Indirect time of flight: how phase becomes depth

A depth image can look stable while reporting the wrong distance. For engineers choosing a depth camera, understanding the measurement principle helps separate random noise from systematic errors that averaging cannot fix. Indirect time of flight measures the phase of returning, modulated light rather than timing an individual pulse. You can use that distinction to evaluate range, compare measurement claims and design an acceptance test before committing to hardware.

How phase samples become a distance

An indirect time-of-flight camera illuminates the scene with periodically modulated light. Each pixel correlates the returning signal with synchronized reference signals. A longer optical path delays the return and therefore changes its phase relative to the outgoing modulation.

The camera does not need to resolve the round-trip travel time directly with a separate stopwatch at every pixel. Instead, the pixel circuitry accumulates charge according to the relationship between the return signal and its modulation reference.

A simplified quadrature measurement uses correlation samples at four reference phases: 0°, 90°, 180° and 270°. For one idealized sign convention, the calculation is:

The opposing samples help reject a common background offset. Their differences preserve the modulated component used to estimate phase. Actual cameras can use different sampling arrangements, reference conventions and calibration corrections; this is a measurement model, not a description of every sensor.

A depth frame may combine several correlation measurements, potentially at several frequencies. The number of samples does not necessarily equal the number of separately exposed images available through the interface. For the timing implications, see ToF exposure time and frame-rate trade-offs.

The equation estimates optical range along a viewing ray. A camera may instead publish depth along its optical axis after geometric conversion. Check the output definition before comparing it with a physical reference. The distinction also matters when transforming point clouds between optical and camera frames.

Modulation frequency trades sensitivity for ambiguity

Phase repeats after a complete modulation cycle. With only one frequency and no additional information, returns separated by a complete cycle can produce the same measured phase.

The corresponding single-frequency unambiguous range is:

The second expression assumes phase noise is expressed in radians. It describes local sensitivity, not a guaranteed camera accuracy.

Using approximately 300,000,000 m/s for the speed of light gives these illustrative values:

Higher frequency makes a given distance change produce a larger phase change. For the same phase uncertainty, that improves distance precision. However, it also shortens the distance interval before phase repeats.

Real phase uncertainty is not constant across frequencies, targets or operating modes. Illumination strength, demodulation efficiency and calibration affect it. A higher frequency alone therefore does not prove better measurement performance.

Multiple frequencies can help resolve ambiguity, but their usable range depends on the measurement sequence and reconstruction method. A mathematical ambiguity interval is also not a usable working-range specification: insufficient return light can limit operation much sooner.

Worked example: distance and phase uncertainty

Consider an illustrative measurement using 20 MHz modulation and a corrected phase of 96°. These are example inputs, not specifications for a catalog camera.

First convert the phase into a fraction of a full cycle:

Now suppose repeated phase estimates have a standard deviation of 1°. The corresponding local range standard deviation is:

At 50 MHz, the same phase standard deviation corresponds to approximately 8.3 mm. That improvement applies only if the camera maintains the same phase uncertainty under the comparison conditions.

A return from beyond the single-frequency ambiguity interval can fold into an apparently valid nearer distance. That is different from a clipped display or an incorrectly scaled point cloud. If a depth map ends abruptly, first distinguish physical ambiguity from the issues covered in depth scale mismatch and close-range point-cloud cutoff.

Separate accuracy, repeatability and valid coverage

Three measurements answer different purchasing questions:

• Bias: How far is the average measurement from the reference distance?
• Repeatability: How much do repeated measurements vary under unchanged conditions?
• Valid coverage: How much of the required surface produces usable measurements?

A camera can repeat the same biased distance very consistently. Conversely, its average can agree with a reference while individual frames are too noisy for the task.

For example, suppose a reference surface lies at 1,000 mm in the coordinate being tested. Across 100 frames, the measured mean is 1,004 mm and the sample standard deviation is 2 mm. The observed bias is 1,004 − 1,000 = +4 mm, while temporal repeatability is described separately by the 2 mm standard deviation.

Do not report that measurement simply as “2 mm accuracy.” State the bias, repeatability, sample count, target, operating mode and reference uncertainty.

For an axial-depth output, a perpendicular plane can provide a convenient reference across the image. For ray-range output, distances to that same plane increase away from the image center. Compare like coordinates rather than treating this geometric difference as an error.

Test multiple distances, surface reflectances and image locations. Include normal startup behavior and the manufacturer's documented stabilization conditions. If valid coverage changes after power-up, investigate depth coverage changes associated with IR exposure before attributing the difference to phase calibration.

Why background rejection does not eliminate ambient-light noise

Opposing phase samples can cancel a steady background contribution in the ideal calculation. They do not remove the photon shot noise associated with that background.

Bright ambient illumination can therefore increase depth noise even when its average contribution is subtracted. It can also consume detector capacity and cause saturation. Once a sample clips, subtracting another sample cannot reconstruct the missing information.

