured light
What should you compare on two iToF datasheets before buying?
Read five fields in order — modulation frequency range, frame rate, pixel pitch and optical format, BSI or front-side, and the documented ambient-light test — and you can rank two iToF datasheets in ten minutes, because every other spec follows from them.
- Modulation frequency range. The top end sets depth precision, the bottom end sets the unwrap window. A 2–120 MHz part pairs a 75 m coarse window with a 1.25 m precision window; a part fixed at 100 MHz caps the cell at 1.5 m however well it resolves.
- Frame rate at full resolution, with its exposure conditions. A 50–60 fps figure means little if your bright cell forces 10 ms exposures; check what exposure the spec assumes.
- Pixel pitch and optical format. About 5 µm on a 1/4.5-type sensor keeps optics small; about 10 µm buys charge capacity for bright cells.
- BSI or front-side. BSI wins photons, and in a bright cell that means exposure headroom.
- Ambient-light rejection, with its test conditions. Which mechanism — pixel-drive-class cancellation, filter, wavelength — and at what irradiance, distance and reflectance it was measured. Compare those numbers with your scene.
- Interface and integration. A MIPI module feeds an embedded compute board; an industrial housing with a standard interface feeds a controller. Decide the cabling and compute path first.
Which depth camera should you start with?
If your computed unambiguous window covers the working distance and the ambient light is indoor-level, a Depth / ToF camera is the shortest path from this arithmetic to a live depth map. Filter Depth / ToF cameras in the camera finder, browse the 3D sensors category, or start at the camera step of the configurator. A current example of the Depth / ToF class in the shop:
- Type
- Depth / ToF camera
If the scene is shiny, dark or outdoor-bright, send the layout through the contact page and we will help you weigh iToF against structured light and stereo. The 3D vision guides cover the rest of the pipeline, from point clouds to robot hand-off.
Questions
What modulation frequency does a 4 m deep field need?
The unambiguous window must exceed 4 m, so f_mod ≤ c/(2 × 4 m) = 37.5 MHz. Run roughly 20–30 MHz for the coarse phase (a 5.0–7.5 m window), then a second capture near 100 MHz for the fine value if the sensor supports multi-frequency operation. A single 100 MHz modulation would fold that 4 m field into a 1.5 m window and read nonsense.
Why do outdoor ToF designs choose 940 nm instead of 850 nm?
Solar irradiance falls into a water-vapour absorption dip near 940 nm, so the sky delivers several times less light there than at 850 nm. The dip deepens in humid air and weakens on very dry days, which is why outdoor trials belong in your worst-case weather. Silicon responds more strongly at 850 nm, so indoor designs often keep 850 nm while outdoor designs move to 940 nm plus a narrowband filter.
Does frame averaging reduce iToF depth noise?
Yes, for temporally random noise: standard deviation falls with the square root of the frame count, so averaging 4 frames halves depth noise and 16 frames quarters it. The cost is latency plus motion smear between frames, so average static or slow scenes only; for fast motion, look at in-sensor capture schemes instead of post-processing.
Can an iToF camera measure black or shiny parts?
Black matte parts reflect too little of the infrared flood, so signal drops and depth noise rises until pixels report low confidence; mirror-like surfaces throw the flood away from the sensor or bounce it off nearby walls, which appears as multipath ghosts at wrong depths. For such parts, structured light at bench range or a laser profiler is the dependable choice.
Is a narrowband filter still needed when the sensor has pixel-level ambient cancellation?
Yes. The cancellation circuit keeps the sun's DC charge from saturating the pixel, but every photon a filter blocks never becomes shot noise at the sense node. A bandpass of roughly 10 nm FWHM at the VCSEL wavelength plus the cancellation scheme stack: the filter cuts noise, the circuit protects the phase from the residual DC.