Assess ambient tolerance using the actual scene and operating mode:

• Compare valid-pixel fraction under representative low and high ambient conditions.
• Record both temporal depth variation and systematic distance shifts.
• Inspect the available amplitude, intensity, confidence or saturation outputs according to their documented meanings.
• Include both bright and dark target surfaces within the required range.

Visible brightness alone is not a reliable indicator of interference. The relevant factor is light reaching the detector within its optical passband. Also, a narrow optical filter cannot reject unwanted illumination that falls inside that same passband.

Longer integration may collect more useful return signal, but it also collects more background and can increase clipping. Shorter integration may prevent saturation while leaving a weak target too noisy. The related guide to IR depth-image noise in bright light addresses that diagnostic branch.

Multipath produces convincing but incorrect depth

The simple phase equation assumes the measured return follows one dominant path. Real scenes can send light from the camera to a wall, then to the target, and finally back to the camera. A pixel may receive both direct and indirect returns.

Those returns combine as modulated signals with different phases and amplitudes. The phase of their sum generally does not represent either path accurately. The reported depth can be smooth, stable and wrong.

Concave corners, reflective fixtures and nearby bright surfaces deserve particular attention. Transparent or translucent material can introduce additional path mixtures or internal scattering. At an object boundary, one pixel may also mix foreground and background returns; this is a related mixed-pixel problem rather than necessarily a wall reflection.

Use controlled scene changes to test the suspected cause:

1. Keep the target and camera fixed, then remove a nearby reflective object from the measurement scene.
2. Cover a suspected reflecting fixture with a suitable low-reflectance material without modifying the camera or its illuminator.
3. Move the target away from a corner while preserving its camera distance as closely as practical.
4. Compare interior surface pixels with pixels near boundaries.

If the reported target distance changes when only the surroundings change, investigate multipath or stray light. Averaging more frames may reduce random scatter while retaining the underlying bias.

Low return from thin or textured targets is a different failure mechanism. Compare those symptoms with close-range depth dropout on fuzzy surfaces and thin parts before applying aggressive confidence filtering.

Choose a depth principle before choosing a model

Indirect time of flight is useful when a full-field depth image fits the task and its measured performance meets the tolerance. It is not automatically the right answer for small dimensional defects or difficult optical materials.

Browse the 3D sensors category and use the camera finder to identify candidates by measurement type. Do not assume that a category label establishes accuracy or suitability for your scene.

The MV-DT01WMU is cataloged as a ToF sensing camera. The supplied catalog information does not establish its indirect-ToF architecture, modulation settings, measurement accuracy or ambient-light limits. Confirm those details in the manufacturer's documentation before applying the calculations above to that part.

The MV-DPS200P-02 is cataloged as a projected structured-light camera. It provides an alternative measurement principle to evaluate, not evidence that either approach will outperform the other on your target.

Request documented test conditions alongside any accuracy or repeatability figure. Check whether it refers to one pixel, a spatially averaged region, a fitted plane or a complete measurement algorithm.

Acceptance checklist before ordering

Turn the application into a test specification rather than relying on the appearance of a point cloud.

• Define the required quantity. State whether you need axial depth, ray range, object height or dimensions after coordinate transformation.
• Set the geometric envelope. List minimum and maximum target distances, required coverage and allowable placement variation. Use the 3D camera working-distance and positioning guide to structure this check.
• Specify errors separately. Record allowable bias, repeatability and invalid-pixel fraction rather than combining them into one ambiguous number.
• Use representative materials. Include dark surfaces, bright fixtures, edges and concave features that remain in the production scene.
• Record the complete operating mode. Capture frequency mode where exposed, integration settings, frame rate, filtering, confidence thresholds and calibration state.
• Test the lighting envelope. Repeat measurements under the expected ambient extremes without changing unrelated settings.
• Challenge ambiguity and multipath. Include farther background surfaces and nearby reflectors, not just an isolated reference target.
• Check neighboring active cameras. If several illuminators share the scene, evaluate interference and consult shared timing for multiple depth cameras.
• Preserve raw evidence. Save depth, available quality channels and configuration details so a later software change can be compared fairly.

For a static qualification test, collecting 100 frames per condition is a practical starting point, not a universal statistical requirement. Report the frame count and retain per-pixel variation instead of publishing only a heavily averaged surface.

When the usual answer is wrong

A higher modulation frequency is not a remedy for multipath, saturation or incorrect coordinate interpretation. Likewise, a lower temporal standard deviation does not demonstrate lower dimensional error.

If the task involves clear material, strong reflections or a tolerance close to the measured uncertainty, require a representative sample evaluation before selecting the principle. A different viewpoint, fixture geometry or sensing method may help more than additional filtering.

Keep the camera's illuminator, optics and protective features within the manufacturer's documented operating conditions. Do not alter emitter drive or remove protective components to increase return signal.

Use the camera configuration starting point to organize the hardware selection. If the catalog does not state the needed measurement conditions, contact applications support with your distance envelope, target samples, ambient conditions and acceptance limits. Those details make a useful evaluation possible before an order is placed.